Ladies and gentlemen, I'm very pleased to welcome you to NVIDIA's Annual Investor Day. You guys see a lot of green here happy Saint Patrick's Day to everybody. It also happens to be our GTC. You can go to th e next slide, please. I have the honor of presenting the most exciting slide in our presentation. It's our safe harbor statement. You can look at it, our forward-looking statements financial measures. My last slide. just overview of the agenda. We divide the day into two different parts. The first half starts with Jensen our CEO, then we talk about a couple of our businesses. We're gonna take a break in the middle where we can go and see the demos at our GTC conference. Second half of the day will end with, obviously our finance presentation and our Q&A.
With no further ado, I'd like to invite our Co-founder, President, and CEO, Jensen Huang.
I was expecting way more jokes. Way more jokes. You know, he wanted to be a stand-up comic, right? How did that work for you guys? That's right. Now he's in IR. I don't know what that means. See, I could have been a stand-up comic. You guys, welcome to our Analyst Day. This is a day I always look forward to a couple of reasons. First of all, of course, it's GTC. GTC as you guys, I hope you guys were there. GTC is really about developers. It's really about all of these developers that come together once a year to share their work in high-performance computing, where high-performance computing is a tool that enables them to do their work. They share their breakthroughs. They're inspired by each other's ideas.
They look at the course, all the papers that you guys are gonna see today. We focus on this particular time each year, and then we present it to each other. We inspire each other to do even better work. Together we could dream about the future. Some of the things that you guys will witness throughout this week are just people who are dreaming about the type of problems they can solve or the type of breakthroughs they could achieve or some new application they would like to bring to the world or I mean, it's really about dreaming the future. For us, what really exciting for me is what's really amazingly exciting is the opportunity to work with them and to enable their dreams. To work with them to enable their dreams.
The second part is Analyst Day. It's right next to each other for a very good reason. It's an enormous amount of work for all of you, I know. You're enjoying the GTC conference. You're learning about a lot of new ideas. Then, of course, right in the middle of it is our Analyst Day. The thing that is really great about putting the two together even though there's a lot of work to do for everybody involved, it is that you get to see us in our rawest form. This is what really inspires our company and I said today, and I mean it, that this is the beginning of it all.
If it wasn't because of all of this work here with developers, why would there be applications or end markets or end, you know, end results that were somehow created? Why would there be large markets for visual computing? Why would people need visual computing? It all starts right here. This is an opportunity for you guys to look at it right at ground zero and see some of the work that's about to change the world several years from now. You guys have heard me talk about NVIDIA for a long time. Long time for some, a very long time. I've been CEO of NVIDIA for 24 years, as you guys know.
I've had the opportunity to invent with all of the great people at the company, the modern version of computer graphics. We are the largest computer graphics company in the history of mankind. Computer graphics. We are the largest computer graphics in the history of mankind. We define modern computer graphics. We invented modern computer graphics. We also reframed what computer graphics mean some 10, 15 years ago. We expanded the scope of visual computing. We call it visual computing. It's not just about the generation of images from all the data that's inside your computer but it's understanding the world and understanding the images that come into the computer. Visual computing is a very broad field today, as you can see. We've reshaped our company in several ways.
Whereas the first NVIDIA that you guys met was really a PC graphics company. We built a chip that was connected to PCI Express. It was Windows compatible. It was VGA compatible. It supported the APIs that were standard in the industry. Largely, we try to make our chip compatible with all those interfaces, HDMI and so on and so forth, and make it as fast as possible. Largely for the first five years of our company, maybe arguably. Even 10, our endeavor was solving those problems, making computer graphics faster and faster, which is very important because you need to experience it in real time. That in itself was an enormous contribution already.
We were an industry-standard component, PCI Express, compatible with Windows, compatible with DirectX, compatible with OpenGL, largely interchangeable. Because of our incredible execution, because of the people that are inside the company, we were able to stay a step ahead the entire way. We started out as one of 50 companies. We eventually ended up being the only company, the only standalone computer graphics company in the world, the largest the world's ever seen. In no time in the history of mankind did one company play a role in computer graphics in such large and broad markets. I thought that the strategy for the first 10 year arguably perilous arguably extremely dangerous, but we executed with quite a bit of flair and we became quite large.
Well, being a component company, being a PC graphics company, fueled our growth in the first 10 years. At some point, with every component, doesn't matter what component you are. It could be a video component, it could be an audio component, it could be a graphics component. It doesn't really matter what component you are. In the end, if you're an interchangeable component, whether it's a DRAM component or flash component, doesn't matter what component you are, whether Ethernet component or USB component or it doesn't matter what component you are. It could be a camera component, it could be a computer vision component, it could be an ISP component. Doesn't matter what component you are. In the final analysis, if you're a component company. Eventually components get good enough.
This is a problem that has been characterized in a lot of different ways. Well, I thought Clay did the ultimate job of describing it. Christensen over at Harvard did the ultimate job of describing it. The Innovator's Dilemma. At some point, technology becomes good enough. Doesn't matter what component you create. We discovered that not only is that true. In fact, we've always believed that visual computing is a very difficult problem. It's a very difficult problem.
The problem of expressing information in a visual way and understanding the visual world in a computer way, that problem is so grand and so great that we ought to go find a few vertical markets, engage it deeply, solve deep, deep problems, and go build a company that is not a components company but a visual computing company that engages four vertical markets or a few vertical markets. We chose four. These are the four, we selected for the various reasons that we'll describe. We selected four to deeply engage. For those markets, we will be the domain experts. We're not a component supplier. We will be the domain experts. We will know more about those fields than any company on the planet. We will understand it at the technology level. We'll understand it at the software level.
We'll understand it at the ecosystem level. We'll not only support standards, we'll drive standards. We are a partner to all of the ecosystem partners. We will evangelize that market. We will solve problems for that market before that market even realize it has a problem. For the markets that we serve, we will be utterly experts. We were not going to be just horizontal component suppliers, that we were going to be domain experts in several vertical markets. Gaming is one of them. Gaming is one of them for several reasons. One, we love gaming. I still remember in 1993 when we first started the company and nobody invested in gaming. There were no VCs that invested in gaming with the exception of one, turned out to be my investor as well.
They were the only investor that understood what gaming was about. Electronic Arts had 16 employees. Video games was a very weird thing. Everybody asked me, "What's our killer app?" I said, "Video games." They said, "Okay, but tell me what your killer app is." I said, "Video games." They said, "Okay, startups don't depend on startups." Their point is that it's a young company. There's a lot of different ways you can think about that. It was, it was utterly correct. Nonetheless, we helped build that industry. Now video games is a $100 billion market. It is expanding still. Everybody's a gamer. Anybody who is born today, anybody who was born in the last 15 years, a gamer today. Gaming is obviously a very large market. It's also one of the most technologically challenging markets.
The reason for that is because we're trying to create virtual reality. We're trying to create virtual re-reality. We're trying to convince you that when you're playing Madden football that you're actually playing Madden football, that you're actually that team, you're actually that player. The animation and the graphics is so real, you can't tell the difference. I still believe that in a very short time, another 10 years' time, you are gonna watch a football game, and you're gonna be watching a graphics generated football game. You have no clue which one's which. That day is coming. Well, we think that there's a market for that. We think that the problem's extremely difficult to solve, and that has been one of our vertical markets. We know more about this industry than nearly anybody.
From the technology to the ecosystem to the markets itself on a global scale. Cars. Over 10 years ago, we came to the conclusion that the car is going to be a supercomputer on wheels is going to be software-defined. You can never write all the software necessary for every single car at the time that it was taken off the lot. If there was a way, if there was a company who could provide value to this particular vertical market, that would be us. That the car will be the ultimate visual supercomputer. We were wrong for about 10 years, and then last year we became right. This is going to be a very large market. I believe there's gonna be a lot of computers inside cars. Those computers, there'll be feature detectors, there'll be computer vision detectors, there'll be all kinds.
There'll be sonars, there'll be radars. There'll be all kinds of things like that. There will be one thing for sure. Some cars will have sonar. Some cars will have LiDAR. Some cars will have computer vision, some cameras. There's one thing we know for sure, that every car will have a very powerful computer inside, and that software is going to define the features. Anybody who has a Tesla, anybody who enjoys a Tesla now cannot understand how to drive a Porsche Cayenne. I have no clue how to drive that car. What are all these knobs and dials? When am I going to get my next software update? That's my only question.
Anybody who drives a Tesla today, the nanosecond that you see that OTA, if you didn't sense a little bit of joy that you're about to discover some new feature that Elon worked on our behalf, if you didn't experience some amount of joy, I would be very shocked. The car is going to be a rolling supercomputer. It's going to be software-defined. Companies who are great at software are going to be great at building these cars. It's going to be about building computers on wheels.
Well, there was somebody who said, I think it was Alan Kay that said, "Anybody who's serious about software builds their own hardware." It's true because most of the software that we're building on these platforms are really intricate things. I need to OTA this thing. I'm gonna support your car for as long as you shall live. That's not a minor promise. The reason why NVIDIA's gamers are so in love with us is because I support their GPUs and their experience for as long as they shall live. I've been asked, "Jensen, how long are you gonna support my drivers?" I look them in the eye and I say, "For as long as you shall live." It's true.
We've been supporting an install base of GPUs now 200 million large, and we update new drivers for them, and we bring new joy to them every single month. One of the reasons why NVIDIA is so loved, the dedication we have to that architecture. We can't do that. We can't do that. We can't do the OTAs. We can't update that architecture if we don't own that architecture. You can't update that architecture if you can't own that architecture. If you're not sure, just ask yourself, how many of your cell phones and how many of your tablets, Android phones and tablets, are you still waiting for Lollipop? If you're serious about software, you build your own hardware. NVIDIA does that enterprise. This market, capturing, realizing people's ideas, realizing people's imaginations, transforming it into products, that is a classic visual computing problem.
This is the oldest of visual computing markets, the workstation market. Our market share here, obviously very, very high. It's a market we care very deeply about. All of the sensibilities and the promises that we bring to gamers, we bring it in spades here. If you have a car that you designed 20 years ago now, and if it were to be contested in some way, you got to bring the database back. You can trust that if you had a Quadro, that database will pop right back, and all of the pixels will be there. Our dedication to that market all reflected in the market share. How is it possible that a company has 90% market share on the free market? How is it possible we have a 90% market share on a free market?
From top to bottom, just about every industry we serve enterprise graphics. One of the things that we did, we imagined all of us are taking our work with us. I do more work in a mobile state than just about any other state. I'm doing more work everywhere else except my desk. Now, it turns out that most designers can't do that. Why can't they do that? Because you can't pack a workstation with you. You can pack a basic PowerPoint PC with you but you can't pack a workstation with you. It's too much. Too much data, too much computing, too much graphics, too much everything. We thought, wouldn't it be great if we just virtualized and throw it into the cloud? Now we move the computer to the storage. A lot of wonderful things can happen in addition to that.
I'm sure Greg and Jeff will talk about that. As a result, instead of moving storage data to the computer, we did the opposite. Instead of moving data to the computer, we did exactly the opposite. We did the miracle and put the computer next to the data. The world's most interactive graphics computer, we now moved all the way across the network to the computer. The benefits are fantastic. HPC and cloud. You guys saw that's what HPC and cloud is precisely what GTC is about. It's precisely what GTC is about. You guys see the enthusiasm here. You see the growth here. The numbers are big. The numbers are big. There is no question now that accelerated computing, which was questioned all along the way, is unquestionably valuable today. Accelerated computing, GPU-accelerated computing. Questioned all along the way.
Maybe it's a fad. Maybe it could be replaced by FPGAs. Maybe multi-core CPUs will come along. They'll just add more transistors. All true, accelerated computing, and the reason why it exists. Is more profound than that It's more profound than that. The architectural nuance, more deep than that. There is a very specific reason, a collection of specific reasons, why accelerated computing is the right way to do it for a lot of applications. We prove it year in and year out, year in and year out. If you wrote a CUDA application seven years ago. That exact CUDA application, if it turns out to be John Stone, who's doing some really amazing work at UIUC on molecular dynamics and trying to figure out a cure for cancer.
If you're that guy, well, it turns out John has written a CUDA application seven years ago, every single year we just double the speed. We're about to increase it by a factor of ten with Pascal. Unbelievable. Port once, write once, sit back and do research. Sit back and do research. Sit back and change the world. That's really ultimately what they want to do, high-performance computing. We still have an OEM business. I mean, from all of this technology, all of these markets that we serve, we serve with the processor, an enormous amount of software, maybe a system associated with that, maybe a cloud service associated with that. Whatever, whatever is the best manifestation of that solution, whatever is the best manifestation of that service, we're gonna help somebody solve a problem.
Whatever is the best manifestation of that, we will create that and go to that market. Sometimes it's related to a graphics card. Sometimes maybe it's in the formation of a system. Sometimes it's in the formation of a system that goes to retail. Sometimes maybe it's a cloud service that's hosted at Amazon. Okay. We look at each one of these vertical markets on first principles. We then of course invent the technology underneath. Those GPUs are still valuable for PCs. We fight hard in every opportunity we get, and we seek out opportunities where GPUs, graphics, visual computing still matters. Maybe not for office laptops where people want it to be as thin as possible. Still, there are mobile workstations maybe they're gaming PCs, maybe they're PCs that they would like to target at digital content creation.
Maybe they just want it to be the highest performance PC that they offer. Same thing with tablets, same thing with other devices. We still have OEMs and of course IP. We have the richest portfolio of visual computing intellectual property on the planet. We have made greater contribution in the last 24 years to modern computer graphics than any company. Arguably all of the companies combined. This is our field. We're serious about IP. We're serious about patents. We invest a lot of money filing the patents. We invest a lot of money inventing the patents. We invest greatly, as you guys all know. As a result, we have a war chest of intellectual property that keeps us out of harm's way on the one hand. We can also, of course, leverage on the other hand. We are very serious about that.
Those are our five markets. Here's the scorecard. We typically, because of where we started the way we describe our business to you, changed over time. Now we've reshaped the company. We're gonna describe to you we still report the numbers in various ways, but we're also gonna describe to you our company's business from those markets, from the viewpoint of the markets. As you can see, the efforts we've made in the last several years to engage those markets have resulted in real growth. Not only just real growth really significant growth. This is a growth company again. There is no question NVIDIA is a growth company. Gaming has grown 36%.
It is now very, very, very clear to all of you that although we are a PC gaming company, we're not a PC company. We are a gaming company that uses PC as the platform to deliver that capability. PC is like air. PC is like water. PC is like electricity. PC is essentially a utility for most people. The fact that we provide GeForce on top of that utility doesn't make us that utility company. We're a gaming company. We're a gaming company in that business, growing 36%. There's a reason why it's growing so fast. You see it all around you. PC gaming, because it's largely open and because everybody has it is culturally friendly. Before you buy a game console for your child, you're gonna buy them a PC. It is utility friendly. You need it for social anyways.
You need it for work anyways. You need it for study anyways. You need it for all kinds of things that you do anyways a dding a GeForce to it turns it into a mighty game console. It's open. It's also open for innovation. It is the first platform to have invented massive online multiplayer games. It was the first platform to have innovated MOBA, turning multiplayer into a battle arena, the invention of esports. It is also the platform where VR will be innovated. One innovation after another innovation in gaming starts with PCs because it is arguably completely flexible, arguably completely open. Okay? Lots of reasons why gaming is going to continue to grow. Production value is growing all the time. At some point, it reaches critical mass.
The reason why, well not just it's a great movie, but the reason why they can invest hundreds of millions of dollars in Star Wars is because they know there are a lot of people who are going to come and watch it. That DVD and Blu-ray and movie theaters are abundant. There's a lot of different ways for you to enjoy Star Wars. They can invest $1 billion and still make it back. Well, turns out you can now do the same about gaming. You can invest $100 million on Call of Duty. You can invest $100 million on Assassin's Creed. You will have confidence that you can get it back.
The reason for that is because the PC platform has reached a critical mass so large, and there's such a large audience, you can now increase the production value. The benefit that we get that most movie mediums don't get is that when you increase your production value, it increases the need for GPUs. Today, you guys saw Kite running on one GTX Titan X. Craziness. 8 billion transistors to play that trailer. Well, if the production value of games continue to grow, we're gonna continue to increase our ASPs. ASP of our GPUs is defying Moore's Law. It has defied Moore's Law now for many, many years. The reason for that is because production value drives higher demand for GPU technology, which therefore drives higher ASPs. We're benefiting from a larger install base.
We're benefiting from gaming become larger than ever, and we're also benefiting from the fact that ASPs production value is increasing. VR is gonna take that to new levels like we've never seen before. Auto has been growing 100% a year. 85% is Jensen's version of CEO math, 100% a year. Okay. Just doubling every year. Enterprise, our market share is very large, and so our growth has slowed a bit. Well, there's two things that we need to do. We need to innovate and invent the future, and then we need to innovate and invent the future. We have two examples of that for you.
We invented GRID so that you could virtualize that graphics and put it into the data center and mobilize everybody and provide designers, workstation designers, digital content creators to provide them the freedom that we all have. Could you imagine? We're gonna invent technology that's gonna free people. They don't have to be burdened. They don't have to be strapped to their desk anymore. Second, we're gonna take computer graphics and design to a level that is just incomprehensible. We're gonna show you some pretty amazing stuff. I think we're gonna show them some pretty amazing stuff. I haven't seen the demos yet, but if they're gonna talk to you guys about this new thing that we're working on I won't ruin it for you, but it's pretty amazing. Okay? We're gonna innovate. We're gonna innovate for growth.
Third, HPC, as you can see, we're growing very nicely. I really do believe at some point we're going to reach the tipping point, then growth can very well accelerate. Here's the thing that I see. It's already been growing very fast. It's not easy to grow the space because it's all about software. You have to develop the software. You have to develop the software, and then it gets deployed. You're seeing the benefits of deployment. Now, as you can see, the groundswell of new software that's being developed for GPUs is growing all the time. It's growing all the time. It's going to show up somewhere. It's going to show up somewhere. If you develop the software, it will be deployed. You develop on a Titan, you deploy it on a Tesla.
You develop it on Titan, you deploy it in the cloud. Tesla's all over the place. Okay. Lastly, our OEM business. If the PC market is down, we'll be slightly down. If PCs become more thin and light, it puts pressure on the number of GPUs that are adopted. We still have a very large, sizable OEM business. My sense is that over time, it will decline. Hopefully, and my expectation is the growth businesses not only outgrow its decline, but the quality of the business is far better. Not only can you see that the growth rates are now big on large businesses, the quality of the business is much better.
We're the only computer graphics company in history that has made its way and invented its way from arguably the platform for computer graphics, which is a workstation, as we became a PC. We innovated and invented our way into two very challenging environments for computer graphics. GPUs are the largest, most complex processors that humanity makes. It is the GPUs we create, Pascal, consumes more engineers, more transistors, more time than any processor that is built anywhere else. Well, that computational horsepower has a burden. It needs a large envelope. It needs large space. Energy efficiency is not the friend of parallel computing. Yet over years, we've challenged ourselves to drive the energy efficiency so high that we can now put GPUs into the smallest of mobile devices.
We have made it so energy efficient that you could populate hundreds of thousands of GPUs in a data center, and it won't melt. These data centers are now powered by GPUs because it is now the most energy efficient way to do processing. We had to invent those things. Graphics cares about latency, and we made graphics so fast. We fought the speed of light and every layer of code to the point where we can now put graphics away in a data center, and you feel like it's right next to your desk. That made it possible. Those two fundamental inventions, virtualization of graphics, put it into the data center. Extreme energy efficiency, so I could put it in the data center in large numbers and also put it into a mobile device.
Those fundamental inventions made it possible for us to turn mobile cloud, which is unquestionably the most powerful force in the history of computing, to turn mobile cloud into our friend. If we didn't do that, we'd be fighting mobile cloud for the rest of time. No computer company will be a computer company in the future if they're not in mobile cloud. We've turned mobile cloud into a platform of advantage. We've turned mobile cloud into an innovative platform for us. One where we can now inspire new ideas instead of all day long worry about it. I am so excited and so proud of the company for making that investment, look where we are now. Mobile cloud, big and growing businesses. The ultimate difficult challenge. You can put storage in the cloud that's easy.
You can virtualize computing, that's easy. Computing is virtualized all by itself anyways. That's what virtual memories are for. Virtualizing graphics, inherently hard, no one has ever done it until we did it. We invented virtualization for graphics. We made it possible for GPUs to exist in the cloud and mobile environment. Okay? Now they're both growth platforms. Very quickly, we have several winning strategies that I described one, to focus on vertical markets. Not a whole bunch of them. The ones where our expertise, the visual computing expertise our company has utterly vital . Without it, you will not survive. Without it, you add no value. Without it, nobody returns a phone call. In these vertical markets, our expertise is utterly essential. Number two, engage in these vertical markets, be thoughtful about the products that we bring.
These platforms has to be much, much more software rich, the way you can be software rich and solve real problems is by deeply understanding that domain. We are domain experts. We know the problems they need to solve. We know the problems they will need to solve, We know the problems they wish they knew that they will have to solve. We're solving those today. Two, data center growth and mobile growth. Those are the four basic strategies of our company. Vertical market. Add a ton of value. The way to do that is being a domain expert. Don't forget, mobile cloud is the future. Mobile cloud is the future. We gotta go ride that wave. If we can ride that wave, we can expand our market dramatically. I'm gonna show you just very, very quickly a few examples.
This is an example of us focusing on a vertical market and not just thinking about it as a component. Titan ultimately is a component. Titan X is the most complex component the world's ever made. A normal company can't make such a component, we don't even stop there. We turn the PC into a game console, if you will, within. A game console within the PC. We made it as powerful as a PC, as easy as a game console with a click of a button. With a little click of a button. It's called OPS, Optimal Playable Settings. Okay? With OPS. Optimal Playable Settings.
With just a click of a button, it will understand the nature of your PC, understand the nature of the game, and set all of those complicated settings all by itself for you. There's a whole layer of mathematics that go on top of our chip. We call it GameWorks. Fire, smoke, water, hair, explosions, collisions, particles, cloth. Well, that's called physics. That's called Newtonian physics. Well, it turns out physics is really, really hard to replicate. Shadows, light rays, God rays, volumetric rendering. These are all effects of real life that are just incredibly hard to duplicate, mathematically intensive. We have a fantastic team of computational mathematicians who then architect, create, design these things. We're like the modern version, the real-time version of Industrial Light & Magic.
We create these special effects, and we put it into these things called GameWorks, and we license them to game developers, and they end up in games. Two benefits: one, the game becomes much, much more beautiful, and two, because the production value is higher, more people play it. If there is a subsequent benefit to ourselves, it therefore consumes more GPU horsepower. We do it for the first two reasons. Okay? We invented GPU accelerated cloud. I already told you about that. We are in full production now. I think VMware announced, I forget which day, but sometime this week, that vSphere 6 is now in full production. We have quite a few trials fully deployed and pilots running. 1,800 or so that we're counting. There are probably others that we haven't counted because we don't see everything in enterprise.
It's a very large space. We captured about 1,000 the year before, we're growing at quite a large click. I'm quite excited about this. We have now set up the ecosystem. From all of the servers and all the OEMs, all of their sales force, we call it OEM sales enablement. We have set up servers, OEM sales forces, OEM marketing teams, the software companies above it, VMware Citrix. All of the application providers, we have now created the condition by which this entire ecosystem can go to market. We invented this platform. We invented this capability, now we've harnessed the entire and organized the entire ecosystem of enterprises around the world to be able to engage literally every company in the world. Quite excited about this. Use mobile to revolutionize cars. Use mobile to revolutionize games.
I already talked plenty about cars. We're gonna talk plenty more about cars. I'm just gonna jump ahead to one that I'll introduce to you. You've heard me say before that mobile is more than phones. Mobile is more than phones. I said it over and over again. Mobile is more than phones and mobile is more than phones. I think it's apparent now that mobile is more than phones. It's hard to hear that mobile is more than phones when phones is the only thing growing, and that whole industry was being revolutionized. Here's the thing that's gonna happen. Just as mobile cloud has revolutionized the way we use our mobile devices because it put a computer into your hands. The only way to do that is with mobile technology, very low power technology, small technology.
It's connected to app, applications in the cloud. The benefit of an application world connected to the cloud is you have no idea what's about to be invented. Anybody who's cynical about it can't be cynical anymore. Look at the inventions that have happened along the way. Somebody all of a sudden invented the first app which killed off a device, a GPS device. A map killed off a GPS device. Adding a small sensor into a camera with an app kills off a discrete camera. Of course, nobody carries music players anymore. That was just really just the beginning. All of a sudden, Twitter comes along, and Uber comes along, and all these great Yelp comes along. I wish somebody would solve all that noise that goes into Yelp. I love Yelp. I just wish it was more pure.
If you guys have an answer for that'd be great. All of a sudden, these apps come along as a result, we didn't think of it. Nobody thought of it except those innovators. As a result, one after another app changed the way we enjoy our day. Well, I believe the same thing's about to happen for cars, and the same thing's about to happen for the way you enjoy television. We've been working for some time now. In fact, this was the first SHIELD we wanted to build. For a variety of reasons, it just took a lot longer. One of the reasons is waiting for an operating system that is connected to the world's largest store. Android TV is really important.
I think Android TV is gonna do the same for smart television that smartphones has done to mobile devices. It's gonna be connected to a very popular store. You're gonna be able to download applications. As you just heard Elon said today, when you download applications, you have to design a computing platform that has a lot of horsepower. The reason for that is because you just don't know what applications you're gonna download. If you want this platform to have some meaningful life, and I don't know about you guys. But my cable set-top box is in my house for 15 years now. There is no hope that you can create a computing device that can accept applications for 15 years. If you don't start off with a fairly reasonable platform and then start.
Our view is once you create such a device and applications organically enter into the world's largest store, and if that user interface was compelling, a lot of things could happen. We decided to build SHIELD. It is the world's most advanced set-top box. It is the world's most advanced smart TV device. It is the world's first 4K smart TV device. It's powered by Android TV. It's powered by Tegra X1. It's connected to the world's most popular store. Of course, because it's Android TV, it has a really fantastic user interface called Voice Search. Our hope is this. Before I show it to you, our hope is this.
Our hope is that we will help redefine the television experience, that there are 40 million over-the-top set top boxes already, that this is going to be the way people enjoy televisions in the future. Everybody will have a smart television. There are a lot of televisions in the world that don't have that capability. As one of the earliest ones, the best one, and the one that has the ability to unify entertainment for you, meaning whose television shouldn't play games? We've already established everybody plays games. What television shouldn't enjoy games? We brought those two. We turned a game console into an app.
Just like we turned a music player into an app, a camera into an app, a GPS device into an app, we turned a game console into an app and connected to an invention that we made called GRID. Okay, Without further Ado, guys, let me introduce you to SHIELD.
Thanks, Jensen. What you're seeing here is the Android TV UI running on SHIELD.
Actually, his name is Sridhar. His name is not actually Sridhar. I see Sridhar every week. We're always playing with Sridhar. For a moment there, I thought his name was Sridhar. His name is actually Sridhar. Sridhar, here, take it away. I'm sorry.
All right. What you're seeing here is the Android TV UI running on SHIELD, and it just looks beautiful. The whole UI is now, you know, optimized for a nice 10-foot lean back experience that you can control using a game controller or a remote. Because it's Android, all your, you know, Google Play movies, music, everything that you purchased on your Android tablet or phone is now immediately accessible in your living room. It also provides you really nice recommendations for movies, TV shows, music that's based on your past browsing history. This is the Recommendations tab on top, and it basically pulls content from all these apps that are installed on your SHIELD device. The thing I absolutely love about Android TV is its powerful voice recognition feature.
Instead of typing in long strings of text in my game controller to search for content on YouTube or Google Play movies, now I can naturally speak to my TV using voice recognition. We have a microphone built into our SHIELD controller and our SHIELD remote, so I can basically use that and talk to my TV. Let me show you how, show me movies by Robert Downey Jr. There you go. It goes ahead and pulls all the movies that Robert Downey Jr. has acted in and lists it in a nice format here. It also shows me YouTube clips that Robert Downey Jr. is featured in, so it's really powerful. It's pulling in content from all the various Google sources. I like the movie Avengers, so I'm gonna go ahead and click that and go to Google Play to purchase it.
Since I've already purchased it, I'm gonna resume playing.
Starting the video stream .
It's being streamed from the web, so it'll take a few seconds for it to buffer up to go to full HD resolution.
It was quite a rush around here finding Evans.
Another cool feature about Android TV is, it has something called as info cards. Here, for example, I can pause the movie, and it immediately recognizes the actors on screen. It recognized that There's Chris Evans on the screen, and since, Scarlett's face is, you know, facing sideward, it just says, "I just missed Scarlett Johansson." I don't know much about Scarlett, so I'm gonna go ahead and click on her info card to find out more about Scarlett. It says that she acted in Avengers, and she's the Black Widow.
Right.
It also lists all the movies that she's acted in, the YouTube clips that she's featured in and what, you know, what are the other cast members of the Avengers and what people search for when they search for Scarlett Johansson. It's a very powerful way to, you know, get information about your favorite actors, music and stuff, through Android TV. Like Jensen Huang mentioned, SHIELD, the amazing feature about SHIELD is it's the world's first 4K playback device.
Just a quick comment, Sridhar. One of the things I really love is the day I took it home. You know, most devices when you take it home, when you first plug it in, it's a brick. Okay. It's a digital brick. The thing that's really cool about this device is you go home, you type in your Gmail address, everything shows up. All my music showed up, all my videos I've, you know, purchased, all that showed up. My preferences are there. It's actually really interesting.
All right. SHIELD is our only living room device that is capable of playing 4K videos. Not just 4K videos, 4K videos at 60 frames per second. No other device on the market is capable of doing that. Let me show you what 4K 60 FPS looks like. That's 4K 60 FPS video, a lot of content is coming. You know, content providers like Netflix, Amazon, Hulu, have all talked about bringing amazing 4K content to the market. SHIELD will be ready for that content. Obviously, being a living room entertainment device, it's not just about videos, it's also about music. Like I mentioned, I can go to music players and use my voice search to search for stuff. For example, Journey. Goes ahead and pulls up all the albums of Journey. Let me go in here and select the greatest hits.
Let's play one of our favorite songs here. Now I can go back to my home screen and like, do other stuff like watch photos. I'm gonna show you a bunch of photos that we captured at our Project Inspire event, which we held last year. This is basically an event that we do every year to help out the local community. We basically select a local school or a community, you know, place and try to, you know, improve it by rebuilding it, planting trees and stuff like that. Really fun event that I really enjoy going to every year. That's Android TV and SHIELD as a great media playback device in a nutshell. Obviously, we built SHIELD as a gaming device as well, and it is made to game. Multiple ways to game on SHIELD.
You can either download great Android games from the Android Store, or you can stream games from the web through our NVIDIA GRID cloud gaming service. SHIELD is powered by our Tegra X1 mobile processor, which is the most powerful mobile processor in the market today, and it uses our Maxwell GPU. You can get some amazing games. What you're seeing here is Crysis 3 running natively on our SHIELD device. It's amazing. It's amazing that a game that required a high-end, you know, GeForce PC just a few years ago is now running on a mobile device that is thin and sleek and can sit in my media center amazingly.
This architecture is exactly the same as the architecture of GeForce, it runs exactly the same game engines. Unreal Engine that I showed you earlier.
Okay.
Unity Engine. All of these engines, it doesn't matter which engine it is, could be a Crytek Engine. In this case, this is a Crytek Engine. These engines are exactly the same produces exactly the same pixels as GeForce, which makes the portability of it incredibly easy.
That's Crysis 3. Let's look at another way to game on SHIELD.
We recently launched our NVIDIA GRID cloud gaming servers, through which you can stream amazing PC games at full 1080p resolution at 60 frames per second. Let's take a look at a really nice racing game that's being streamed from the cloud. This is GRID 2. A good way to test the latency of any cloud-based gaming is to look at racing games like this, which are very latency sensitive. Take it away, Ben.
Sector one is where you really need to push yourself. Good. Let's push for first.
Let's see if Ben can make 1:58. It's 1:43, 1:44.
Okay, boss. You got a nice early lead.
Now, ladies and gentlemen, you saw two no small feats. You saw two no small feats. That was 1080p, 60 hertz. You know that current generation game console have a hard time keeping up with that, and the reason for that is because that's 2x the performance of a current generation game console, a state-of-the-art game console. It's a click away. Click and play in 60 seconds. Click and play. No downloads of 10 hours. You don't have to run to Best Buy. You just click and enjoy, just like you do with Netflix, just like you do these days, you know, with Amazon Prime. We're gonna do for games what Netflix has done for movies. The second not-so-easy feat was Ben's achievement just now. Other companies have 10% free time for inventions. We have 10% free time for gaming. All right, very good job.
SHIELD. We believe that SHIELD will revolutionize your TV experience. It is the world's most advanced smart TV device. We believe that the future of smart TVs is about all kinds of applications, including amazing games. In doing so, not only are we changing the way you view television and hopefully bring to everybody the benefits that smartphones has for mobile devices, but we're gonna dramatically expand the reach of games. Notice what we're doing here. You've seen this strategy before. This is just an even larger version of that. This strategy is exactly the same as adding a GeForce to a Windows PC and turning Windows PC into a game console. We're doing exactly the same here. We're adding Tegra to a smart TV platform and turning that smart TV platform, of course, into everything that a smart TV is, but also a wonderful game platform. Okay.
Revolutionizing smart television, the SHIELD Android TV console. We're gonna put it on sale. It'll be available in May, and the base product is available at $199. We're gonna be announcing peripherals and its pricing and so on and so forth as we get closer to the launch. Let me give you an update on IP. I've already mentioned the importance of IP to our company, the importance of invention. We are the world's largest visual computing company. We're the most innovative visual computing company, and this is an area we care very, very deeply about. You guys can see that visual computing and its reach is growing all the time. That can't happen without invention . We care very, very deeply about fundamental invention in this field.
We have a treasure chest of great IP. We're serious about monetizing it. We're in litigation with Samsung and Qualcomm. We don't have anything particularly to report today. The case is moving nicely through the courts. On June 22nd and June 23rd, we hope to have the first of the ITC hearings. This is a very important case for us. Largely our IP strategy is focused on this at the moment. Okay? Growth. All of these new things that I've talked to you about over the years. I think that we've been saying the same thing over and over again. Today, of course, we get the chance to reframe it again. This reshaped NVIDIA with our new strategy and our way of engaging these vertical markets deeply.
Rather than being a component company, we no longer think about how many GPUs are sold. That's the component view. We no longer think just about what's the ASP of GPUs. We care about both of those matters, but that's not enough anymore. What's much, much more important is what is the market opportunity within that market that we serve. For the vast majority of the time that you've known us, the market opportunity for GeForce is how big is PC GPUs. That's not the case anymore. The question that we have to ask ourselves now is how big is the gaming market? Because we're not engaging PC GPUs, we're engaging the gaming market. How big is the gaming market? We have two ways to engage it.
We engage it through PC, which is still growing. Surprisingly large. It's gonna continue to grow, I believe. We have every evidence it's gonna continue to grow. We have a new way of engaging that game market through SHIELD. Second, the automotive market. Rather than selling components into the automotive industry, we're selling computers into the automotive industry. Instead of thinking about it as a GPU, as an SOC, think of it as a mobile supercomputer. Think of it as a mobile supercomputer. Rich in software, rich in capabilities. Some of the capabilities are related to. Of course, infotainment and beautiful digital clusters. Some of the capabilities are related to surround vision to help you see around the car and some of the features are related to deep learning.
As we go towards self-driving, we don't have to get there without spinning off and not get the benefit of spinning off a lot of capabilities along the way. We'll get there. We'll get there when we get there. We're gonna spin off a lot of capabilities along the way, okay? Don't be distracted by the fact that we're working on self-driving cars. It's the same thing as we're distracted by trying to solve photorealistic rendering. Along the way of photorealistic rendering, along the ways of virtual reality, along the ways of being able to replicate everything that you see right now, we're gonna spin off a lot of great technologies along the way. Auto is not about chips, auto is about auto computers. Enterprise. High-performance computing. The way we go to market with Grid, increasingly software-rich. Increasingly software-rich.
The processors that will be in the cloud will be many and growing. We see those as large markets. Okay. That is my part of the afternoon. I appreciate all of your attention. It's been fantastic talking about our company story with you guys all throughout the years. I wanna thank all of you for your support. I think that the opportunity for Q&A is afterwards. Am I right, Arnab? Okay. All right. Next up is Jeff Fisher, head of our gaming business. Fish?
Thanks, Jensen.
Thank you, everybody.
Hello, everybody, welcome to another year of Investor Day. I want to talk about PC gaming. Wrong button here. Want to talk about PC gaming. We had a good year in GeForce gaming. I think many of us were talking last night about what an amazing year it was. We've seen growth in virtually every region in the world. Especially in China and Southeast Asia. $2 billion-plus revenues past year for GeForce gaming business. 36% year-on-year growth. We also launched Maxwell, our certainly best GPU ever and the most advanced GPU ever built. Maxwell delivered 2x the performance of a prior generation GPU and 2x the performance per watt, the 2x the power efficiency of prior generation GPUs. Today, Jensen announced Titan X, our new flagship Maxwell GPU, the fastest GPU we've built.
Titan X is priced at $999. It's a product that we've engineered in-house. We will ship out to the channel, to all of our channel partners as a complete product. It really bears a resemblance of a high-performance gaming product. I can pass this around. Why don't I try that? Although you guys can really appreciate the quality of the gaming product we deliver. See if it gets all the way around. Please, that one does not work. I don't want to see it on eBay, not making it around and see it on eBay later today. Maxwell has been great for our business as well. First, let me back up. As I talk about the PC gaming market, let me back up a little bit and talk about gaming overall.
There's roughly 1.7 billion gamers in the world today. Jensen mentioned virtually anybody over 15- 20 is a gamer, and it's true, about everybody is a gamer. This is roughly 20% of the world's population playing games. They play on a multiple of different platforms, including console, and we can debate at length what the future of consoles is. Today, they play on console, they certainly play on PC, and they play on mobile cloud devices. PwC's entertainment survey last year showed that video gaming, number two to cable subscriptions, but is ahead of virtually every other type of household entertainment spend. Ahead of books, ahead of movies, ahead of music, and continues to grow. Take for example, the movie Transformers: Age of Extinction that came out last year was the number one grossing movie last year.
Number one. I know I didn't love it either, but a lot of people did. Globally, in 15 weeks, it grossed $1 billion in sales. Contrast that to Grand Theft Auto V that came out last year. Grand Theft Auto V did $1 billion in sales in 24 hours. It broke every entertainment revenue record in history. That was just on one platform. That was on console. Grand Theft Auto V, it turns out, comes out next month for PC, we're really excited about that to help drive really the PC gaming community. Games are driving household spend. Gamer, gaming is a huge market and continues to grow year-over-year in all regions in the world.
Well, one more thing, one more point I wanted to make. Game Developers Conference was two weeks ago, up in San Francisco. It's kind of the mecca of the gaming industry for game development. There's always an annual survey where they ask the attendees, "What platform do you intend to target next year?" A PC again, comes out as number one. Target for game developers attending GDC. Mobile cloud is number two, a distant 25% says console. PC remains and continues to be a very relevant target for game developers in game development and a platform for gaming. Let's take a look at PC gaming in particular. 330 million core PC gamers in the world, the numbers are much bigger than that when you just talk about casual PC gamers.
These are core PC gamers. Gamers that play hours and hours a week on PC. Gamers that spend money on games. $330 million, I think Newzoo is estimating this is growing at about 10% a year for PC gamers. DFC, which tracks software revenue to PC gaming, shows that PC gaming is a roughly $27 billion-$28 billion business today, and that's growing at a CAGR of about 8% and should continue to grow for the next several years. Steam, you may or may not be familiar with Steam. Steam is the number one digital store for buying games online. Steam's active user base has grown from roughly 50 million in 2012 to 125 million last year, and this is targeted at PC gaming. There's roughly 4,000 titles in the Steam catalog.
Turns out they're not all Windows. As you may know, Steam is also developing their own gaming OS for PC called SteamOS, and there's roughly 1,000 titles of that 4,000 that are targeted at the SteamOS Linux gaming. It isn't just Steam. There's also new gaming genres like MOBA that Jensen mentioned. Multiplayer online battle arena that has become really the turbocharger for esports. League of Legends, the number one game in the world, has roughly 27 million daily players. 27 million people are playing League of Legends every day. League of Legends is a PC-only title. The reason for that is there is a genre, a class of gaming, and certainly competitive gaming, that wants to play on PC. Why keyboard and mouse?
It gives you a level of gameplay that is nonexistent through a controller on other platforms. PCs were very relevant to future of gaming and today and current gaming. What are a couple of factors that are driving PC and continue to make the PC platform growing? I know I talk to you each year at these events, and the number one question is how and why does this continue to grow? Every year it does, and it is not in New York. It is not. While the U.S. is growing, the majority of the growth is outside of the U.S., in Europe, in Russia, in Asia, China in particular, Korea, Southeast Asia.
One of the real turbochargers to this has been computer gaming as a competitive sport, esports. Roughly 200 million global fans of esports. It's a pretty amazing number. These are people that are playing or aspirational gamers or people watching online. They flock to events. The League of Legends Championship, which was held last year in Korea, which was really the birthplace of competitive gaming. It was held last year in Korea in an outdoor soccer stadium, a World Cup stadium. They attracted 45 million 45 thousand fans to fill this outdoor arena to watch an esports competition in Korea. They came from all over the world to watch.
The gamers that are competing in this in this championship are celebrities among the gaming community. On top of that, there were about 30 million people that tuned in online to watch the championship. They not only come to these big events, they come to small gaming events all over the world, and we've sponsored many of them in Taiwan and China and the U.S. As little as 50 to 100 people in iCafes to several hundred to 1,000 in gaming venues. esports is big and growing, and the audience fans are rivaling some very well-known established franchises like the NHL, the NFL, and even Formula One. This is a diehard base of gaming fans. Another factor fueling PC gaming is the social aspect.
Twitch, which was roughly started in about 2011 as a gaming channel, and it just caught fire. If anybody wants to go watch live stream of a gamer, anybody as from an amateur to a pro can broadcast their gameplay on Twitch. Well, Twitch has grown to roughly 1 million broadcasters, people broadcasting their gameplay, gamers. 100 million viewers who are tuning in at any time of day, find their favorite gamer or just their favorite game, watch someone for enjoyment, entertainment, or to learn some techniques, watch them play. 100 million people watching 16 billion minutes of gameplay a month. As you know, Twitch was bought by Amazon last year for $1 billion. Amazon sees this as a very important cultural and social vehicle for getting in touch with this enormous gaming community. YouTube.
YouTube is a place that you can clip and upload your brag clips or complete game playthroughs. Now, I know your kids or maybe some of you who are stuck playing a game or want to know exactly the right technique to take down the boss in a game, you can go to YouTube and see some of the best people or just amateurs teach you how to play through it. Now, you think your kid is actually on Khan Academy, but in fact, they're on the Battlefield 4 channel trying to figure out how to get through that last level. There's roughly 15% of the content on YouTube is dedicated to video games. The content, the gaming content on YouTube generates about twice the engagement of any other content on YouTube. People will watch it and re-watch it.
They'll comment on it. They'll talk to their friends about it. They'll share it. YouTube is also an important social platform. In China, where social gaming has driven this iCafe culture. We've talked about iCafe with you for years. There's roughly 150 million iCafes around China where people gather to play with their friends. 150,000. Sorry, where people will gather to play with their friends. Social gaming. Game streaming is just now starting to pick up in China. yy.com, which is a huge social network in China for chat and streaming music and other things, has now started to lean into game streaming. In fact, we have a relationship with YY that with a single click, you can now, from a GeForce PC, broadcast your gameplay up into YY.
This is an exciting area of growth for social gaming and streaming. Competitive gaming, social is all help building and driving this gaming culture, PC gaming culture worldwide. Jensen mentioned it a while ago. We've talked about this for years. It's an important ingredient in the PC as a piece of gaming hardware. It is open and scalable. The PC is a living, breathing thing. Every gamer can build the PC that they want. They can configure a PC for the right performance and price that suits their needs. DIYer, they can build it or they can buy it. In fact, DIY is so important to our business, we have a channel on our website called GeForce Garage, where we teach best techniques to gamers who want to build their own PCs around the world.
It's like car modders. Kids used to mod their car. I know I did. Now they like to mod and update and tinker with their PC. One of the challenges of a PC that is so configurable is compatibility and optimization. Every PC has got slightly different memory, CPU, hard drive, GPU, monitor resolution. How do you make sure that a game that is being developed for PC runs best on that PC? Well, we've leaned into this problem and have been working on a game platform around GeForce. We no longer just sell the GPU, although it is the heart of what we sell, GeForce GPU, and a range of GPUs for multiple gamers. On top of that, we've built a client we call GeForce Experience. Jensen had mentioned it earlier.
GeForce Experience is designed to make your games play best on your PC. I'll give you an example. Today, Battlefield Hardline launched from EA. Brand new episode of Battlefield, tons of gamers ready to download it and play it. This morning, our 57 million GeForce gamers who have GeForce Experience installed on their PC got a notification that there's a game-ready driver available for Battlefield Hardline. Click here within our client to download and update your PC and get it ready for Battlefield Hardline. The issue is that Battlefield Hardline has been years in development. Our engineers have been working with team. We've been QA-ing it. We've been making sure that it runs best on this wide configuration of PCs.
We want to make sure that every PC is ready for Battlefield Hardline when it comes out, and we want to make sure they have the latest driver, the latest configuration of OPS for GFE the day it comes out. We align what I'll say is our OTAs around the most important games that come out in the industry. There's roughly one a month. You get a notification, update your PC get ready to play Battlefield Hardline. The developers really love it because they know the consumers are going to get the best experience on a GeForce PC. Another aspect of GeForce Experience is OPS, click to configure. Of course, there's many settings in Battlefield Hardline. We want to make sure it's set exactly for the best experience on your PC. Also we're developing share technologies inside of GeForce Experience.
There's a DVR capability that you can turn on that will record your gameplay on your PC with no performance impact. If you decide then to go back and clip and upload to YouTube so your friends can see it, or you can teach somebody about a playthrough, you can do that automatically. There's also a single-click share to Twitch in GFE. In addition to just keeping your PC finely tuned and optimized, we're building GFE into a share platform so you can start to connect and share your gaming experiences with your friends. Finally is GameWorks. Jensen had mentioned it. Tony's going to be up in a little bit to share some of what's going on in the labs of what I'll say is NVIDIA's Industrial Light & Magic.
He has a team of some of the best visual effects engineers and artists in the world at NVIDIA inventing new libraries, new ways to render real life on GeForce GPUs. Lighting, smoke, fire, hair, whatever it is in characters that will get us closer and closer to cinematic realistic effects in games. Tony is working with our architects and game developers to bridge that gap and deliver next-generation gaming experiences on GeForce GPUs and make sure they work, that we've got game-ready drivers when the game launches as well. That's GameWorks. Our promise through the GeForce platform is to deliver the best possible gaming experience on every GeForce PC. GeForce is a platform. Okay, let's talk about some of the growth drivers for GeForce GPUs inside the PC gaming market. The first one I want to mention is notebooks.
Gamers today want a portable way to game. They would like to take their gaming on the go, dual purpose platform to study with at a college, or just take and meet up with their friends and game. Historically, GPUs have been very power hungry. It's been very difficult to build a portable notebook. In fact, we call them transportable or what I used to call them are desktops in disguise. These were very big, bulky notebooks that the portability of them was roughly to run from one power outlet to the next. You had a rough about a half an hour of playtime on your gaming PC. Well, Maxwell changed all that. Launching Maxwell has enabled a new generation of sleek sexy gaming notebooks from a number of new players. This is a notebook built by MSI, Microstar.
This notebook is thin, it's light has got long battery life. I'll leave it up here. I don't know. This one will definitely show up on eBay if I pass it around. It's thin, it's light. It has a GTX 970 class. It's one of our highest-end GPU class GPUs in it. These notebooks have resolutions of 1080p up to 4K. They're designed for 1080p full HD gaming at 60 frames per second of all the latest games. Long battery life. They come pre-installed with GeForce Experience. As you can see, we have seen a real acceleration in the growth of gaming notebooks. It's not just new entrants like Microstar and GIGABYTE. Alienware, Dell is shipping Maxwell high-end gaming notebooks, thin and light. Lenovo as well.
We're seeing tier one OEMs now getting more and more into the gaming space. This segment is alive. They are the same. They play the same games, League of Legends and Battlefield 4. The new gamers who want portability are picking up gaming notebooks. One of the drivers of GeForce GPU in PC gaming. Another, as Jensen alluded to, is production value of games. This is really important. When Tony gets up and talks, you'll see some of the future of effects that are going into games, new effects do drive GPUs. More importantly, as an ecosystem and industry shift, new consoles launched at the end of 2013. PlayStation 4 launched at the end of 2013 that was roughly six- seven times faster than PlayStation 3.
As reluctant as I am to stand up here and talk about console, what's important for you guys to understand is the PlayStation 4, the new generation of consoles have raised the baseline. The floor, if you will for the target of triple A games. When triple A developers are targeting a low baseline, they have to stretch way up to take advantage of a high-end GPU. Now that baseline has raised six to seven times. Triple A games are targeting a new class of performance. You look at a game, The Witcher from 2006, in the same era as Xbox 360 and PlayStation 3. The next generation of Witcher. We just started marketing it on GeForce. It's gonna be available. I think, in the end of April.
You can see the fidelity of the image of the current version of Witcher moving closer and closer to well, I won't say reality because we still have a long way to go, but a much more realistic effect, taking full advantage of the hardware that's available today that wasn't available back in prior generation console days. Triple A game production value is increasing, and a large driver of that is this new baseline set by console. Our GTX 960, which was launched in January, is roughly the equivalent experience 1080P, 60 frames per second on a PC. I'll call that a baseline for PC. This is where triple A games over the next several years are gonna be targeting for that sweet spot for their baseline experience, GTX 960. What does that mean for GPUs?
I'll use CEO math here, because sitting on Jensen's knee for the last 21 years, I've learned a few things from him. Including what CEO math means. Okay, I wasn't sitting on his knee. Standing by his side. You look at the installed base of GPUs. Yeah, that was kind of creepy. You look at the installed base of GPUs. 960, which we just started shipping in January, there are roughly 100 million GPUs in the installed base that have a lower class of performance than GTX 960. As the AAA games continue to push past this new baseline set by consoles, this is an opportunity.
This continues to be an opportunity, which is what we've seen over the years, but continues to be an opportunity for this installed base to continue to be motivated to upgrade their PCs. We think Maxwell, the performance class of Maxwell going forward is between that and the production value of games is a great opportunity to drive GPU, GeForce GPU over the next several years. Finally, new gaming experiences. The current class I just mentioned, the new baseline, 1080p gaming 60 hertz, PlayStation for GTX 960, is what's really driving the business today. 1080p screen is really the screen of choice for gamers. It is probably the most popular resolution, full HD, among gamers today.
If you build something immersive for gaming, the gamers will come. They want more. They want a better experience. They want a more realistic experience. We've already seen that with 4K gaming. Gamers are upgrading to get to a higher resolution screen. At GDC two weeks ago, we all experienced something pretty mind-blowing with virtual reality. A number of the game developers wanted to demo virtual reality, the latest generation, some pretty amazing virtual reality demos. Epic, Valve, Crytek all had developed these amazing demos to show at GDC in a VR with a VR headset. The problem was there wasn't a GPU on the planet fast enough to drive it in a way that it was completely immersive.
If you've ever had a VR headset on, I encourage you, we've got two set up in the exhibition area. Because we can't crowd around the screen, it's one person at a time. If you have the time, I hope you will experience it because you will see, I promise, the future of gaming if you have an opportunity to put it on. It's pretty mind-blowing. The prototype VR headsets that were shown at Game Developer Conference are roughly one megapixel per eye. It's about 1 K by 1 K. You have to run those at 90 hertz. It has to be fast frame rate because imagine putting your phone right up to your head. I mean, literally, it is right up to your eyes. It's gotta be fast. It's gotta be responsive.
It has to be high resolution for you to be really tricked into believing that you are there and not get a little motion sick. At 1K by 1K, 90 hertz, it is pretty amazing. They're still prototypes. In order to drive these next generation demos with this version of VR, call it baseline you needed a Titan. I was in the room when Tim Sweeney, who's the head of Epic Games, called Jensen and said, " Hey, dude. I need a GPU. I can't show my latest generation demo in the VR headsets at GDC next week without some hardware." And he knew we were working on Titan X. Our plan was really to surprise everybody and launch Titan X today. Tim's a great friend of ours and Epic is an important game development partner.
We did decide to show up at GDC with Titan X, and in fact, it wasn't just him. The guys at Epic or the guys at Crytek and the guys at Valve also needed high performance GPUs to show their VR demos. We, it turns out we had about 20 Titan Xs, so we populated the demos at GDC two weeks ago and announced its existence. Saved all the specs for today, announced its existence. You need a Titan X a $1,000 GPU, our highest end flagship product to really enjoy a basic VR experience today. It's about 2x the horsepower required that you would need on a 1080p 60 hertz. Imagine to get to what I'd call virtual reality speed of light, I mean, really the immersive experience you're gonna want.
Just from a pixel perspective, 4K by 4K, 90 hertz, which is what you would want to have a fully immersive experience, is about 12x the horsepower of 1080p today. You add a demo like A Boy and His Kite, where you're not just talking about the number of pixels, but the quality of each pixel. Each pixel is immersed, it's rich. You're talking about 24x. I believe the future of PC gaming is very rich. Pushing numbers of pixels, pushing quality of production value of content. We've got many years to go before we can deliver on something that would be real life in real time. Lots of headroom for us in this industry. That's what gets us really excited in the GeForce team in engineering.
Sure, every day we get motivated by competition. We wanna stay ahead. More importantly, we really want to deliver next generation gaming experiences to gamers. It is there. It is possible. It is within our reach. There's a long roadmap of things we have planned. There's a long future of things we can deliver, we're excited about, and are just not technically possible today. That's really what motivates and inspires the gaming organization side of, inside of NVIDIA from day one. Okay, before Tony gets up, I think, really shows you some cool demos, talks about our GameWorks strategy, NVIDIA's, effectively our Industrial Light & Magic inside of inside of NVIDIA, let me quickly summarize. PC gaming is big, and we see it continue to grow. We really do.
Esports, social, scalable, production value of content, portability, all factors that are gonna continue to inspire people to upgrade and move to the PC as a gaming platform. We are building GeForce is a platform within the PC. We are gonna continue to add value to GeForce gamers. They're not just inspired to buy a new GPU because of speeds and feeds, but they enjoy the entire GeForce gaming experience. Finally, growth. Notebook gamers, production value of content, and really important new immersive experiences for the future of gaming that can really only be delivered on PC. That's my story on PC gaming, GeForce and growth. I'd like to introduce Tony Tamasi, my brother-in-arms, who leads our content dev team and a band of really brilliant engineers and artists in the GameWorks organization.
Good afternoon. I only have a couple slides, and hopefully I can get to show you some cool technology. As Jeff mentioned, I manage the GameWorks effort for NVIDIA. The strategy is pretty simple, and it maps directly to what Jensen talked about earlier. We've got the industry's largest investment in terms of engineers and artists whose sole mission is to advance the state-of-the-art. Their sole mission is to increase the production value of games, to make games better, to solve the hardest problems, to create technology and middleware that's kind of at this intersection of engineering, art, and science, and really at the very forefront of real-time graphics R&D, which drives gaming forward.
There's no doubt that games of 10- 20 years ago were compelling, fun games, but they didn't have anything near the involvement of today's games or the realism of today's games. Part of that is the technology of gaming has advanced and production value has advanced. Gaming is now a huge business. It's the largest business in the industry. Jensen mentioned that many games are investing movie-like, blockbuster movie-like budgets in building these games. Hundreds of millions of dollars go into the production of a modern AAA game. It's an enormous undertaking, but it has enormous rewards. When games like Grand Theft Auto do $1 billion in 24 hours, I think they're kind of onto something. The output of this group, the work, the body of work that they produced, is embodied in the GameWorks libraries.
These are simulation SDKs, technologies, algorithms, toolkits. These are pieces of technology that we license to game developers, so they can integrate into their games and advance the state-of-the-art. The way we come up with these is that we work, and we have been working with game developers and publishers for roughly two decades. We know all the key engines. We know all the key engineers. We know all the key game developers because we work with them on a daily basis. We know their pain. We know the challenges that they're trying to face. We know the problems that they haven't yet solved. We pick the hard ones. We try to work at the very edge of what's possible, because frankly, that helps move the industry forward. We're always a little bit beyond what today's games are doing.
As we're trying to be there, so when the next game comes around, we've developed the technology that they're after. Lastly, we build tools. These are complex pieces of software that are being built. Video games, modern video games may be the most complicated pieces of software being built. Essentially, they sit on top of a game engine, which is essentially an operating system for real-time graphics and gaming. Millions of lines of code, all of which has to run in a fraction of a second to produce a realistic image and be enjoyable. They need tools like IDE integrated debuggers and profilers the ability to work inside of say, Visual Studio, look at your code, find the bugs make it go faster. We do these across a variety of platforms.
Our main platforms are of course, the PC and Windows and Linux, as well as Android. Our mission is to advance the state-of-the-art, to kind of push the boundaries to increase the production value of games. Ultimately, if we're successful games will continue to advance, the industry will get larger, and as those games advance, they'll need increasing horsepower and GPU richness to drive them. We've worked tirelessly with all these developers. This is just a sampling of some of the games we've shipped over the last year. This actually happens to be 12 different game engines that GameWorks technology has been integrated into. These are some of the biggest games in the industry. You might have heard of World of Warcraft. They ship with GameWorks technology in their game.
Unreal Engine 4, CryEngine, Call of Duty, some of the largest gaming franchises in the industry have GameWorks technology integrated into them. We've integrated into those core engines, which tend to be reused for many games. While this represents a dozen engines, we've worked with many dozens of games over the last year, all of which integrated some GameWorks technology to advance the state-of-the-art. That's it for the slide piece. Now we get to the fun piece. One of the most compelling things that we get to do at NVIDIA is invent the future. One of the biggest problems in gaming is lighting, particularly in real-time. Video games have gone to enormous efforts to be very clever with the way they compute shadows and the way they compute light and reflections. A lot of that is tricks, hacks, frankly.
In particular, one of the problems is that to compute bounced light, you have to pre-compute it today in games. You have to kind of set your scene up and then offline render a what's called a light map or a hemisphere of light that you can look up into. The problem with that is that as you move around in the game or the game changes or someone blows a hole in the wall and the sunlight streams in, well, the lighting needs to change, but you've pre-computed it, and you can't do that. We wanted to tackle that problem. What we came up with is a technology called VXGI. It stands for Voxel-Based Global Illumination. What we did was we created a volumetric representation of the game world using voxels. Think Minecraft, something like that.
Then we do a form of ray tracing called cone tracing to bounce light in the world in real-time, so that not only do you get the normal kind of direct light or the direct shadow that you're used to in basic games, you get interactively bounced light. So, The lighting system can be recomputed every frame. You can get accurate reflections, you can get accurate shadows, and you can get some really stunning imagery. Let me go ahead and go to the demo here and kind of step you through it. What we have here is the VXGI library that we've integrated inside Unreal Engine 4. This is a shipping game engine. This code is available today to game developers. It's being used by game developers today.
In fact, when Epic changed their business model for Unreal Engine 4, they took what is quite possibly the most powerful game engine tool, and they made it free, which makes incredibly high production value games available to literally everyone, and this technology is integrated into that platform. What you're looking at here is essentially, I'll call the diffuse component of light. Let's kind of cycle through it. Here's diffuse light. What you see here is no textures, no direct light. You see kind of the green glow and the indirect stuff. You might have seen images like this before, called ambient occlusion. In this case, we're actually computing all the colors of light, the indirect bounce. If you look at what games traditionally do with just direct light. Let's go ahead and switch over to that.
This is what a video game something like Doom. For example, might have looked like. Very dark. You get direct light. Of course, light bounces, and in the real world, when it bounces off a surface. It is influenced by that surface, and it can change the color of the light. In a traditional video game, you get only direct light. You get this kind of very dark, stark looking world, and there's no color bleeding. There's no attributes of real physical behavior. If we combine that with VXGI's indirect lighting and the direct lighting, you start to see a much more realistic world. You get light that bounces. You get green tints from the wall reflecting and bouncing onto the floor. This is all being done in real time. You get real reflections.
That little object there, as it moves around, the reflection on the floor is correct. One of the other advantages that we get with VXGI is traditional games will cheat with reflections. They'll just kind of look up into a cube map, or they'll precompute it. We can actually, because we're bouncing light a reflection is nothing more than a bounce. Let's go ahead and cycle through the different reflection modes. That's kind of the standard way Unreal Engine 4 would do things. They kind of fake it. It's kind of specular trickery. You see that little sheen on the floor, but you don't actually see the objects. Now let's go ahead and turn VXGI on. Now you actually see you're picking up some of the material. You're picking up the metallic surface off of the wall. You're getting a real-time computer reflection.
This kind of technology is integrated in games today, and I expect you're going to see games shipping using this technology by this fall. We introduced this as part of Maxwell towards the fall of last year. We've integrated into game engines now, and I think you're going to see game engine shipping with it really soon. This is pretty cool stuff. This is at the very bleeding edge. This is real-time global illumination, running integrated into the game engine at 60 frames per second. That's pretty cool stuff. While as cool as that is, that's not actually the future because that's here now. One of the great things about what we get to do at NVIDIA is we get to work on the future, the problems that are just out of reach for game developers. If we can come back to the slides.
One of the problems that's always been just out of reach of game developers is to do what I'll call accurate clothing simulation. Typically, in a game you get what I'd call plastic clothing or plastic hair for that matter. Everyone wears a helmet or they're all bald. They all look like Jeff. Or if they have hair, it's plastic hair. Sorry, that was probably a little, you know, too close to home. Their clothing tends to look kind of plastic. They wear a lot of armor and things of that nature because to do physically simulated clothing turns out to be really, really hard. The clothing itself tends to move and flow. It bounces. In particular, layered clothing is one of these kind of unsolved problems in real-time physics. Cloth is thin. It moves around.
As you move, it can interact with folds of other pieces of cloth or itself. The solver, the mathematical solution for that, has been out of the reach of real-time graphics. Well, we've actually solved that problem by actually modeling not just the cloth, but the air itself in the physical simulation. For the first time, we're going to get real multi-layered clothing simulated in real time. This happens to be a system that we built on. It runs on the GPU. It will be coming to PhysX soon. This is a little bit of a taste of the future. It uses about 400,000 polygons for the dress. It runs entirely on the GPU, and it's about 40 times faster on the GPU than it would have been on the CPU.
Let's go ahead and take a look at that. This is all in real time. Just to prove it, let's go ahead and change the cloth material of her dress. There we go. Cycle through it, you know. How about the one with the texture on it? I kind of like that one. There you go. All being computed in real time. This is the kind of basically simulation that. In fact, clothing designers use, and they did it offline in many cases. They'd render a frame in minutes or hours, and they'd render the simulation and then they'd look at the results. We're actually now doing similar style physical simulation rendering entirely in real time. The dress flows, interacts the folds. It corrects itself, you know, and then you can of course change it and stuff like that.
This is, like I said, a little bit of a taste of the future. I think in a year or two, you'll start to see games that have complex clothing that looks right as opposed to the plastic clothing and the, you know, helmet heads that everyone currently experiences in games. That's just taking that next step further towards production value and realism in games. Okay, let's come back. One more super hard problem. In fact, this tends to be a problem that is really two problems in one. Fluids. In most games, you don't see a lot of water that's interacted with. You don't see a lot of fluids that are part of the game. They'll trick it.
You'll see lakes and rivers and waters, and you'll see people walking in a river. It'll look like a wake, but really it's just a circle of textures behind. There's no real interaction there. That's because fluid simulation is computationally enormously expensive. It's a volumetric computation. Not only is the fluid simulation complex, but the ability for one simulation to interact with another simulation is a real problem. People hack it in games today. Having your rigid body, your bullets, interact with other things in the game world turns out to be physical systems that historically just don't play nice together. It drives up engineering costs and it makes the implementation of those things in the game just impractical.
We've developed a system called FleX, which is a unified physics system, which basically can build a variety of physical simulations out of particles. These are special particles, but they can represent cloth, they can represent water, they can represent rigid bodies, and they can all interact with each other, and they all run great on the GPU. What we thought we'd do is try to put together a technology demonstration that demonstrates a surface here. You'll see a bubble a water balloon with a volumetric fluid simulation inside it and then shoot it with a bullet. There's, roughly speaking, about 400,000 particles in a volume being simulated here, and we're actually ray tracing the result in real time on a Titan X. Let's go ahead and take a look at this demo. Kinda like the slow-mo bullet time. Yeah.
One of the great things about doing things in real time is you can control the time, and you can accelerate it and you can look at it from any angle. That's just beautiful. That's the kind of graphics that people are used to seeing being done kind of offline. Sometimes you'll see people doing the same kind of thing with a camera that shoots 10,000 frames a second. With the physical simulations we're able to do now on GPUs and the really high-quality rendering we're able to do now on GPUs, we can start to approximate much more realistically real-world properties. Not just physically based rendering is kind of all the rage these days but actually physically based behavior. Things will react properly, simulations can interact with each other.
With the horsepower that we're at, some of these really tough problems that game developers have found impractical, we're at the verge of making them practical and possible in games, which is pretty cool. I can't wait to see, you know, games where we have even more realistic Oh, I guess we'd use that, blood splatter. Mud kicking off of tires, you name it. Clothing simulations where the cloth can be torn and interact with the game. These are the kinds of things they're working on at NVIDIA with GameWorks. Again, kinda drive that production value up going forward. Hopefully you guys got a little taste of the future. Like I said, VXGI is really here now. That's that global illumination technology. These two technologies are really kind of a peek out of the labs, so it's kind of a special treat.
These are things that are gonna be implemented over the course of this year and probably shipping in games next year, whereas VXGI is a this year kind of thing. That's what I've got. GameWorks, it's about driving the production value of games up, making the GPU an increasingly critical part of the gaming equation, driving the industry forward so that we get realistic games and graphics in real time. Thanks. Who's Is Jeff, are you up next or is it Greg?
It's Jeff.
Okay. Next, let me introduce Jeff Brown. He's gonna tell you all about, I believe, cloud and GRID. Is that right?
Enterprise graphics. I might've stolen your thunder. I apologize for that.
Thank you.
Thank you, Tony. While a lot of us are graphics heads, you know, Tony, Greg, myself. History at Silicon Graphics and Apple and HP, and that kind of research and technology is just amazing stuff. In the spirit of Arnab's history of the Stand-Up Comic, now for something completely different, we're gonna be talking to you about our Enterprise Graphics business. My name is Jeff Brown. I'm responsible for externally what we call Pro Viz and Design, internally what we call Enterprise Graphics. Greg Estes is gonna join us. He's our VP of Enterprise Marketing, so he spans over not only Enterprise Graphics, but also HPC. He's gonna introduce a brand-new technology initiative that we believe is gonna be the future of growing our Enterprise Graphics business around technical compute or technical graphics.
Before we jump in, let me just introduce the two platforms. One of them that you know and love, Quadro. This is a historical legacy for us, a dynasty. I think we calculate that Quadro has been around for about 15 years, half again as long as Silicon Graphics, for example. 15 years of absolute dominance. It's because we are extremely customer-facing, application-facing, customer-facing. NVIDIA is the place to be if you care about graphics and visual computing, and we care about our customers and the work that they're doing. The paradox for them is they wanna push the boundaries in technical graphics, but they're incredibly conservative. These are people whose work is driving earnings of companies or research, medical, energy, et cetera.
Quadro, our traditional workstation business, it goes to market through branded global workstation OEMs. A lot of times we say that Quadro's actually constrained by the channel that it goes to the market through. Very, very serious business, very serious customers, and we bring a lot of innovation to them over the years and a huge amount of commitment. Our newest enterprise graphics platform is what's called GRID. Jensen described the challenges to bring GPUs to the data center, Power efficiency, virtualization. We had to invent ways to basically fractionalize a very, very complicated graphics engine. We had to figure out ways to deliver those pixels in a very, very low latency, low bandwidth kind of way. The two platforms, we call them virtualization and visualization, you're going to see they're very related.
NVIDIA GRID delivers the promise of Quadro in a virtualized data center, sort of way. Before we go forward, we're gonna look back at the two businesses. Enterprise graphics, we did $833 million last fiscal year. Quadro, as we've talked about really is the market leader in professional graphics. We really stand alone with that responsibility. We had a record year with Quadro. Quadro is mission-critical to all sorts of customers, and Quadro is responsible for bringing dozens and dozens of aircraft, cars, millions of medical radiology records. Very, very mission-critical, right in the heart of technical engineering workstations and applications. We're gonna bring Greg Estes up in a bit to talk about how we're gonna expand this market.
Great lead off from Tony's presentation on GameWorks. The huge benefit to being at NVIDIA with a massive research department is that we can leverage not only the economies of scale of building these processors, but we can leverage a huge amount of research. One of the holy grails, I'm not gonna give it away for Greg, one of the holy grails in professional visualization has been the ability to do something called physically based rendering. How do you create a physical digital prototype that makes it much faster to get to market, fewer iterations, and better decisions up front? The basis for this is something that's called physically based rendering. It's a concept that's been around for a while.
We're applying a ton of technology and a massive ecosystem to solving this problem and delivering that to millions and millions of professional users. That's on the Quadro side. On the GRID side, this is a tremendous anniversary today. A year ago at GTC, Ben Fathi from VMware the CTO you may remember came on stage, and we talked about a co-development to bring vGPU our virtual graphics technology into vSphere, which is the 80%, 800-pound gorilla in enterprise virtualization. Today, actually, I'm sorry yesterday, VMware actually shipped the gold bits. vSphere 6, every version. Every copy now ships with NVIDIA GRID vGPU as of yesterday. Jensen talked about the trials, doubling year-over-year, and that's really only a fraction of the customers are gonna experience the benefits of virtual GPU in vSphere.
We're going to talk about what those problems are and how that really solves the problems for those customers. We expect to see explosive growth. Last year's GRID growth was all built off of the work we had done with Citrix, going forward and they represent less than 10% of the enterprise virtualization market. VMware represents 80%. We did all that basically without the VMware engines engaged. What we're going to do now is we're going to break this up into two sections. We're going to talk about the enterprise graphics market as a whole talk about Quadro first. Want to use this kind of as a map for you all to think about who we believe our customers are for enterprise graphics. If you're visual, you could actually imagine a triangle here.
This represents the 725 million professional users in the marketplace: PCs, laptops, workstations. We believe that this is our addressable market for pro graphics, whether those are, you know, PC graphics, workstations or virtual graphics. Quadro is laser-focused at what we call the designer market. That's an installed base of somewhere around 25 million-30 million users. These users all use a tool or multiple tools. They might use a CAD tool, a digital content tool, a film tool, a medical tool, and they are doing the mission-critical work within enterprise. This has been our target for Quadro for all these years. It's a sizable total available market.
We are right now about to launch a vision for how we believe that we're gonna grow this TAM even further and our GPU penetration into this market even further. I'll come back after that and talk about how GRID allows us to address this and push down and expand our overall graphics enterprise graphics marketplace. Without further ado, I'll introduce Greg Estes.
Thank you, Jeff. I'd like to talk about this area of physically based rendering, which is what we think is gonna be a great opportunity for us to address a larger part of the market and bring something that's been the dream of designers to the marketplace. People have been working to be able to visualize the creation that they wanna build for many years and have been able to accomplish that in limited ways. But those limitations have caused them to sometimes build creations that the first prototypes weren't what they expected or they created a building that had lighting that had caused problems in certain ways. If you take a look at this beautiful image here, you can think about I guess there's three ways maybe that you could create something like that.
One could be if it already existed, you could take a photograph of it. If it didn't already exist, you could have an artist design it, use Photoshop and other painting tools to create it. Or you could model it in computer graphics and then try to understand what it would look like when you went to go build it. In doing that last one, you would be able to then take that model and use that model in your manufacturing process or your building process to construct it. Ideally you would have a visualization that would allow you to understand when you went to go build it, what it would really look like. To do that involves some fundamental things. First, you have to, of course, have the model and that could be something simple.
If you look at our customer Boeing with the 787 Dreamliner, there were more than 2.3 million parts to that airplane. The models can be extraordinarily complex. The second thing that you have to have is you have to understand what the materials are, right? The materials are going to affect when light strikes it and interacts with that material certain things happen. That energy is either absorbed or it's reflected or refracted. That light, as Tony mentioned and showed you, bounces around in different ways, which is all about not only what it's like to feel when the product is built, but also what everything is gonna look like because that light bounced around.
If you look at the other example there, which is a real place actually. It's in Portland, called the Portland Armory, where those lights are and what it looks like if you're building a building, it could affect the safety or if you're building a product. It could affect how beautiful it is or the feel that a customer will have for it. All of these things interact together to come together to allow you to understand what your products are gonna look like, which is of course important to the 25 million designers that Jeff talked about at the top. Last year at GTC, our customer Honda, for the first time at great expense, more than $2 million, and great effort, was able to visualize a car on stage with Jensen in real time.
An entire automobile, slice it arbitrarily, understand what it looked like, and be able to see that and manipulate it in real time, which if you can imagine if you're an auto manufacturer is an enormous advantage in being able to bring products to market not only quicker because you can build fewer prototypes but less expensively and ultimately more accurately. You can imagine what would happen if, for instance you built a prototype and you found out that when light hit the windshield and with a certain material for the dashboard, that glare occurred. Well, you would wanna know that before you tooled anything and before you built it. Having an accurate simulation of the rendering of how light works and interacts with those materials is critical to that process.
We have a vision of bringing this capability not to one customer with a TAM of one much as we love them, but to millions of designers. We're using multiple of our technologies, software, Quadro hardware, beyond Quadro into the data center that we'll talk about more in just a moment all together to have these products work together and then work with a series of partnerships to bring this capability again 2 millions of users. It starts with bringing our rendering software, our physically correct, physically accurate, physically based rendering software called Iray, which is already used in some design tools. We're gonna bring that broadly across the market. The second thing that we're gonna do is to create what we call a Material Definition Language.
What it is a way to describe not only what a material looks like, but the height of the material, it's got surface properties and how light will interact with it. Now we have something that measures light correctly, and we have something that measures materials correctly, and we're gonna put that into a format that is usable across all of these applications that we're going to bring to the market through Iray. Then we're gonna make that scalable from laptops into high-end workstations and out of that and into the data center. You can use it if you're sitting in an airport somewhere or in a hotel room, or you could be able to tap into your network and multiple high-end GPUs rendering a very complex design in real time.
The heart of this is Iray 2015, which again is our physically based renderer which calculates how light works. The second part is this new Material Definition Language. You'll be able to create materials and there's a file format that allows you to bring those materials across into other applications. Now, excuse me, that may be something that you would have assumed would already work in the marketplace. It doesn't. Today, it is generally not possible to have a material work in one of these 3D design tools, then send it over to another worker in another part of your company, have them use a different tool for that and get exactly the same result because each of these applications has their own way of doing things. What we're doing is bringing to the market through Iray is to have a common way to do that.
Because it's all physically correct, the materials are correct, the way we calculate light is correct, you will be able to get a correct answer which is critical for these 25 million designers. We're gonna do that in a way that's built specifically for Quadro and Quadro VCA. The Iray software knows that you have a Quadro in the system and can scale seamlessly and elegantly to multiple Quadros if you have a workstation that's able to support. The highest end workstations now are able to have up to four Quadros in them. Then go outside of the box into the data center with our new Quadro VCA product, which has the power of eight of our highest end Quadro M6000s, roughly the equivalent of a Titan X, the professional version, if you will.
They can put eight of these in this data center ready packaging and then scale that from there. As a whole then that again brings us the ability to have interactive physically based rendering. We're working with a whole series of partners throughout the year throughout 2015. Partners like Autodesk that have architectural tools like Revit and design tools like 3ds Max, those that are used by entertainment customers and auto styling like Maya. You've got Dassault CATIA. CATIA is a great example. Dassault is one of our great partners, and they have multiple products that we work with in them. CATIA is about the highest end styling and design system that you can have. If you were gonna go build a nuclear submarine, you would use CATIA to do it. Excuse me.
Most of the auto manufacturers use CATIA or Siemens NX, which is one of their main competitions. We're working with the big boys and the most complicated problems, but we're also working down in the volume parts of the marketplace as well, like 3ds Max, which is used by hundreds of thousands of users and even Daz 3D in the lower right there. That tool is actually free. We're building this community and this ecosystem all around this idea of bringing physically based rendering to designers. The lower bar there is a very important part of the equation. Remember I said materials were super important part of this because that's how you tell how light is gonna react with things, which gives your correct result.
Our partners there, I'll start with Oldcastle in the middle. Oldcastle makes a library of different materials. They think of themselves as being the envelope around architecture. If you were building a building and you wanted a certain kind of glass or perhaps masonry, they have a library of materials that you can go get from them and buy them. Then when you put that in your model, it's physically accurate. You know what it's gonna look like. Allegorithmic on the right, it makes a series of software products that allow you to create your own material.
It's used in the game industry, for instance, because if you wanted to do a minotaur and it was drooling, there's no material for minotaur drool, or any kind of a custom design that you need to be able to do, right? You can do that using these tools. What's super exciting for particularly auto manufacturers is our partnership with X-Rite. Do you know X-Rite? You might know their subsidiary Pantone very well. They are the world experts in color and materials. X-Rite has a hemispherical scanner that not only measures what the light that comes off of the material, but the height map as well. It knows the texture of it as well.
If you're, I'll make up an example Ford, and you're designing the next generation Ford Fusion, you can take the materials for the seats, you can give it to X-Rite, they can scan it, you can get a digital file back, and now you have exactly that material that can be taken into the digital realm. Again, through the use of Iray and our Material Definition Language, it can go across all of these tools and your designers can have that exact material. You can make that specification for all the different partners that you're working with or all the parts of your design system. All of this is working together based around our new Maxwell architecture to accelerate it. A great example of that, a real simple benchmark is this particular scene of the BMW Z4.
It was more than twice as fast, 2.4 times faster when using Maxwell and this new version of Iray versus last year's version of Iray and our previous Kepler architecture. The results for our early customers for this have really been outstanding. Here's an example from Audi: "With NVIDIA Iray, the time I need for a 3D model to a meaningful image is extremely short, and the creation of physically based renderings, I've never had these fast turnaround times." We're providing real value to these customers that have these real needs, and we're doing it in a way that is gonna expand our market at the same time. Value for us and value for our partners and value for our users as well. We have an expanded market opportunity here.
We've got this broad ecosystem of partners, and we're scaling it from the laptops to the data centers. I'd like to turn it back to Jeff to talk to you about the rest of our data center strategy and our second growth strategy, which is around GRID and bringing this capability into the data center. Thank you.
Thanks, Greg. Just one note, we're not gonna do these demos here, but off on the exhibit floor, which we're gonna take you to, I believe right after we're done, we've got examples of Iray physically based rendering plugged into the applications that Greg just talked about, as well as a pretty amazing spectacle we call the Death Rays. We invite you to check that out. The segue here is into data centers. You can see everything that we're doing on the Quadro side, and we believe we're very enthusiastic based on feedback from our customers. This is gonna drive GPU growth in the professional space. Everything we're doing here basically lifts and goes into virtual workstations, what we call vGPU, virtual GPU, which is the NVIDIA GRID enterprise graphics virtualization platform.
Back to this icon here. Everything we've talked about so far with respect to designers where Quadro, its addressable market sits, that 25 million user designer. We've never been able to penetrate down into these sub-segments of users in the enterprise or business computer space. Virtual GPU gives us the ability to bring Quadro quality application performance compatibility into a virtual environment. We're partnered with people like Citrix and VMware and Microsoft, they've been coming from the bottom. Server virtualization and the first generation of VDI has really saturated this task worker space where the applications are not graphics rich. They're primarily web or text-based. We're looking at a couple of different segments now where we believe that GRID enterprise graphics is gonna penetrate.
Some of our early customers certainly are workstation users that wanna take advantage of the benefits of virtualization, which is to move to the flexibility using any device, remote access security. Part of the GRID opportunity matrix certainly is designers. What we're really excited about is we're seeing very early indications. I shouldn't call them early anymore. When you're up to 2,000 trials, it's no longer early, but within this power user space. Roughly the way to think about it, for every designer for every engineer in CATIA designing engine for Boeing's Dreamliner, there's 20-30 people downstream that use that data in some sort of way, documentation, service. There's whole sectors of users that are smaller businesses or more utility kinds of applications in the enterprise.
This is where Autodesk AutoCAD sits, Bentley MicroStation. Our initial target for GRID is this 225 million user base. Again, our motivation is to push down. VMware's motivation is to push up, and that's why we are so well synergized in the market. Speaking of that, VDI is not new. Virtual desktop infrastructure is not a new concept. It has, however been a failure outside of that task worker marketplace. You ask yourself, why is that? The promise of VDI is so great. The promise of VDI to the enterprise is mobility and freedom, being able to work anywhere, being able to get more flex time out of your users, being able to manage your contractors and your interns with more security.
IT managers love it because they're not managing thousands or tens of thousands of workstations and PCs and laptops. Users can bring their own devices, and they get managed through a single pane of glass, is what enterprise IT calls this. Everything's managed as a virtual workspace. Finally, security is utmost, you know, on the radar for global companies. Security. Jensen talked about the benefit of virtualization is moving the data and the compute to the data center. The advantage of that is performance and flexibility and provisioning, but also guarantees you security. Your data never leaves your corporate data center. All you're sending down the pipe are pixels. You're sending pictures down the pipe. It's the ultimate security for your most important information your car design your aircraft design, your Star Wars next.
Here's where VMware, Citrix, and NVIDIA are just so tightly aligned. Now last year, as I mentioned we made a commitment together with VMware to bring vGPU into vSphere, which is their server virtualization hypervisor. As of yesterday, we achieved that goal. Every copy of vSphere ships with NVIDIA's vGPU. Everything we've done so far in terms of lifting GRID off the ground has been done basically with our very good friend Citrix, but they represent about 8% of the market for hypervisors. As of today, our ability to address the market has increased tenfold. All of our trials, all of our initial pilots, all of our initial deployments have been either early access or with Citrix. That gives you an idea of context. They are the gorilla.
They've now incorporated vGPU. We've got a ton of trials going on. We're well past 2,000 trials that we know about. We've gotten to the point right now where we're not only managing our lighthouse and trial accounts, we've got lots of channel partners, distributors, system builders, the OEMs, all their sales teams enabled. We're really starting to see lift with the business. What we wanna do real quick is jump to a demo. There's a couple things we're doing with VMware in terms of go-to-market. We went through a really great early access, and it was called Direct Access Program with VMware. One of the ways that we got enterprises and users to experience GRID before it was available widely was we set up something called Try GRID.
We literally set up our own virtualization stack with our own virtual workstations. We invited users and enterprises and software developers to try GRID. It was a trial. You could register through a very simple web form. We give people different access times. VMware loved that idea because it is actually relatively difficult to set up a server with all the software, you know, all the security. Together with VMware, actually today, this very day, we're launching NVIDIA & VMware Test Drive. Try GRID with the vSphere stack. This is gonna be open. We had about 20,000-25,000 users try our own proprietary Try GRID.
Going forward, you know, amplified through VMware, we're really optimistic that this joint. Try GRID vehicle for people to test and trial is going to be a phenomenal lead development and actually initial qualification platform as well. This is co-hosted, co-marketed. We collect all the leads and share them with VMware. We distribute those out to our channel, either our VARs, systems integrators, OEMs. Milan, if you could flip over. What we are going to show you is we are going to show you what Try GRID looks like. You very quickly register, you submit, very low bar to enter here. The benefit here is we gather a lot of leads. Milan is going to show you what Try GRID looks like. He's on a Macintosh. He's on a laptop.
He has launched a VMware vSphere vGPU Windows desktop, fully accelerated. To Jensen's point, here's a bit of the inception mind bend here. Sure, he's connected over this hotel's really pretty poor Wi-Fi connection that virtual workstation, that virtual machine is connected at 800 MB, 800 Mb, because it's sitting basically on the hub of the internet. His access to data and to files and applications is lightning fast. He's got a three year old MacBook Pro running a lightning fast PC. He gets application availability and compatibility while he's running on his machine. You can see that here. What he's running here is a very simple, it's called WebGL, which is Google's next generation browser graphics engine.
Even anybody who's using a browser is starting to feel the need for graphics, even in your everyday use. That's what we call WebGL. It's a benchmark for running browser graphics. Digital Ira was the star of GTC two years ago, I think. This was running on our highest end TITANs two years ago. Now you're seeing this run on a virtual workstation. This is a fraction of a GPU running up in the cloud giving you performance that's actually better than that keynote performance for Digital Ira was just two years ago. Making this kind of performance accessible in a virtualized environment. Thank you. What we're showing you here is an application from a company called Esri. Esri, it's called the GIS application.
You may have heard of them. Really famous. Every department of transportation or urban planning, government uses, this is kind of Google Earth on steroids. The data here, this is a multi-gigabyte data set running on his MacBook Pro. Access to a huge amount of data. This is the city of Philadelphia, I believe. You can see this is completely a sub-meter LiDAR topography data. Used by all sorts of government analysts or, you know, local department of transportation workers. If you're gonna drill a pipe or find a main, this is what you're gonna use. Again, running virtually over the net off of Try GRID. Esri is so excited about this. They're using this actually to beta their next generation software.
Very cool benefit to the ISV because they don't need all of their user base to have the latest, greatest hardware, and they can show their latest, greatest software at full throttle. That is Try GRID based on vSphere and vGPU. This goes live. If you wanna try it, we can provide you with the URL, and you can experience it for yourself. It's a really, really amazing tool to show the world the value of virtualized graphics. Thank you, Milan. Placeholder. Real quickly, what I wanna take you in sort of a staccato manner through a couple of case studies. We've been managing through our direct business development team with VMware a series of lighthouse accounts, and wanna kind of illustrate the use models and the kinds of customers that are seeing value out of virtual graphics.
PSA Peugeot Citroën, one of the largest auto manufacturers in Europe. They've got sites in Spain, Vélizy , France, South America. Their CTO is actually here. He loved listening to Elon Musk. Their objective was they've got all these distributed sites. They wanna manage one data center. This is a very traditional Quadro customer running CATIA. Their goal really was remote worker access. This was productivity of their engineers over remote sites, global design teams. They achieved that with GRID. These guys were very early pioneers of graphics virtualization. They tried everything, but only GRID really solved this problem for them. That's one very sort of classic, obvious customer case. This one I love. These guys, STV a global architecture firm based in New York and Philadelphia.
They did the World Trade Center in New York. They did the Denver Airport expansion massive projects global footprint. They were not a Quadro user. They were using an Autodesk application. Still had some graphics, you know, intensive needs they were not a Quadro user. They weren't buying workstations. They didn't need to, they need GRID gave them what they needed. It gave them the application compatibility. It gave them the productivity because their teams, their architecture all over the world. They set up mobile offices on site reduced their costs, increased their manageability. All in all, you know, GRID, massive breakthrough for STV. We love these guys. They're actually presenting here as well their second year in a row presenting their success story in one of our breakouts.
Another one of my favorites, Saint Lawrence Academy, which is a pilot school part of a local Bay Area, Catholic diocese, K through 12 example. In this case, they're teaching their students Adobe, Creative Suite, Creative Cloud, photo editing, video editing, web development, very GPU-intensive application. What they can do here, the beauty here is that with GRID, they can teach their students, they don't have to worry about viruses. They can basically clean the images, you know, at the press of a button. They can also give their students access to way more resource than these students may either have or they may have the budget for through provisioning. Students can also get access to the virtual machines from home.
You know, again, we had no Quadro business in the education market to speak, and this is a phenomenal example where virtual GPU brings a lot of value to K-12. They're the pilot school for this entire Catholic diocese district. Another one, all these are really amazing stories. MetroHealth, which is a HMO in Cleveland, Ohio, focused on geriatrics. They actually implemented VDI. They were an early Citrix VMware success story. What ended up happening is they upgraded their software, their radiology software, and their entire system broke. It just broke. They could no longer view the radiology images. This was a higher resolution version of the same application, their EHR records system. What they did is they had existing server infrastructure. They plugged vGPU in there.
They ran a beta version of VMware, and they were up and running. This solved a problem where they just fell off a cliff and their doctors weren't able actually to do these diagnoses with their new software. Really, really great story here. You can see new industries, new markets, but you're seeing a very consistent theme in terms of the value proposition for virtual GPU. They see, you know, the kind of user base expanding as well through this technology. That's great. Obviously, you can tell we're really excited about physically-based rendering expanding the traditional Quadro market. I wanna note that everything we do in Quadro lifts into our virtual workstations as well, so there's this secondary benefit. GRID is at the point right now where it has lifted off.
The business is at a doubling right now. The number of customers in trials, pilots, deployments has really achieved lift off. It really is the essential catalyst to make this real. Again, we're super well aligned with VMware with Citrix, with all the system builders. It really does expand our addressable market for virtual GPU for graphics outside of our traditional enterprise graphics target, which is that designer market. We're seeing early indicators that that is in fact true. We're very, very optimistic about the overall enterprise graphics market. Again, I'm gonna hand it over to Arnab and invite you to come see the demos on the floor and thank you for your time.
The bad news is we're only at the halfway point, but the good news is Jensen already did everybody's presentation, right? It's terrible. Anyway, we'll have to work on the comedy next time a little bit more. Right now we have the break. We're going to walk over to the demo room to show all the demos, but before that we have coffee and drinks. If you guys need that, we can go get that. Thank you very much. Good afternoon. You guys can all sit down. Jensen can do what he wants. It's his company. Anyway, we're just starting out the H2 of our day. We have three more presentations, and we'll have Q&A after that. I'd like to introduce Shanker Trivedi to talk about our HPC and cloud opportunity. Thank you.
All right. Thanks very much, Arnab. As Arnab said, I'm Shanker Trivedi. I'm responsible for our enterprise sales and industry business development organization. All of the things that we do in front of the customer, all customer-facing activities. I joined NVIDIA about six years ago almost to the day. My mission has been to grow our server data center business. Basically, I came to NVIDIA with about, you know, 25 years of experience in enterprises and data centers, server systems, and middleware, working for companies like Sun Microsystems, IBM Corporation, and Fujitsu Corporation. Additionally, I spent about three years as the head of marketing and corporate development for a small cap enterprise software company.
It was a great privilege and very exciting for me to come and join NVIDIA with a view to sort of growing our enterprise business. We started with HPC. I work very closely with Jeff Brown and Greg Estes on the graphics, enterprise graphics side. I work very closely with Ian Buck and Sumit Gupta, who are on the HPC and computing side. Today, my mission is to talk to you about our plans for HPC and cloud. You know, last year was a fantastic year for us. Tesla had yet another record year. We grew, our revenues are now $279 million. As Jensen mentioned, we grew about 53% year-over-year.
It was a phenomenal year. The Kepler architecture GPUs are just awesome for enterprise computing, for data centers. In the previous generation of Fermi, we made very, very significant strides, but Kepler has just been transformational in terms of its architecture. It's designed for the data center and for enterprise computing and doing multiple tasks at the same time with Hyper-Q and, you know, multiple parallelized threads. It's been phenomenal. Along the way, you know, when I started six years ago, we had no business at all in data centers. There were no servers. You know, we systematically and methodically have created this new category of accelerated visual computing for data centers. It has been a journey. We've had to create the market.
We've had to create the ecosystem. Today I'm gonna share with you our plans, our strategies for growth in two kind of segments. One being high-performance computing or HPC, the other being cloud computing. If you think about it, you know, Jeff Brown talked about the third leg of the stool, which is the enterprise graphics virtualization. All of those give us a strong wedge into the data center. The data center worldwide overall is a, you know, about a $50 billion market. We are poised to address at least a $5 billion growth opportunity. As, I'll share with you. Having said that, it all starts with as Jensen put it so eloquently both in his keynote and as in his introduction, it all starts with a developer.
Without a developer, there are no applications. There are no uses for whatever product you're making. There are no customers, and therefore there is no data center. You know what's fascinating, when I first joined, you know, we had about 400 people or so at our GTC. It was like a science fair on steroids, with lots of poster boards and lots of excitement. Today we have, you know, 4,000 or so people at GTC. The way I kind of think about it, and GTC. Of course, is a developer conference.
The way I think about it is, you know, this is all about helping people write their master's thesis, helping people defend their PhD thesis and produce, you know, awesome research and helping people do postdoctoral work, you know, of the kind that Jensen was talking about in his keynote, which all about creating a new type of application, a new category, a new way of doing work. Okay. You know, what's interesting is a lot of people have this conception that, you know, it's all about CUDA, and CUDA happens to be some language. It could not be further from the truth. It's all about a platform. What we've been doing is making it easier for our developers to capture the, you know, to make parallel programming easier.
We basically, everything we do on the developer side is all about helping our people, helping and making it easier to program the GPU. We have, you know, ways of making ordinary programming languages like C and C++ and now Python and you'll see IBM is doing Java for example. All these programming languages, they need to be accelerated. To strengthen our own organization, we bought PGI, a compiler company. Compilers are critical to our future as a developer platform.
We've propagated the standard for accelerated computing called OpenACC, where you can take long, complex, difficult legacy codes and relatively easily parallelize them using directives so that they can be run on a single system, on a multiple system, multiple GPU system, on a cluster of GPUs, and indeed, on multiple racks of processors running in a data center. The second point there, which is all about development tools, you know, we have great profilers and debuggers, and these, together with our partners' products, you know, companies like Allinea and Vampir. Our partners' products have to work not only on a single node, on a single GPU, but also on a single node, across multiple nodes, on the cluster, on the rack, and on the data center. These things need to scale when you're solving real problems.
Lastly, as you heard, we also make it easier for people to program using a huge number of accelerated libraries. In your industry, random number generation rand is a really important mathematical function. We provide an accelerated version of rand called cuRAND. We do linear algebra. We use deep neural networks with cuDNN and Fast Fourier transforms and so on and so forth. The bottom line on all this is a few years ago only a few years ago, we had just a handful of applications that were accelerated, that could take advantage of the GPU in the data center. Today, we have over 300, and the rate of adoption is increasing. You can feel it every day. You can see what's happening.
A startup, you know, CEO just met me outside the room, and he expressed how he was working with Apache Spark MLlib and accelerating that. That's a new instance of GPUs being used in an exciting area. Developers are everything. Developers are embracing the platform. As a result of which, the first market that we chose to go after was the high-performance computing market. This is, you know, a $15 billion server market, according to IDC. Our estimate, what you see is because of the huge number of new applications that are coming on stream, because every time somebody buys a new HPC system, the latest research shows that more than half of these new HPC systems will have accelerators installed in them.
Our view is that the opportunity for us is conservatively probably around the region of $3 billion. You know, I like to always speak, you know, doing market research and, you know, having IDC or Gartner or somebody at Dataquest validate these market sizes is one thing. The other thing is the practical side of it. When I sort of go into a project and I say, "Okay, if the customer is spending $100, am I getting $20?" If the answer is I'm getting $20 or more, then I think $3 billion is a good target market opportunity for growth for NVIDIA. In many cases, we get more than 20%. Of course, we don't always have accelerators.
You know, I think $3 billion is accurate and it's very, very significant. We're best known for, you know, the headline always is the fastest computers are accelerated by NVIDIA Tesla. Last year we announced, you know, our new partnership with IBM. You know, that was the new news at last year's GTC. This year, you see the fruit of that partnership. The U.S. government has decided, after a long, important evaluation, to build the two most powerful supercomputers in the world using NVIDIA's Tesla architecture. These systems will incorporate, obviously, an IBM power CPU, NVIDIA's future codename Volta GPU. The most important thing is they will incorporate the high-speed interconnect, which is called NVLink.
These systems will be at least 100 petaFLOPS if not 300 petaFLOPS in terms of performance and they will come in at an energy level below 10 megawatts. If you do the math, you know, you're looking at one of the most energy-efficient supercomputers in the world, way more energy efficient. Today, we already have, you know, NVIDIA Tesla powers, the most powerful supercomputers that are the most powerful energy-efficient supercomputers. This one will be even more energy efficient than that one. The key point here is, it's very, very high extreme performance combined with extreme energy efficiency. That's, you know, that's just the highest end of the pyramid, so to speak. That's the pinnacle. That's the lighthouse.
You know, what's also happened is we've democratized high-performance computing. The way I like to think about this is, you know, in 2009, in my alma mater, which is the Indian Institute of Technology, and I went to the Delhi campus. In all of the Indian Institutes of Technology, there were no supercomputers zero supercomputers. Today, if you visit India, every single Indian Institute of Technology has a local supercomputer, right? Whether you go to, you know, you go to Clemson, you go to Indiana, you go to Purdue, you go to Seattle, you go to San Diego State, et cetera, et cetera. Everywhere around the world, in every nation, in every university, in every research laboratory, because it's now more affordable and more accessible, researchers and research communities are installing mini supercomputers.
A couple of weeks ago, I was with my life sciences team in Washington, D.C., we did a seminar at the National Institutes of Health. You know, National Institutes of Health invest about $5 billion in grants to researchers, both intramural within the campus and outside the campus at various medical colleges and so on. You know, this lady had called a seminar and she'd invited all the program managers. About 70 people turned up into this beautiful conference room. We went around the room and, you know, there were so many people said, "I'm in cancer research. We have 40 K40s.
I'm doing the genomics department and we're just installing 75 K80s. You see my point that, you know, around the world, research laboratories, governments, are adopting Tesla high-performance computing systems increasingly. What's most interesting is it's not just in government. You know, even two years ago, in industry adoption of Tesla was very, very small. Handfuls of systems. Today, if I go around, you know, the manufacturing industry because of the availability of applications, such as Ansys and Abaqus and Ansoft and, you know, CST and Agilent EMPro and so on. You know, the adoption of Tesla high-performance computing by manufacturing industries is increasing. Not all of these companies have big systems right now, you know. Because they will always follow the method that Jensen talked about.
You know, they start with a trial, they then go into a pilot, they make a budget, you got approval, you go into a deployment over three-five years in your data center. These are long adoption cycles, but all of these manufacturing companies are using Tesla high-performance computing. For example, at the highest level to design and make and simulate their new products. In financial services the story I like to tell is, you know, it's initially taken on by people who have the most pain. One of your colleagues, actually a bank in this room, had so much pain they had to complete their risk management in eight hours because you do around the clock, you know. You have to complete that risk management calculation within eight hours.
This particular bank was using 4,000 of its servers to do that calculation every day for eight hours. Because the calculation was becoming more complicated, it was taking more than eight hours. They had to put another 400 servers to get it done and this could just go on and on. Because this gentleman had so much pain, they decided to replace this estate and use 400 Tesla accelerated servers. 4,000 servers down to 400 servers. Saved themselves a lot of money, of capital saved themselves a lot of energy for deploying the system and gets his work done on time. That's kind of the typical adoption of Tesla for high-performance computing, both in governments and national institutions and education as well as in industry.
You know, the reason people adopt Tesla is it's not just a graphics card, you know? When I started, you know, and we had. Like, four servers and I went to one of the other big banks, you know, we had four models of servers that supported Tesla. I went there and I sat down with this gentleman and he told me in the most colorful, flowery Irish language that my Tesla-based server would never enter into this bank's data center. Because at that point of time, we had no way of running it in a cluster. We had no way of integrating it into the data center infrastructure management stacks. We had no way of doing management alerts and diagnos ability and reliability and utilization and all of the attributes that a data center manager requires.
The way we now think about it is it's all about the data center platform. We accelerate. We provide facilities for our customers, our server customers, to accelerate the performance of their application on the fly. It's not just the raw performance you talk about. We can do energy limited performance. We have this feature called GPU Boost. It's not just hardware, but hardware and software that drives system performance. These systems, the second big problem is you have to move data or programs across the nodes, across the network. The way you work with the interconnect on the system becomes extremely crucial.
A key thing that we do is we have this facility called GPUDirect, which does RDMA and allows you to transfer data very, very quickly from one subsystem on the data center to another subsystem. People can orchestrate it and manage it and deliver predictable performance to their customers. Lastly, we fit into all of the infrastructure management. In infrastructure, you need to think about two things. One is the actual hardware, you know, the systems and the network and the utilization and the metering and reliability and serviceability. It's the application deployment, the application provisioning, the application management. You know, if you now need to reboot an application, you can't turn down the whole system. You've got to provision it dynamically, restore it with checkpointing, and so on.
We, you know, in the five years, we have developed Tesla into a platform for accelerated data center computing. That's the reason why the adoption of Tesla in data centers is increasing. One other thin g for some customers, we are offering even enterprise support services. These applications are now so critical, so mission-critical, that people want to contract with us for the support services so that they're guaranteed a response time. They know when the fix is gonna happen. We back support them for, you know, multiple generations of software stack and so on. We're building up a support services business. The upshot of all this is we've moved from a handful of servers to all of the world's top-selling server OEMs supporting Tesla.
All of the leading server companies in the U.S. like HP and Dell and IBM and Cisco. All of the specialist HPC companies that build supercomputers like Cray and Fujitsu and Bull and SGI. All of the major Chinese server companies like Huawei and Lenovo and Inspur and Sugon. Also all of the major Taiwanese companies like Tyan and, you know, Supermicro is both in Taiwan and here in the U.S. and ASUS and Quanta. Today you can get a Tesla-based data center server or a rack from just about anybody. We have like, more than 100 server models with more than 350 options on those server models to configure Tesla for your data center rack or server.
We not only do we support, you know, the standard x86 architecture, you know, hundreds of these servers, increasingly, we are supporting the Arm processor architecture as well as the Power processor architecture. What I mean by that is a Power processor gets connected over PCI to a Tesla GPU, and the applications are accelerated. Today you can buy, you know, Power-based servers with GPUs from IBM and Tyan, and you can also buy Arm-based servers from that incorporate processors from companies like AppliedMicro and Cavium, and buy them from system vendors like E4 and Cirrascale. My prediction is that the number of providers of non-x86-based servers is going to increase quite dramatically over the next few years.
Above all, the common thing across all of them is the applications are accelerated using the developer tools and the developer platform that I talked about in the second chart. The thing that's new and really exciting is whenever you look at this, you know, Jensen mentioned Amdahl's law. You always have to have a balanced system architecture. One of the key bottlenecks in the servers has been the PCI bus. We have alleviated and upleveled that bottleneck dramatically with the introduction of NVLink. Now we are reaching out into the market to all of the different server OEMs of various types to increase the adoption of NVLink.
With NVLink, we will now start to be on the server motherboard, it'll be much, much easier to plug in a Tesla processor like a Pascal. For example, into the server processor. IBM has already announced Power based servers with NVLink and Pascal. So, they're actively taking them to their customers. Today, Quanta Computer QCT, Quanta Cloud Technology, announced an NVLink-based server, the first-ever x86 server, you know, available with NVLink. The reason why Quanta is significant is because Quanta is an important provider of systems to cloud computing companies. We'll talk about cloud computing in a moment. I now wanna just transition into why do customers buy Tesla? What's driving the adoption? Why are they buying? In high-performance computing, the primary reason people buy is to accelerate their scientific discovery.
You know, this great example from Nature a couple of years ago is all about taking the HIV virus and actually understanding its detailed molecular structure. I think last year, they used a supercomputer to get a detailed molecular structure of the H1N1 virus, you know, which is the bird flu virus. Last year, the bird flu affected large parts of China and Southeast Asia and allowed drug companies like Glaxo to put out variants of Tamiflu to market faster to prevent the spread of the disease. This year, the same problem is afflicting India, and the local drug companies are using this great science this great scientific discovery to actually save people's lives, okay?
That's kind of the overarching reason why people buy, and we can call, you know, It may not be scientific discovery, it may be technical discovery, it may be design discovery, but at the end of the day. It's helping the scientific and technical community, the creative class, to discover more things faster, to discover better things faster. The economics of the purchase decision are, I've summarized them in the following ways. First of all, because the application works much faster, you need fewer servers, right? Because you need fewer servers, even though each server is a little bit more expensive, you end up spending less money. Your capital outlay, your total cost of acquisition is lower.
The other thing that happens is because these servers are significantly more energy efficient, you end up consuming less energy. As many of you know, the most, the highest cost of operation of a data center is the energy cost. If you save energy cost dramatically, that's a big reduction in the total cost of ownership. More often than not, you end up doing a lot more work, right? In this example, and these examples are meant to be illustrative, you know, every customer won't have the exact same results. This is an example of the AMBER molecular dynamics simulation. Basically, you know, the simulation is of molecules, and the output is the amount of simulation in nanoseconds per day.
One day of using this computer, this server, you know, the 32 servers, in 1 day, it'll simulate 58 nanoseconds of how the molecule is going to react when reacting with another molecule. In this example, you know, your cost goes down. Before Tesla, you know, your costs were $200,000. You bought 32 CPU servers. You spent 21 kW of electricity, and you did 58 nanoseconds per day of work. After including incorporating Tesla, you spend $14,000. You buy one Tesla K80 server. You consume 1 kW of energy, and you do 220 nanoseconds a day of work.
About four times the amount of work. The whole theme here is it costs you less because your application runs faster, you have fewer servers, you consume less energy, and therefore your operating systems operating cost comes down. More often than not, you end up doing more work. Now, the amount of more work, you know, in many of my customers in supercomputing and in government, the way it actually works is they have a fixed budget, you know, $20 million or $8 million or $5 million, whatever. It's a fixed budget, they're trying to optimize their acquisition cost and their operating cost within that fixed budget. With accelerated computing, they can get a lot more done for the same amount of money, and that's the significant value proposition for Tesla and HPC.
Now what's happening with, as you see with somebody like Quanta saying, "Hey, we're doing NVLink for our new servers." What's happening, as Jensen spoke about extensively is the big thing that's happening for cloud computing customers is deep learning, and it's best exemplified by the Google Brain application. The Google Brain experiment was actually done over 1,000 servers, about 16,000 cores. It cost about $5 million out of Google's estate and consumed around 600 kW of energy. In that, they did about 1 billion neurons. I think each neuron was connected to 1,000 other neurons. That gives you an idea of the scale of the problem.
Using these deep neural networks, using convolutional neural networks, this is of course, Andrew Ng's famous paper. Just one year later, they simulated 11 billion neurons. Okay? 11 billion neurons using only 16 Tesla servers, 64 Tesla processors, and the cost was like $200,000, right? You can kind of see, you know, the same theme. You know, you save a huge amount of money because you can do, you know, you can do it with far fewer servers consuming far less electricity, you end up more often than not doing more work. By the way, in this example, I put it down conservatively at about six times the amount of work. There's some argument, you know, between engineers as to whether it's 11 times the amount of work or four times the amount of work.
It doesn't matter. The key point is it is more work for less money. Notice, you know, once you're down to, you know, $200,000, you've started to democratize this, as Jensen described in his keynote. You're making it more accessible to everybody. What's happening is deep learning. The trend is really, really clear. You know, what's driving deep learning is, first of all, the maths exist. Secondly, the applications are now accelerated with the GPU, right? You know, it is possible to do deep learning using an accelerated application. Secondly, this is probably the most significant there's huge amounts of data that are now being made available to people.
You know, everyone knows sort of like, you know, I think it says around 10 exabytes of data is uploaded every day to the internet. You know, there's just huge volumes of photos and, you know, half of this data, by the way, is video and YouTube and so on and so forth. What's equally important is, you know, retailers have data about your shopping habits in store and online. It's the data volumes are just ginormous. If everyone on the planet had their genome sequenced and every genome, if you did the 26 chromosomes, and every one of those chromosomes, by the way, today, there's a paper today about the origami of the genome and how they've used the GPU to really visualize that genome. The datasets are huge.
The United Kingdom has said every single cancer patient's genome will be sequenced and will be provided in an anonymized public database for all researchers everywhere to use as a contribution to humanity so that, you know, people can find personalized cures for cancer or genetic issues, and so on. These datasets and the combination of that with this amazing deep learning algorithms that are now accelerated together with the democratized power of the GPU's computing capability, I think is gonna drive deep learning significantly. Let me illustrate to you. You know, unfortunately, IDC or Dataquest or, you know, Gartner or Forrester does not have a market study with statistically significant data to show exactly how big this deep learning market is.
Let me give you a sense of what it is. You know, it isn't just about recognizing cats. You know, some people have this soundbite, "Oh, yes, the Google Brain experiment, and, you know, what they figured out was the computer recognized that this was a cat." I mean, that's not the significant thing. Let's take, for example, facial recognition. If you've been following the Tsarnaev trial, you'll see that the entire sequence of what happened on that day and how the terrorist bombings were done has been stitched together using multiple data sets. All of them are temporal, meaning timestamped data sets from so many different sources to synthesize so that we have a picture of what happened exactly on that day. The implication of that is it's all about public safety.
We need to be able to simulate. It's not just about anti-terrorism or, you know, prevention of crime. It's even about, you know, tsunami warnings, earthquake simulation. What's actually gonna happen if there is an earthquake? How do we make sure that our people, the public, are safe? This image classification, this facial recognition, these algorithms are tremendously important and will be used by lots and lots of, you know, defense agencies, police, security forces, municipalities, cities, you know, retailers and so on and so forth. It isn't just about image. You know, most of today and a lot of what you've seen in the, in the show floors is about image because it's easy to see images. It's as applicable to sound and language as it is to image.
A great example, you know, it's not just about, "Okay, Google," and you sort of speak to Google and, you know, isn't it amazing that now Google recognizes an Indian English accent? I mean, I just find it so cool, right? That I can actually speak in an Indian English accent and it understands what I'm saying. What's also interesting is music. If you have an application like Shazam. For example, or, you know, just a few bars of music. You can play into Shazam and it recognizes not only the artist and the tune, it can even tell you. For example, this is Mahler's Fifth Symphony as conducted by Michael Tilson Thomas instead of, you know, Herbert von Karajan, for example. It's that good at recognizing and classifying music, and music is a complicated problem.
You know, the top line is kind of all about, you know, consumers and regular people and public. You know, the bottom row is much more about the enterprise use of these. There are a number of companies already on the show floor here at GTC showing medical in-image interpretation, using deep learning, and this is everywhere. The amount, the quantity of medical images whether they're X-rays, ultrasounds, MRIs, CTs, you know, and so on, is quite amazing. I remember four years, nine years ago or something. I accidentally fractured my toes because I was running to pick up my daughter from a volleyball game, and I crashed into a door and I fractured two of my toes.
I'm sitting in Palo Alto Medical Foundation, this, you know, lady who's obviously the podiatrist, she knows how to look at an X-ray. She looks and says, "Oh, yes, your toe is fractured." After about five minutes, I said, "I think you mean two of my toes are fractured." Because even with her experience, she mistakenly did not see that two of my bones were broken. If this can happen with a simple fracture on X-ray, imagine the complexity of interpreting these images when you're looking for, you know, organ patterns and tumors and changes in cells and so on, and growths of cells. This is a complex problem with huge implications.
Lastly, I wanna highlight, you know, Netflix wrote a paper about how they're using deep learning for, and Tesla for recommendation engines. You know, the cool demo that you saw. A few weeks ago, I was in China with iQIYI, and this is the kind of Hulu and Netflix of China. They are part of Baidu, and they make their own programming and, you know, Chinese kind of soap operas and, unfortunately in China, people don't pay, you know. Most of their revenue. They have lots of subscribers, but they don't pay a lot of money. They have to recover a lot of their money from advertising.
They're looking at deep video for recommendation engines so that while, you know, the show is being streamed, they can have dynamic content being inserted either into the frame of the video based on who you are and what you're looking at and what your preferences are and so on. That's a kind of new way of thinking about, you know, deep learning for video, and I really hope and I'm very excited to see that they're working on this project, and I hope it fructifies. I'm pretty confident it will. Deep learning is everywhere. I don't know how big this market is. I can sense, About three weeks ago, we were in Washington D.C. with my team, and we went to all the security services. We showed them the deep learning demo.
We sat with them. They have, of course, you know, amazing images, right? You can imagine these amazing images. Very, very high resolution sort of images. Deep learning is obviously applicable to them. We know already that, you know, industries are adopting, regular You know, this is not just about cats and recognizing cats in YouTube videos. This is about doing real computing, and it's gonna be a huge market, and we're very well-positioned for it. Let me just sort of close off with where we are today. The first key point is big customers, big companies are adopting deep learning. This is a small sample, of large corporations, what I call industry leaders, that are using deep learning. You know, IBM is one of the best speech recognition companies and they're using it.
United Technologies, you know, they don't make driverless cars, self-driving cars, but they do make autonomous vehicles like drones and planes and, you know, equipment for our defense forces, and they're using it for occlusion detection, which will help the autonomous vehicle better recognize their surroundings because of deep learning. And of course, United Technologies also by the way, owns Otis Elevator, so maybe they're working on improving those driverless elevators that Elon Musk talked about in the keynote today. You saw, you know, the demo of Skype simultaneous translation. You know, Skype, which is owned by Microsoft, and how you can speak in English and hear it in Spanish, and somebody else is hearing the same thing in German. That's all done using deep learning algorithms.
Big companies are adopting deep learning. Equally importantly, lots and lots of smaller startup companies are driving innovation in deep learning. Here's the thing. Because it's more affordable, you know, Wired put out a headline saying, "Now you can make Google's, you know, $5 million artificial brain on the cheap." In other words, you know, for just a few thousand dollars, you can buy the Titan X and set up, you know, a deep learning system. All of these startups, it doesn't cost a lot of money to be really innovative. What's happening is the areas are, you know, image and video, voice and language and sound, big data. There's some awesome demonstrations of large data sets. There's a company, for example, GIS Federal, I was looking at their demonstration.
They've got the Federal Aviation database, the live database of all the plane journeys happening over the U.S. right now. You can now draw a polygon or any sort of, you know, circle, whatever, around an area of interest, and you can select all of the flight data information within that polygon. When it was running on a normal Oracle database, this took 92 hours. What you'll see in the demo is it takes literally seconds. You know, you can see it, you can see exactly where your flight is. You can get all the flight data that's currently available from the FAA right there for you. Large data sets and analytics around large data sets. There's lots of industry-specific things going on as well.
Many companies, as Jensen pointed out as well in his keynote, many companies are offering deep learning as a service. You can, you know, other companies, other startups can build deep learning applications using their platform as a service. What's most interesting, especially since you guys are investors and you represent that community is many of these companies are being acquired by the large corporations. Just the other day, you know, IBM bought a company. IBM Watson Group bought a company. DeepMind was bought by Google. Cooliris was bought by Yahoo. Madbits bought by Twitter. Wit.ai bought by Facebook, and so on. What happens is, you know, the startups do the innovation.
They use the our developer platform to develop, you know, cool deep learning applications using this fancy algorithms, and then they get bought by large corporations. The upshot of that is this is leading to rapid adoption by all of the cloud computing data centers. You know, it's difficult to know exactly how many data centers, you know, Google has or Apple has or Facebook has. You know, I did a study of trying to figure out exactly how much Google spent last year on data centers, and it turns out to be somewhere between $3 billion and $7 billion, right? I asked Quanta, and Quanta said they sold 1.1 million servers last year, right? Most of their customers are in data centers.
You get a sense of, you know, these cloud computing data centers are very, very large. They consume lots of energy. The fact that many of the applications they're now running, given the large data sets, gives us a tremendous opportunity for rapid adoption in these cloud computing centers. Every single one of the companies listed on this chart are all using a significant amount of Tesla processors already. Some of these companies are just sort of, you know, hosting companies or, you know, infrastructure as a service companies, and others are more application companies. I'm pretty confident that a lot of software as a service companies will also start, you know, using the Tesla accelerated computing platform for their cloud computing data centers in the future.
I think I'm exactly it says, 0.06. In summary, we created this new category of acceleration by focusing on developers, by helping them speed up their applications. Because of that, we went into the data center with a new concept of accelerated data centers. We focused on high-performance computing, where we have now already deep penetration in government and significant penetration into industry. We're seeing this new growth opportunity in deep learning in the cloud computing data centers. I hope that I've shared with you know, how remember, I showed you the value, you know, why people buy. The application runs faster with fewer servers, consumes less energy, a lower cost of acquisition, and because of the lower energy, a lower cost of ownership.
That's the reason that this adoption is increasing, and we have positioned for strong growth in the future. Thank you very much. To introduce my colleague, Rob Csongor, who runs our automotive business.
Thank you, Shanker. By the way, I don't think it's 0.06. I think it's -0.06. Six minutes over. Okay. Hi, guys. I wanna I'm Rob Csongor. I'm the GM of the automotive business. You guys have heard a lot about automotive. I know you guys, track it a lot, and we had a lot of conversations last night. I got a lot of questions and, you know, a lot of the questions focus on why is NVIDIA in this business? What are the things that are happening? I guess, what I'd like to do in the presentation is walk you through a number of things. First of all, why are we excited about this business? You know, what are the metrics that we look at to measure our progress in it?
What are the things that we look at to drive the business for us? What are the things that we look at? What are the problems that we look at that we think we can solve? Then, finally, I'll summarize with our product strategies our business strategies and what the future looks like. Maybe what I'll do is I'll start with the end in mind. I'll start from the perspective of why are we in this business. If you guys look at NVIDIA and you look at the history of NVIDIA, we always focus on markets that are fundamentally have a very complex problem to solve.
If we had intercepted the car market about 20 years ago when NVIDIA was started, I think we would have been trying to solve the problem of how can we accelerate the turn signal or the radio in the car, right? There just wasn't a lot of electronics or computing in a car. What's happened is that the car is very rapidly going from a very simple electronics device to what we refer to as a supercomputer on wheels. There's a number of things that are driving that. If you could just start, you know from this morning, you look at what everybody's saying, you look at all the news, without exception, every car maker in the world is working on a self-driving car, okay? Think about it. A car with a radio to a self-driving car, okay?
That's like going from a hot air balloon to landing on the moon, okay? In order to do it, the amount of problems that will need to be solved are going to be enormous. There'll be lots of people involved. There'll be lots of vendors. There'll be lots of suppliers. There'll be lots of partners. There'll be lots of applications. There'll be an enormous amount of things that have to be solved in order to make this happen, just like landing on the moon. At the core of it, you're gonna need an enormous amount of technology, and it's gonna be a major disruption. At the end of the day, that's where NVIDIA always focuses.
We focus on segments of the market where there are visual computing problems and where there's enormous problems that can be solved, and that's why we're fundamentally excited about it, okay? Now, on the path to the self-driving car, there's a lot of things to do. You don't just go from a hot air balloon to a self-driving car. There's lots of things that have to be developed along the way. There's lots of intermediate steps, right? What I wanna do is I wanna walk you through kind of what are the things that we're involved in, and then how do we see the business developing, how do we see it evolving, and then how do we measure ourselves, okay? First of all, you guys know that we have a very deep engagement in automotive.
We've already talked about some of it. There's one thing that I wanna point out here. You know, if you look at the image there on the left-hand side, that's the M6000, Quadro rendering, ray tracing. When we talk to automakers, the fact that we are involved in many other industries is a huge advantage. Think about it. If we were only solely focused on automotive by itself, right, there's no way that we would justify creating the investment that's required to build a self-driving car. Because NVIDIA is involved in so many different industries from medical research. All of the lessons we learn, how do you cost reduce a computer? How do you create scalable architecture? How do you create one software that can be leveraged from one business to the next? How do you do deep learning?
We're involved in all these different fields, what NVIDIA can do is leverage those experiences, bring them to automotive, where what we're doing is essentially building another supercomputer. This one just happens to be on wheels, okay? In this case, what you're seeing is an example of that. We go to a car company, the architecture that's used to design the car is the same architecture that's used in the car computer, which is the same architecture that's used in doing simulation and the same architecture that is used in the data center. The only difference is that in the data center, it's 20,000 cores, in a PC, it's 2,000 cores, and in a mobile chip, it's 200 cores, right?
The leverage of the same architecture and the software across all those different things means that a car maker can leverage an enormous amount of investment. The more these car computers grow in complexity, the more the cost of developing them is. NVIDIA's fundamental value proposition is always to reduce the overall cost of development. And as a car computer gets more complicated, that's a very important value proposition. Now, by the numbers right now, NVIDIA, we're at about 8 million cars on the road, about 25 million more cars coming. All right? I picked a few images here just to give you a flavor of what are some of the cars that are coming. The top image there is an image of the Audi TT. That's kind of a cool car that's coming.
I'm sorry, it's a cool car that's on the road right now. What this car does is it takes the infotainment nav, it integrates it into the cluster, and you have an integrated cockpit. Audi calls it the virtual cockpit. The image on the left there is interesting also. I think since last time we talked. You guys may know that we announced Honda as a customer. Honda is interesting for a number of reasons. I think some of you asked last night is NVIDIA only focused on high-end? You know, are you only gonna be in premium cars, right? Honda makes the Honda Civic. Of course, it's a car that a lot of people drive, okay? In that particular case, what's interesting also is that this is the world's first Android car.
The more complicated the car gets, the more car makers have to look at the overall cost, right? They have to look at not just what is the cost of one part and one component of a car, but as you know, 80% of the cost of developing in a car computer now is software. They know that, again, that NVIDIA has expertise in Android. Android has lots of things built in. You can download it. It's free. You can have over-the-air capability, and you can bring this into a car almost immediately. This is a time-to-market strategy, and in this case, software is the key thing that matters. These are some interesting dynamics, right? These are some interesting dynamics in the market that are driving us that again, that NVIDIA has some unique capabilities and as a result, we're able to drive our business.
I wanna share with you some of the growth metrics that we measure, right? I picked some of the ones that I think are meaningful, okay? I'll talk about some of the dynamics that are underlying these growth metrics. Of course, revenue, our business is growing very rapidly, and of course, due to car makers engaging us. You're seeing a proliferation in the models. Now, there's really two things that are occurring within the business. First of all, car makers know that if you develop a car computer, it's much better to take that car computer and leverage it. If you can build it once and leverage it into multiple models, then this is a good thing. What you're seeing in models increasing is really two things.
You're seeing the proliferation of a base model computer, which is going into multiple models, and you're seeing new design wins, okay? For example, one car computer developed originally on Audi, now that car computer is inside the Volkswagen Passat, or it's inside the Volkswagen Golf, okay? You're seeing both design wins as well as proliferation of models. Both of these things, we think, are important because, again, the cost of developing these car computers is going up. It's not enough to have a chip. A chip is just a start. If you look around this conference, this conference is not about hardware. It's about researchers and developers. It's about how do you take this chip and then build applications on top of it. The automotive industry is no different.
One of the metrics that we use to measure our progress is to look at how many people out there are developing. What kind of applications are they developing? Computer vision, graphics, capability, all that kind of thing. If you walk around GTC, you're gonna notice that automotive has a much, much bigger footprint in GTC than it ever did. What we're showing on the bottom left is SDK shipments, the software developer kit for in the automotive business, it's called Jetson. These are development kits we send out, and then we support it, just like we do in the gaming business, just like we do in the professional visualization business. Same thing.
It's that same strategy that NVIDIA's become excellent at, we're leveraging that now into the automotive world because, once again, it's a complicated computer, and it requires lots of applications. The last chart, ADAS deep learning engagements. You obviously heard a lot about deep learning this morning. ADAS, for those of you not familiar with the term, is Advanced Driver Assistance Systems. Another way to say that last chart is beyond graphics. If you look at our growth and our revenue, it's all infotainment today. It's all graphics. A big, big, big part of the future, and if you look at self-driving cars, is gonna be deep learning, and it's gonna be computer vision, and it's going to be assistance systems and computation.
When you look at the investment that we make into these GPUs, and you look at the investment, it's not just graphics. It's about computation and about parallel problems or parallel applications. This is, I think, one of the other things that's pretty significant to take note of. One other thing I would mention, as an indicator, you know, as an indicator of what's happening for us within the automotive space is just to walk around GTC, right? If you walk around GTC two years ago, I think, you know, we had maybe a few guys. Here at GTC, and I really encourage you to do this, we're gonna have 27 automotive presentations.
The head of zFAS at Audi is gonna be talking about self-driving cars, what they're doing with NVIDIA, and then the next early trends for deep learning. Elektrobit is gonna be talking about how to design next-generation human-machine interfaces. How do you extract the power that's within a processor, a GPU, and create information that's functional and is also visually delightful? How do you measure human-machine interface, right? These are all things that we're gonna be talking here. Honda is going to be presenting on how to create a virtual crash test. There's just a wide variety of different presentations. There are 15 automakers here at GTC.
Honda, Toyota, Ford, Daimler, BMW, Audi, Tesla, Hyundai, just all over the map from all over the world. eight automotive tier one suppliers. You guys know that in the automotive space. There's an entire ecosystem, massive companies. When NVIDIA delivers solutions to the market, we partner with tier one's. Tier one's have been working in the automotive world for years. These are people like Bosch, Conti, Denso, Delphi. They're all here, and they're presenting. Software companies, press and analysts, and if you wanna take a spin in a car or take a look at a car, there's I think 20 cars here that are showing various innovations in GPUs.
It's, I think just looking around you here at GTC will probably give you a sense of what's happening and a flavor, you know, beyond slides of really what's happening within NVIDIA's automotive business. At the end of the day, again, what's driving all of this is that the car is becoming a supercomputer on wheels. We've talked about that. I said it before, but what does it really mean? Essentially, there's several trends within the automotive market that are driving the need for more processing. First, the proliferation of displays. More and more pixels. We talked about this at CES, and what I wanna do here is share with you some empirical examples of what's really happening out there so that it's more concrete. Secondly, more cameras, more sensor devices.
All of the inputs from these sensor devices and cameras have to be processed. It's pretty complicated. The bar is rising continuously. Without exception, I mentioned every car maker is trying to solve the self-driving car problem. Without exception. All of these computers need a number of things. They need parallel processing. They need extensive software. It's software architecture. It's not just one piece of software. There's system software, middleware, application software, lots of things. Finally, it needs deep learning technology. The reason why is that you simply cannot iterate your way to a self-driving car. You need something that works in parallel with the existing solutions that you've implemented. We'll talk about that. Now, the question is: what's the TAM, right? A supercomputer on wheels.
Well, from our perspective, the answer is within the next 10 to 15 years, every car will have one. 20 years ago, when NVIDIA was started people said, "What's the TAM for 3D graphics?" The answer was, of course zero . You know, I say, "Well, you know, how many computers eventually will have it?" People would say, "Well, I don't know, you know, maybe 1%." The answer now, 20 years later, is 100. 100% of all computers have 3D graphics, right? It doesn't mean that NVIDIA is in all of them. In the areas where visual computing matters and 3D graphics is pushed by segments of the market that need it, that's where NVIDIA is gonna play. This fundamental trend is required in order to push in order to push us.
Okay, let me talk about some of the market trends. By the way, this is a real this is a postcard I picked up at a rest stop on the Autobahn, and I thought it was kind of funny. It basically refers to a crappy nav system only not in those words. What it really highlights is this, okay? Before you get to a self-driving car. This was perfect. By the way, I saw this article then I went into the message boards, and I was reading the comments, then I actually, I extracted this one message because I thought it was so perfect, right? "This is why everyone hates in-car navigation systems.
Seriously, why don't they just integrate CarPlay or Android Auto and stop charging $1,000 extra for, you know, a nav system for new cars? This is 2015, not 2005. Seriously, I mean, how many of you have a car nav in your car that you know, looks like it's from 2005? You know? Normally, nobody would raise their hand 'cause this is an investor con. You know, everyone's very serious. It's true. What consumers expect is what they see on their smartphone or what they see on their tablet. Because it takes 10 years to get something out to market, the end result is that consumers don't wanna buy a car nav. By the way, where do car makers make money? Accessories. Selling them, selling the nav, selling, you know, your premium sound system.
That's where they make their money. How are you gonna get it from here to there? Okay. Where you see NVIDIA engaging, right, again, is these are lessons learned from other industries. What NVIDIA is good at is taking those lessons, and then we take them to the automotive market. As a result, we can work with somebody like Tesla. We can work with somebody like Audi and accelerate that time to market. Okay? This is a fun, you know, if you're looking at all the different stats and all the areas where NVIDIA is involved in, this is the first one. Okay? What's the validation of, you know, what are some of the proof points about the need for more pixels and more displays and all that kind of stuff? Well, the car makers are not oblivious of this.
You know, we did a very nice rendering for CES that kinda showed conceptually a car that would have lots of displays. Well, the car makers just went ahead, and, you know, they're doing it themselves. If you look at the Daimler keynote at CES, they had a car. It just has pixels everywhere. It was just displays everywhere. Audi had the same thing. The Audi Prologue, I think there were like, seven displays with all sorts of information that can flow back and forth between them. It's not just concept cars. Right? This is the new Audi Q7. This has 4 NVIDIA Tegras. You've got the cluster that virtual cockpit. You have the infotainment head unit. In this case, you have two rear seat tablets or rear seat entertainment. They're detachable tablets. You can sit in the back seat.
You can load your music, take your music, load it onto the car's sound system, play with environmental controls, all of this kind of thing. This is not so. You know, this is not just concept cars. Now, if you're going to develop a car computer, you have to be able to drive the pixels, but it also has to be scalable, right? You can't have a car computer that just costs a bazillion dollars. We're not gonna be able to drive the business by doing that. What we did is we announced a product at CES called DRIVE CX. DRIVE CX is a digital cockpit computer, and we showcased it using a Tegra X1.
What the DRIVE CX, and this is our strategy, you know, the DRIVE CX, it is a complete car computer end-to-end, all of the components that you need. We use this. You know, we show this to customers. We show it to tier 1s. With this, they can deliver a scalable solution. Let me describe or let me tell you what that means. A scalable solution means that maybe you have an entry-level SKU of a car where you only have one display, but you wanna have the same design and use it for subsequent cars. You need to be able to add capability to it. The base level configuration of a DRIVE CX would have. For example, one display and then you can optionally add more displays, in this case, up to three.
You can power 16 million pixels with this at the high end. You can optionally add cameras to this also, I'll show you what you can do with that in a cockpit. The end result is that you have a lot more performance. As a result, you can also create visuals that consumers expect and that consumers would be delighted with. Again, the hardware is not enough. A DRIVE CX is not enough by itself. The DRIVE CX, the hardware the Tegra X1 at the core, one of the things that we are doing is to work with computer makers to not just deliver information to the driver, but to make it with the fidelity and the craftsmanship that they and the effort that they put into craftsmanship in the rest of the car.
You know, craftsmanship is a word that's used a lot at automakers. It's not enough to just have a car that's functional. They put a lot of effort into craftsmanship. The UI lags the car in craftsmanship. What we do are things like this. In addition to the chip and the hardware, what NVIDIA is doing is providing software services, tools, so that we can help the car maker extract the performance of our processor and then to deliver something that is functional and beautiful. Doing this is very hard. Car makers know all about cars. They know all about how to do instrumentation. They know about what information they need to present to the driver. We know all about graphics and 3D graphics.
This is if you went to the demo area and if you go there, you can see this in action, and you can see the tools and the software that we're providing. This is also, by the way, I think a reflection of the evolution of our business model. I think the evolution of the business model, and you know that in other businesses, we start, and a lot of people view us as a chip company. At the end of the day, after they've worked with us for a while, they think of us more as a software company. I think the automotive business is no different. The end result of this is that you can deliver something like this instead of something like that. Okay. The image on the left, I think we're all familiar with.
The image on the right is actual real navigation application that's running on a DRIVE CX and Tegra X1. There's no reason for a consumer. I think the end result of this is that you wanna develop something that the consumer's gonna love, not that they hate. Right. If you have something like this, you're not doing something. You're not developing 3D just for the sake of having 3D. In this case, visual cues of buildings, cities, intersections are important visual cues that give a driver much more immediate understanding of where they are. In this case, you use lighting and 3D graphics to reduce the clutter on the outside edges of a map and to focus the driver's attention on the center. Right. There's lots of subtleties and lots of things that you can do here. Again, you just have to have the capability.
In addition to graphics, He goes, "My bad." All right, the other thing that we demonstrated, and you can see it here at GTC is because of the Tegra X1 has so much capability. You guys know that when we launched Tegra X1, we doubled the graphics performance. In addition to that, we also dramatically increased the 16-bit floating point capability, the computational capability. The image signal processing capability of Tegra X1 is amazing, and that's all for doing video. What you can do is You know, how many of you have seen Top View or Surround View in a car? Or maybe you have it. You know, like, you're parking and it looks like there's a camera floating, right?
I don't know if you've seen it, but the quality of these top views is actually pretty lousy, right? The reason is what you're doing is you're stitching the video from four separate cameras: left, right, front, and the rear. You have to stitch them together really well. You know, there's, you know, the ground might be uneven, you get lots of distortion. What we are developing is a best-in-class around view. We use the GPU to do lots of processing of the overall image. It has a shadow. The wheels turn. It looks very, very visually realistic. In addition to the value of the image processing that we're doing, because we can process this in a Tegra X1, it means that you no longer have to have an external box.
Today, Surround View is an external computer in a car. By processing it in a Tegra X1, you now can save $200 to the automaker. That's an enormous cost saving, okay? This is an example of reducing overall system costs that we can do by providing better processing. By the way, there's one other interesting dynamic of this. You have distributed boxes today in a car, a lot of them can be hacked. If you go to China, you know, a lot of these external ECUs, you know, Chinese people just put an emulator on the CAN bus, and then they can just basically hack it and put something else in there.
I think there's advantages, and you're seeing more car makers move towards a centralized processing model, and Tegra X1 and DRIVE CX would be, of course, ideal for that. Beyond graphics and beyond the cockpit computer, at CES, we announced a second computer called NVIDIA DRIVE PX. NVIDIA DRIVE PX is basically two Tegra X1s. It has CUDA programmability. We demonstrated the ability for doing more advanced computer vision algorithms. Of course, as we talked about a lot today, it is the platform for deep learning for a self-driving car. The first thing that a DRIVE PX does is it allows you to connect up to 12 cameras. You might be thinking, 12 cameras is crazy. Who in their right mind needs 12 cameras in a car? It's actually not that crazy at all.
Today's car has two forward cameras, four surround cameras, two cross-traffic cameras. You have a camera internally that monitors the driver. You have, you can replace the side view mirrors, reduces the drag of the car by 15%, reduces the CO2 footprint of the car. All of these are things that are being worked on right now within the car industry. If you're driving fast on the Autobahn, you wanna be able to pick up objects as fast as possible. Now, today, car makers can use things like radar. They can use ultrasonics. You can pick up radar, but you can't tell what it is. Ideally, what you would be doing is using higher resolution cameras. The reason you can do that is because the cost of cameras are coming down at Moore's Law.
They're coming down in cost every day and much faster than radar. With a DRIVE PX, you can hook up up to 12. Two megapixel cameras, capture at over 50 frames per second, and still not overtax it. This is 1.3 gigapixels per second of processing, and that's what you would use it for. You would use it for being able to connect high-resolution cameras and then be able to process them and capture very quickly. CUDA programmability. What could you do with CUDA programmability? Again, because we leverage and because CUDA is taught at 800 universities around the world and because there's all these applications, there are, for example, if you go online and just do a Google search on rain removal, there are so many different rain removal algorithms that you can use.
You know, what we can do is process, clean up computer images, feed it to cameras, and allow detection, allow safer driving in the car. At CES, we demonstrated a surround vision solution where we can do an auto valet application, right? You can reconstruct a garage and have a car park itself in a garage that it's never been in before, only using cameras, right? Finally, the deep neural network computer vision, which we talked about a lot this morning, right? I think one of the things that's pretty important is if you look at the model and what is NVIDIA's strategy, our strategy with DRIVE PX is to deliver a solution that is a complement to work and investment and solutions that already exist.
That's really what the diagram represents. There are a lot of solutions out there right now, and I got asked about it a lot last night. You know, how do you complement, you know, other computer vision solutions or things that are out there? Think about it, if when we went into the PC market, when we went into the PC, we didn't go into the PC and redesign everything in the PC. There was an x86 CPU, for example, that was doing just fine. We didn't have to redesign an x86 CPU. People are invested in the x86 CPU, and it does its job just fine. What we do is we complement the x86 CPU. I think it's no different in the car. There are some computer vision solutions that are just fine at detecting objects, but it's not enough.
For those corner cases, for example, like as we showed this morning and Jensen talked about in the keynote, when you're doing free space calculation and a bus comes, or there's a car door that opens, or there's a pothole, or there's something, for those types of situations, you need some help. The complementary solution of a deep neural network working in combination with an existing computer vision solution, we think is the ideal platform beginning now. For all of the things that people are working on, the intermediate steps that you work on to get to a self-driving car, things like dynamic cruising, being able to change lanes, auto valet applications, all of these types of applications are steps, and then ultimately you get to the self-driving car.
At the same time, you want the deep neural net to be learning, like the baby hitting the ping pong ball. Right? It's a behavioral thing. You didn't teach the baby Newtonian physics. You just told the baby to hit the ball. All right? I believe that ultimately the task, and what we talked about this morning, the task of building that self-driving car requires a complementary approach to the conventional solution. Because the conventional solution, it's gonna be impossible to get there all right without something else. The thing that I think everybody was waiting for and which I think was addressed today was: when can I get it? When can I develop on it? When can I get started? The answer is, today, or in this case, May.
The end result of this, you know, for developers and researchers, and this is what I think everyone is excited about, is that you now have that platform. Right? You can begin working now. You can begin working today. The DIGITS Dev Box contains the hardware, and it contains the DIGITS software framework that Jensen talked about this morning. All right? If you think about it, and I know I got some questions on this last night, and let me just, you know, mention it briefly in the context of a car. Picture that the box on the left is in a data center of a car maker, and that's where you're doing the learning. That's where you're doing the training. What gets downloaded to the car goes into a DRIVE PX. All right? The deep neural net model gets downloaded to the car.
This is terabytes of data, and this is tens of megabytes of data. All right. When the car learns, and if it finds something that it's not familiar with, it uploads it back to the data center. It learns and then downloads a new deep neural net model to the car, and then not only does that car get smarter, but every car gets smarter. That's how this works. This platform, this development platform, is just a micro version of that future. The learning platform and then the runtime platform in the car. Okay. I'm almost out of time.
What I've done is I've tried to walk you through every step of what I think are the important developments, the important trends, the important strategies that matter to us, that matter to the automotive market, and that are ultimately driving our business. I think at the end of the day, I think to summarize our growth metrics and our growth drivers, we're currently a $200 million, roughly $200 million business growing 85%. 8 million cars on the road, 25 million more coming. We talked about the car being a supercomputer on wheels. I think hopefully, I gave you the context, you know, the depth beyond, you know, that statement so that you understand what's driving it. I think for us, the cockpit business, the infotainment business, the proliferation of displays is a growing business for us.
ADAS, assisted driving, is an expansion beyond that. I think ultimately, as I started, what's driving this ultimately is the moonshot goal that without exception, every car maker is trying to drive the self-driving problem. I think what this is going to do is going to define lots of intermediate solutions. Before you get to a self-driving car, you will have cars that will help you. They'll be semi-autonomous. They'll help you park. They'll help you cruise. Those will be things that will drive our business in the short term until you get to a self-driving car. Okay? Thank you very much.
Good afternoon. I am your last speaker for today, and I'm gonna finish up and talk about the financials. I thank you all for being with us today, seeing through the GTC keynotes, and then spending the day hearing from Jensen and all of the different business units on what we have cooking up for us in the future. I will take this opportunity to kind of recap some of the things that you saw in the individual pieces and how this kind of puts together from an overall financial perspective. I have four goals to deliver today. One focus on our transformation. Where are we? What type of progress have we seen? What do we measure to look at our transformation in terms of where we are? two our overall profitability. Drivers of our profitability.
What have we seen in this last year? And help you in terms of key metrics across there. Shareholder value, and our capital return program. Focusing you on how important our capital return program is, what we look for the program in the future, and how you can help think about how we'll deliver that. Then number four, how to model and think about our business as we move forward. Okay. Those are our four things I hope to accomplish. With that, let's see if we can get started. Okay, let's start with a recap of fiscal year 2015 at a really high level. First, overall revenue. Revenue finished at fiscal year 2015, $4.7 billion. 13% growth, more than $550 million increase over fiscal year 2014.
Most of that is driven from the individual platforms that you heard from today in terms of our driving our growth. We'll talk about that a little bit further in the next couple slides. Secondly, record gross margin levels as well. About a year ago, I stood on this stage as well and talked about again, a record gross margin. We beat that record gross margin again this year. We also had a long discussion in terms of what would be those drivers that you could potentially, possibly see any more gross margin. We did that and growing more than 70 basis points over prior year. Third, profitability. Pick your favorite metric, operating income, net income. I chose EPS. Overall EPS growth, more than 53% increase from the prior year.
Large of this is driven in terms of the revenue you saw on the top line, expanding our gross margins faster than our overall revenue. Thirdly, our continued effort to look in terms of our investments and make sure that we are leveraging every last dollar that hits us in the overall OpEx to drive our operating income. By any standards, looking at our results in fiscal year 2015, an exceptional year, and we're very, very pleased with the overall results. Before I go to the next slide, I thought this would be a great opportunity for us to talk about where we are in our transformation. What I mean from our transformation, you heard from Jensen really talking about why. Why move from a commodity type of business?
How do we focus on the individual platforms as we go forward, and why that's important? One, our platform vision allows us to get closer to our customers. The more tied in you are to the customers, the more that you can help them, the more that you can provide to them. Number two, it provides us that overall diversification of our overall business as we focus on the computing platforms of the future: data center, cloud, and mobility. You can see that by taking what we've learned in our PC overall business, leveraging that strength, and applying it to all those different businesses. The third piece is value. The more that you can provide in an overall platform, the more value that you can provide, the better the overall business model that we have.
You can see that right now in terms of the results that we're providing. I get asked quite a bit, and even three or five years ago, probably the most popular question that we had from this room and this group was: How are your attach rates doing? How do you think about the overall PC market in general? Is it gonna grow? How's your attach rate gonna do? You remember, those are the questions you all asked. Even about 1.5 years ago when I first started, that was the most popular question that we've had. What this chart displays, the gray bars represent the worldwide PC unit shipments growth over the last three years. That's below the zero mark. They're declining. Think you know and have seen that. It's a great market. It's huge. It's enormous. You know what?
It's not all necessary for the key platforms that we've been focused on. Our green bars. Our green bars represent one of our most important PC platforms, gaming, and our revenue growth over those same periods of time where the overall PC market declined. Over that period of three, four years, CAGR 17% growth in our overall gaming platforms for the PC. In this last year, 34% growth. An exceptional performance, even in a market where the overall PC units have not been growing. Secondly, where are we in transformation when we look at our business mix? In this chart, I took an opportunity to look back just two years ago and where we stood in fiscal year 2013 when you thought about our business mix. Our business mix, I'll call it 50/50.
50% of our business focused on OEMs, focused on GPU, Tegra to the OEMs, and our overall IP royalties. The other 50% are growth platforms making tremendous progress even in that time as we continue to invest in delivering these platforms. Let's put us where we are at the end of fiscal year 2015 in Q4. A very different story. It's actually 80/20 now. In less than two years, moved to 80/20. 80% of our business stemming from our growth platforms and the growth across there. What you see is key platforms, high-performance computing, Auto, doubling their size in terms of our overall company mix. You see strong growth. You've seen our gaming platform now become nearly 50% of our overall company. Tremendous progress. Okay. Thirdly, transformation, in looking at the overall revenue growth in totality.
Our white bars indicate where the total company growth rate was. Strong growth in this last year, 13%. Last year coming off of a decline of 4%. When we actually just extrapolate our four key platforms, gaming, enterprise, high-performance computing, and automobile, we actually grew 29% in this last year. Our growth in our growth platforms is double the rate of the overall company. You can even see that in the prior year. Although we didn't actually talk about some of that growth as we saw it. When people ask me, "What inning are you in in terms of your transformation?" It's a hard analogy for me. My sons play basketball, but I get the point of the question. I think we're quite far.
I think you're seeing really the results of a lot of the investment, a lot of the work for us to diversify into these overall platforms. Here is kind of a summary of revenue. Up on the top, focused on our key segments. These segments are still important. These are the form factors of the platforms that we deliver. We'll still continue to look at our GPU segment and also our Tegra segment. They're both extremely important for us. Down on the bottom, and what you had seen from individually the presentations today is how we did in those growth platforms. What you'll see here is we've received now a gaming business that's more than $2 billion in total.
Our high-performance computing, you heard us talk about it's in the $100 million . Here we have it at $279 to help you in a little bit more. We've talked about our automobile business growing to almost a $200 million business and growing quite exceptionally, over 80%. I hope this provides a little bit more clarity, a little bit more understanding in terms of our success, how we have taken these platforms and grown them into reasonable and great businesses for the future. Focus on gross margin expansion. How did we do it? What are those drivers? The last six years, even though everybody every time says, "Haven't you reached the max?" Six years in a row, expansion in terms of gross margin and hitting a record at this point.
Again, strength of course, in terms of the revenue growth, what are those drivers that reach us to this high expansion? Here, the center white is our overall company average. Sit right around 56. If you look to the right of the white bar four different growth platforms. Each one of them higher than the company average. That's what's driving our overall growth platforms. It's rare for a company to name a couple growth platforms in terms of what it's driving. We have four. Also to even have the ability that they're driving in expansion in terms of our gross margin. We're quite fortunate, but again, it's based on really that vision of true end-to-end platforms and vertical integration. At the end, in terms of our gaming, it can be around the company average.
You can see that just because it is about 50% of our overall business. Again, there's a lot of different pieces of that, and we provide value at all different price points and levels, driving our overall gross margin. Our enterprise businesses are even faster, and even higher gross margins than our gaming. Then, as you can see, our IP is some of the strongest. Yes, to the other side of the white bar are gross margins that are not at the company average. That's okay. Volume, a lot of it can add in terms of overall profitability in total. Those are still extremely important strategic focus for us as well. Operating margin expansion. Talked about EPS in the first couple slides.
When you look at our overall operating income, performance as well, hit about $954 million for the fiscal year, up over 40% from the prior year and the highest level over the last four years. We expanded our overall margin on operating income by 400 basis points. Where was that focus? That focus was really on our investments. That wasn't a statement that we did not invest. Of course, we invested. Of course, we continued to make the right moves that we needed to capture those growth markets that you had seen. Expansion in terms of our sales capacity, expansion in terms of our go-to-markets for such important parts of those markets, but continuously looking for efficiencies across the group.
Engineering, a great opportunity using the unified architecture, across the Maxwell line that is consistent to continue to find ways, to find savings in terms of what we deliver. We executed that and really drove our overall operating income. I get questioned a lot. Are we done? Is that the end? No. We'll continue to make the investments that we need to, going forward. Again, we're just gonna be focused on the overall profitability of the company. We'll talk about that a little bit later. Generating cash flow important metric for us. As you know, this fuels, our overall capital return program.
Quite a good year in terms of our overall free cash flow levels at the end of fiscal year 2015 at $783, driving more than 35% growth from the prior year. A lot of focus, again, in terms of the investments that we made, our overall capital expenditures, finding the exact timing of what we needed to purchase, the exact type, and really trying to manage that efficiently across the group. Over the last four years, our overall free cash flows have averaged about $700 million a year, which again, its main use is thinking about our overall capital return program. Our cash balance. Get a lot of questions in terms of how is the cash levels, what is your focus in terms of that.
Two years ago, our overall cash balance in total sat at about $3.7 billion, and we had about 50% of that located in the U.S. We executed a convertible debt shortly after that. Why? To overall leverage better use of our international cash so we could fund our capital return program. We're now where we sit. Our overall cash balance is still at about $4.6 billion. Our net cash is at about $3.2 billion, and we have about $1.7 billion in U.S. cash. Our uses for that is definitely our capital return program, as well as our operations that are, you know, headquartered in the U.S. We'll talk a little bit further. Okay. Our capital return philosophy and what our goal. Why do we have it? Why did we put it in place?
Again, it's a key component of our overall shareholder value delivery, and we're gonna make sure it's a long-standing program for the overall company. The first thing we need to do is to assure we've made the appropriate investments into the business, and we do that prudently as we thought through the investments that we need to think about for the future. Additionally, solidifying a competitive yield, a competitive yield against our peer set, competitive what you would like to see in terms of returns of a dividend. This is a long-term program our dividends, so we need to be very cautious and very thoughtful in terms of our level of dividend. Additionally, we use share repurchases to enhance our overall shareholder value, and capital return program is the last piece.
The number one goal in terms of how we look at this is trying to return the largest percentage of free cash flow that we can, associated with where our cash flows come in and our ability to leverage our U.S. cash flow for return. Let's see how we've actually done. We restarted our capital return program in 2013. That was the start of the dividend at that time, and we also repurchased stock to have about $150 million return of capital. In fiscal year 2014 and 2015, more than $1 billion returned in capital to shareholders over that period of time. The white line refers to our actual shares outstanding, more than a 12% decrease in shares outstanding over that period.
For fiscal year 2016, we've already talked about our intent to return $600 million for that year as well. Not including fiscal year 2016, but since the beginning of our program, since fiscal year 2005, we've returned $3.7 billion or approximately 70% of our free cash flow for our capital return program. It's here to stay. We'll continue to focus and provide as much information as we can in terms of what you can expect for there. In summary, when we think about our commitment to shareholder value, I think our number 1 focus is growth on the top line.
Growth in terms of our platform vision, our growth in terms of expanding to the adjacent markets outside of just the PC platform that you've seen in the past and moving to the cloud, the mobile, and the data center is important parts of our future. Our excellence in operations. Excellence in operations stems in a lot of different pieces. Everything from our operations on how we manufacture our products, how we think of the overall overhead to that, as well as how we think about our overall engineering processes, how we go to market, and the sales. You have our commitment to continue to focus on that, on making sure we're making the appropriate investments to grow these platforms as we go forward. Our overall operating margins grew over 400 basis points. We're very pleased with where they sit, right now at 20.
We had guided again for the first quarter along that same lines. Shareholder returns, again our historical average is a pretty good indication of what you could probably see forward. It might change quarter to quarter, it might change year to year but overall, we will try and the highest amount of free cash flow that we can to return to investors. Modeling our business. When you come up with your growth platforms and you come up with your overall strategy vision, they're generally gonna be multi-year in nature. When we think about our transformation to our visual computing platforms, you can look at that as a multi-year effort.
We've talked about a lot of the expansion of the TAMs through each of the business units' presentations and how important that is going to be for the growth as we go forward. The gaming market extremely healthy. Our continued expansion in high-performance computing, moving into deep learning and our enabling of the partners as we think about virtualization, and with the virtualized GPU. All of these are great drivers of our overall future growth, and we'll definitely see that continue. Two, profitability. It's a balance. As you move into an overall platform type of strategy, you have the opportunity to expand your overall business models into unique opportunities. Now you're leveraging the overall software skills of the engineers to deliver the additional value on the top line.
What you're gonna see is different business models all the way across our overall businesses, each of them delivering a different level of profitability. Profitability growth is definitely our focus. Shareholder returns. We've talked about it. They'll be here for a long term. We're committed to our intention to return $600 million this year and our dividend is definitely a long term. So we'll continue to focus on that. With that, I'm gonna bring back up our business unit leaders and Jensen, and we're gonna open up for Q&A for the presentations. Thank you. Thanks.
Come on, you guys.
Please, we'll have some microphones come across, so before you ask your question, excuse me, let's make sure we get a microphone so everyone can hear. Thank you.
Come on. I'm very proud of you guys. I wanna show you guys off. Come on.
Thank you. Another way to, maybe a lens to look at your financial results is that the company has consistently invested about 35% of revenue into platform development, software development, and the business in general. As you talked about having many of your growth platforms now with higher than average corporate growth margins, and at the point of launching, how do you think about the ability to invest? Do you think investing will increase as a percentage of revenues, may stay the same, or can it scale? Basically, it's another way to ask, you know, what is your potential in terms of operating margins over the next five years?
Would you like to answer?
Me?
Either way.
There's a couple ways to think about growth. First of all, the TAM of the markets that we're engaging are very large now. They're also new markets. The concept that high-performance computing is going to be very important in the cloud is, I think, a foregone conclusion at this point. I mean, we're just feeling it all around us. The fact that we could put GPU-accelerated graphics in the cloud is you're feeling it. It's us. It's all of our partners that are engaged in this platform. It's all of our, you know, all of our OEMs that are part of the GRID platform. You could just feel how vibrant that opportunity is. We're solving real problems. The market opportunity is big.
The first thing about growth is that you have to create something that engages you into a large TAM. However, that large TAM today, you have to think of it in terms of a SAM. Some of that TAM requires additional market cultivation to happen. For example, Shanker talked about the fact that you have to enter it. First of all, you have to convince somebody that your technology is possible to solve their problem. Of course, they will do trials and then go into a pilot, then they go get budget approved, and then before you know it, they deploy. That process in certain companies, enterprises, could be nine months. That process could be a year and a half in some large companies. We have to be thoughtful about how long that would take.
During that time, we could put all the money into it that we want. It won't make a difference because it just takes time for people to go through their own process. Each one of these verticals, whether it's accelerated cloud, whether it's high-performance computing, whether it's automotive, which is a multi-year design cycle, whether it's, you know, so on and so forth, you have to be thoughtful about how quickly it will grow irrespective of your investment. We have to find that measure. It basically goes like we've created a TAM. We can imagine a very large opportunity. We can feel it. We can feel the buzz of it. Then we have to be measured about the rate of investment because sometimes it just takes time, okay? If that's helpful to you. We just gotta find that balance.
Yeah. Can you help me understand?
If I could there was something I was gonna say. Just one second.
Okay.
In no time in the history of our company, in no time in the history of our company have we ever engaged a TAM, an opportunity that is many, many times bigger than the size of our company today. You guys do the analysis of your own TAM. Just kinda rough ball it, and you'll come to the conclusion, whether it's, whether it's our estimate or the estimates of our partners or the estimates of the people that are in this conference, it is a multi-billion-dollar TAM now. It's many times the size of our company, and that's what's really exciting to all of us. I think the growth opportunity is surely there.
We just have to be measured about how quickly to invest and add fuel to it so that we can, as Colette said earlier, continue to deliver increase in profitability. Okay. Go ahead. Yes, sir.
Yeah. I have a question about, maybe particularly on the auto side of how are you different than Mobileye? I mean there's a dramatic difference in valuation. When I listen to your presentation, it doesn't sound that different. What's the difference?
We do actually very different things. We do very different things. They're a computer vision chip that's connected to a camera, and it detects things. It detects things. For example, when I walked into this room, this is an extremely unfair explanation, but I'm gonna do it anyways because it's easy to understand. When you first walked into this room, this room is a smart room. If the lights were off and you're the first person to walk into this room, it detected us, and there was a motion sensor, and dunk, the lights turn on. In the case of today's ADAS, there are several ways to basically perform that same functionality. You could do it using sonar for things that are very close to you. You could do things, you know, you could use a radar.
Radars. In fact, most of the cars that have ADAS today are radar-based. You can also increasingly add cameras that allows you to also perform some of those functionality. It detects a car in front of you, apply the brakes. It detects that you're close to the lanes of the roads. It, you know, you could translate that to a subtle buzz in the steering wheel. That's what ADAS is. That's a radical difference to a self-driving car. A self-driving car needs to understand what you would do in a very large number of very complicated conditions. If you had to describe, for example, what is the algorithm for detecting a car in front of you? The answer is actually pretty clear. It's a 50-year-old problem. Computer vision has been around for a very long time.
If you had to describe how a self-driving car works, what is the unifying theory of self-driving cars? In fact, there isn't one. There's no unifying theory for playing tennis. There's no unifying theory for golfing. There's no unifying theory for driving. It's a behavior that is learned. It's a learned behavior. That's one of the reasons why we believe the answer is to use a processor like what NVIDIA builds that has the ability to run a deep neural network, as I was describing today, that can learn the behavior.
Just as scientists have taught these deep neural nets how to learn the behavior of inferencing, understanding, recognizing an image, the behavior of recognizing a photograph and figuring out what story to tell, we can use that same basic technology to infer what is the right, what is the best driving action to take, okay? It's just a radically different thing. They're a computer vision detector. We are a platform for software. Now, our platform for software has several applications. The application that is driving our growth at the moment is digital clusters and infotainment. Beautiful graphics. What Rob was saying earlier, more and more cars are gonna have richer and richer graphics because there's more and more pixels. There will be no cars in the future that should have dials and knobs and that kind of craziness, okay?
I don't even know how to drive a car with dials and knobs and, you know. It's just crazy. You just wanna have a nice computer, nice and elegant, and why wouldn't you want it to be more beautiful than your Android device three years from now? It's gotta be pretty rich. Graphics, computer, being the basic computer, that's the anchor point, if you will. That's the starting point of what Rob does. On top of that, there is a self-driving autonomous vehicle platform that is an augment to that. His basic business today is infotainment, digital clusters, a supercomputer, if you will, for cars, and then it's augmented with self-driving technology, okay? We're, in fact, a very different. Now, of course, some people find joy in battles. You know?
It's like, it's just more fun to watch people battle. We're not battling with them. We're every platform that we're in for self-driving, they're in too. Every one they're in, we're in. I mean, there's no sense to battle. We gotta just work together and build these self-driving cars together. Okay.
Can you talk a little bit about the PC gaming business in the context of Intel's pre-announcement, cited a bunch of negative factors for PCs? You guys have done a great job of explaining how you've decoupled from that, and your growth is separate from that. Are you completely decoupled from it or, and any of the currency factors or things like that Intel cited, things that we need to be thinking about?
Let's see. I think they cited largely enterprise PC declines. We are in 0% of enterprise PCs. We've been out of that market now for, I think like 18 years. The reason why is because most of you just would rather have a thinner laptop that, you know, and run Office. Largely, I think that that's been solved. We're largely decoupled from that. I appreciate that, Bill. Do you guys want to add something else?
Jensen, you had mentioned that you wanna be to gaming what Netflix is to movies. Is that something NVIDIA wants to run or do you wanna hand it off to another company to actually use that?
That's really a good question. In fact, that's a fantastic question. I thought about that for a very long time. It came down to this I think we ought to just do it ourselves. The reason for that is because there's nobody else with the street cred to do it. First of all, you have to build the basic foundational technology. If we didn't do it as the world's largest visual computing company and the company with so much might in this area, if we didn't build it would have never been done. One, we had to go build it. Well, the next question goes, how do we get those platforms. Those processors and the software stack into data centers?
Well, it turns out data centers all over the world now in Amazon have these NVIDIA processors in them. The only reason why Amazon has it in them isn't because I told them to put it in there. They bought a bunch of GPUs themselves because it's used for cloud graphics for SaaS. It's used for high-performance computing, all these researchers around the world who are doing research. It's being used for machine deep learning companies who are trying to start these platform companies for deep learning, and they wanna host it from Amazon. There's so many different applications that have aggregated demand to these cloud service providers. And as a result, those GPUs are there. Then, of course, we have the additional demand of GRID and cloud gaming.
Now, if you were to rely on somebody else to do this, who would do it? Who would do it, number one? Who has the street cred to do it? Who has the capacity to do it in terms of having these supercomputing resources all over AWS? Who has the ability to go work with game developers to adapt, if you will, all of those games so that they're perfect for streaming from the cloud? Just on and on and on and on. By the time that we're done, really, we just ought to do it ourselves. Of course, the last part of it is, it's such a fantastic opportunity. We've invented all the pieces. We have all the skills.
You know, the folks in Tony's group and Fisher's group. We know the game developers so well. It's just such a fantastic opportunity. Why let somebody else do it? That's kinda how I reasoned through it. We're not a chip company whose destiny is just to sell to OEMs. If I start from first principles and ask myself, you know, what is the best way, most efficient way, what is the fastest way, what's the best way to deliver this value to a customer? I think the answer is I ought to just do it myself. That's kind of where we came down. Here's the other question, the flip side question.
Now that you're doing it yourself, and that you've decided to do it yourself, would you be open to providing the GRID platform for other people to build it? The answer is just about everybody who's even thinking about cloud gaming is using NVIDIA's platform. Okay. The answer is yes. If somebody else can do it better than I can, fantastic. The most important thing is we would just love for this Netflix of games to happen. If it doesn't happen, we'll just do it ourselves.
Okay.
Jensen, your OEM business, if I take out IP revenue, I think OEM revenue declined 19%. My question is, at what rate do you see it declining? Do you see it declining faster at the same rate? Is all of this revenue going away or some part of it is something you could stabilize and keep it, or all of it is going away?
We're winning OEMs. We've won big OEMs. We've lost some OEMs. The number of PCs, mainstream PCs with discrete GPUs is declining. It's a combination of all that kind of stuff. There's no guarantee that it's gonna decline. There's no guarantee that it's gonna go up. The only guarantee that I tell you is it will become increasingly less important to our business. Okay, I think that I've not told the OEM team, you're allowed not to win anymore. I mean, I don't know. Maybe it's just my nature. It just depends on which one of our managers you put in front of me.
The work that I'm talking to that manager about is the most important thing in the world at that moment, and I'll continue to do that. We ought to go win everything we can win. We ought to go win everything we can win. There's no harm in the OEM business. It's just that we can't rely on it. We just can't rely on it. We ought to go build valuable businesses that deliver a great deal more value that has much, much larger TAMs. As you can see, that's why the growth and that's why the increase in gross margins. We ought to go focus on that. In the meantime, my OEM team ought to fight as hard as they can.
Okay.
Hey, Jensen, just a quick question on the game console. You know, one of the reasons why Netflix did so well is that they're able to deliver through a browser. You know, we saw OnLive try to build a microconsole as well, several years ago. Why did you need to build a microconsole? Why not You know, if you look at Google Chrome, right, it can address the GPU directly. Why not, you know, why build a microconsole when you know, like Netflix, if you use it as a compare, they do it directly through browser. Why can't you do that?
Yeah, good question. First of all, we're not building a microconsole. In fact, we're not building a game console at all. You know, first of all, SHIELD is a Android smart TV first. It's not about trying to build a game console. It's about trying to reinvent reimagine the way you enjoy television. We have an opportunity to add a lot of value here. The reimagining of how people enjoy television includes, of course, a smart TV device that's really, really snappy. It's gonna have a lot of apps. It cannot include games, of course. Our objective, our focus is to reimagine, to reinvent, to revolutionize the way you enjoy television. We think we can add an enormous amount of value there because ultimately, video games, snappy performance because of apps and video games are gonna be important.
That's our starting point. What was wrong with OnLive? Oh my God, the list goes on. Comparing what we're doing with SHIELD and OnLive is just impossible. Of course, they were pioneers. I mean, Steve is a pioneer, and he tried a lot of things at the time. He only had a few, you know, things that he could work with. He couldn't virtualize the GPU like I could. He didn't have the benefit of Android TV. There were so many things that Perlman didn't have access to, but he did really great work.
I mean, Steve Perlman is fantastically clever, and it was just arguably, he just doesn't have the resources in his, you know, within his grasp to do the type of things that we're doing, and arguably, maybe a little bit too early.
Okay
I think in the last slide that Colette had up, she had used the term redeployment. I'm just curious if that's related to unified architecture, and if so, how do we think about the potential for operating margins over time as these growth initiatives take foot?
Well, there's two things that you noticed last year.On the level of investment we brought to our growth businesses skyrocketed. Enormous new investments in growth businesses, yet our OpEx stayed flat. The way that happened was the entire management team worked incredibly hard to unify the architecture of two previously separate teams. They were separate for good reasons. Now they're unified for good reasons. We took the GPU team and the Tegra team. We unified it into one. That process started about three years ago. We accelerated tremendously last two years. As a result, instead of doing two GPUs, instead of implementing two flows, instead of implementing 2 chips, we're now implementing 1, if you will, uber GPU. The same GPU that goes into Tegra for TX1 is the same GPU I announced today, Titan X.
The difference is 256 cores versus 3,076 cores, okay? The two of them are both incredibly energy efficient. That was the incredible benefit of unifying the two teams. As a result, we have one architecture, one design, one implementation flow, and we were able to redeploy, and that, I think was the word you redeploy many of the engineers that otherwise we would have had to hire new into incredible growth drivers that you guys are experiencing now. For example, that would be an uber example. I mean a big example. I mean, they ruined the word. It was going so well for so many years. That would be the ultimate example, the big example of what I think she was referring to.
There are other places where we're doing this. The management team is constantly thinking about how we could reshift and reshape, and that's one of the advantages in our company, and all of us have worked together for so long, and we're so comfortable with the idea of shifting resources from one group to another group to another group. You know, we don't think anybody belongs to us. They belong to the projects. They don't belong to organizations. They belong to the strategy. We change the strategy, people move around. Was that helpful to you?
Okay. It's over here, on your left. I wanted to ask a quick question about GRID in the enterprise. The vSphere announcement is to me a big deal. Things tend to move slower in enterprise, but it occurs to me that this might be an opportunity where there's significant pent-up demand. Maybe you could talk about the And you laid out a $5 billion TAM there, which is a big number. Maybe you could talk about the potential adoption curve and what that growth rate can look like over the next 24 to 36 months. Thanks.
Maybe Shanker knows the answer better than I do, but here, let me just go out on a limb and show you how little I know. Let me tell you why. It's actually truthful. The fact of the matter is we've created something new. When you invent something new, you have no idea how fast it's gonna grow. We have some early indicators that shows us that, in fact, the number of people who are trying it, who are in pilot, who have deployed, is now growing, you know, about 100% a year. We know that we used to engage about 10% of the market. We're engaging about 80% of the market, 90% of the market, okay?
Because of the support of VMware and vSphere. We know some things. We know that, in fact, the number of servers that are GRID-certified at OEMs, and it takes millions of dollars to engineer a new server. The number of servers that have been created to support GRID has grown tremendously over the last year. We know that even ODMs who are not branded, who don't call directly on enterprises, are now announcing GRID servers, for example, Quanta. Which means that the OEMs that they serve must be asking them for those kind of architectures. It's going down, you know, all the way through the supply chain. We know that this must be something important. Of course, we knew that it was important before anybody else showed us it was important. That's why we built it. It's simple logic.
It goes like this. We have been GPU-accelerated. Not one person in this room is having an office experience that's not GPU-accelerated since 1997. Since 1997 Intel has been in the integrated graphics business. It is GPU-accelerated. How is it possible that we are enjoying VDI with no graphics acceleration? The answer is because the technology doesn't exist to virtualize graphics. It's a very hard problem. Finally, we have done that. You're right, there is pent-up demand. The problem is severe. People want to be mobile. Enterprises are heterogeneous. You can't control it end-to-end anymore. Your new employees bring all kinds of crazy computers to work, and you've got to make them all, as an IT manager, make them all productive. There are many reasons why there's huge pent-up demand.
We also know that there are many industries where the workforce is incredibly in flux. How the laws treat your contractors, for example, in Germany, has a big deal, as it turns out in how they see this technology. For example, any contractor that sits on approximately the same campus has to be paid approximately the same levels. Of course, there's so much contracting work in some industries. The answer, therefore, is don't let your contracting firms on your campus. They want to virtualize that. There's so many different reasons. The film industry has a workforce that go from 0 to 500 people at the hot times of creating a film, and then it ramps back down to 0. How do you deal with workstations for them?
There are so many different reasons why, industry after industry that virtualization, remote-remoting, the pent-up demand is quite large. How big do we think it is? I think the relatively simple math we did is the question: Suppose we were to virtualize the graphics experience of 250 million people. Let's just kind of roughly call it 200 million people. If we're to virtualize 200 million people's enterprise computing capability, what would the opportunity be worth? That's kind of how they did it.
Okay. We could illustrate how bad we were. I mean, when we launched early access for VMware vSphere in August, we expected 150?
100.
100. We ended up at 500 within two months. you know, if you want to trust our forecast.
Yeah. Because it's so new, it's really hard to tell. It's really hard to tell. I think that the pent-up demand is surely there. I think, I think our solution is clearly a good one. Otherwise, why would all of these companies from VMware to Citrix to IBM to Cisco be out promoting it? It's legitimately fantastic technology.
Hi, this is Shankar from Bank of America. I have a question on the upgrade cycle for PC games. I know you guys mentioned that about 100 million gamers who have systems which have less performance than 960. And you guys work closely with the, you know, game developers like EA and Activision to kinda market once they release new games. How often, you know, or how quickly do the ecosystem gonna upgrade to the next, you know, next greatest GPU once those games come up? Is it every 2 years or every year? How do we think about that in terms of the upgrade cycle as for the PC gaming?
Well, I guess simply we view the installed base of having an average age of about three and a half years, somewhere between three and four years. When there's a big disconnect between new games and what's in the installed base, that'll shrink. There are periods of time that it may extend, but in the gaming segment, I would estimate it at about three to four years.
First of all, thanks for the presentations. Very useful. On GRID, on the partnership with VMware for their customers that take advantage of vGPU, GPU, and your GRID-based servers. Obviously, you guys benefit from the hardware sell. The other component to that is, you know, these same customers renew their vSphere licenses and their hypervisor licenses on a fairly regular basis. The question is, does NVIDIA benefit as well? Do you not only get the hardware economics, but do you also benefit from the ongoing licensing of the overall seats that accompany that as well?
Before I answer that, let me just make an observation. We, because graphics is so complicated, and you guys know, for example, we update our platform software literally every month. There's continuous performance tuning, there's new applications that come out. There are new features that are added. There are bugs to be fixed, so on and so forth. In this new environment, you take all of that complexity, you have to multiply it by one more thing. That one thing is the complexity of an enterprise network. The complexity of an enterprise network. We are going to be intensely software-rich in this area. It stands to reason that there will be a software layer and a software part of the licensing.
What I've advised my GRID team to do is to think about the GPU, the hardware as not their business, and for them to think about their business model in the context of that and not to use the GPU as a hedge, if you will. Hedge is not the right answer, but as a bad habit. You know, the GPU was already sold to the OEMs and the GRID software platform runs on this hardware platform. There's an ongoing quite a bit of care, improvement, you know, continuous tuning that will go on. There will be a software component to the business model.
Okay, great. Just one question on the automotive business, obviously, great growth there. You know, there's two parts to it. There's the penetration, which I think, you know, Rob gave us some good ideas about penetration on a go-forward basis. The other component to that is content uplift. If we rewind three years ago when we were talking about, you know, number of Tegras per car to sort of looking up the value chain as you guys layer on more software, and firmware and talk about compute-level solutions, how do we think about the dollar content step-up over time?
Do you want to take it?
I was listening to you. Here it comes. We're always making fun of each other, guys. One of the things that you'll notice. That's why we're able to work with each other for like, 24 years, okay?
Entertainment.
Entertainment and abuse.
It's not the word entertainment.
Here's the simple logic. Here's the simple logic. Today's cars are largely powered by embedded controllers. Calling them computers is like calling my rice cooker a computer. It does have a computer inside. We don't think of it as a computer. A computer has several characteristics about it that makes it much more valuable to us over time. For example, three days ago I got an OTA from Elon, and man, it gave me so much joy. There's no question there were more features. Apparently, in two more days, he's gonna give me another OTA that's gonna give me an amazing amount of joy. I'm not sure what it is, but I'm not gonna tell you anyways. That is what a computer is.
Your computer, your PC downloads applications. It's getting better all the time. That's why we use our computers for so long. My tablet, my SHIELD tablet is man, one of my favorite computers. I use the living daylights out of it. I'm using it for all kinds of applications that I didn't realize existed when I first created it, right, a few years ago. A computer has a living, breathing nature to it. Therefore, a computer is always increasing in computational capacity. For a very, very, very, very, very long time, the PC industry's ASP continued to grow until one day, of course, in the case of the PC, a lot of the computational went into the cloud, okay? That's the reason why it's starting to decline. The computation hasn't declined.
It's just shifted to the cloud, okay? If you were to measure Intel's ASPs, their ASP is clearly increasing as a company, but just not for PCs. I believe, now in the case of the car, it's just the beginning of that cycle. It's got to be 20 years before we're going to put all of that stuff in the cloud. In the meantime, that rice cooker is about to get more computational capability. I do know that the processors, the type of processors we have in Teslas need to be a lot faster. I know Elon wants it to be a lot faster because all of the software that he's piled on it ever since has caused the performance to not be as good, not be as snappy as when it first started. Okay?
When that happens, of course, the ASPs will increase. When the number of pixels of displays increase, of course his ASP should increase. I think over time, that will pan out. I believe that.
Jensen, a question on your IP revenue stream. Obviously, I think you're doing about $260 million a year currently. And given the situation with Intel and also, I guess you mentioned that there's no update on the litigation. I'm just wondering if you assume the best case scenario from the litigation, I don't know what that is, but just hoping that you could help us with that. Do you expect the current run rate to sustain in the best case scenario?
I think that we should answer that in one of two ways. First of all, the question is what do I believe? Then second, what should you assume? What I believe is this: I've always believed in first principles, and I believe on first principles. We have made greater contributions and invented more modern graphics technology than all of the other companies combined. I believe they're using that IP. We have no trouble with that idea. Now, we have no trouble with people buying our chips, which is my IP. It's just IP. The chip I didn't make, chip TSMC made. It's their IP. My IP, the NVIDIA IP is a bunch of software that's sitting on top of the chip. It's no different than the patents. It's no different than all of the innovation that isn't sitting on a chip.
We're a technology company, finally. I have no trouble people using our chips, so long as they compensate us fairly for it. We did work very hard for it, and our engineers worked really hard for it. It's upsetting for everybody if, including our shareholders, if we don't make an attempt to monetize it. I think on first principles, number one, we have an incredible treasure chest of intellectual property in this field, and a lot of people are using it, and we are serious about being fairly compensated for it. Okay? That's the first principles. Now, the question is, what should you assume? Well, I think, first of all, that's a hard question. I have no trouble with you guys assuming it's nothing. Just think about it from that perspective.
When everything shows up, it'll just be upside. Maybe it's just a lot easier as a way to think about it. Okay? Does that make sense?
Yeah, it does. Thank you.
If anybody questions the value of our IP. If anybody questions the value of the IP, meaning, "Gosh, Jensen, that IP can't be worth very much," I would like to do something on behalf of all of us. I would like to make a tender offer of $100 for all of that IP. It's worth nothing to everybody anyways, if you guys wouldn't mind, I'll just buy it for $100. How about that? How about $100 million? Not one of you would sell it to me. Not one of you would sell it to me. There's the test. Isn't that right? Free market test. Let's go to $1 billion. I'd buy it. That's the free market test. Obviously, it's worth a lot. Obviously, it's worth a lot.
We just have to go prove it. The way to go prove the value of IP is the way we're currently doing it. If you're serious about it if you believe in yourself go litigate it.
Great. Maybe for Colette, you said you have about $1.7 billion on shore. What's the minimum cash balance you need to run operations? Maybe, Jensen, given all the M&A in the, you know, semi industry. What's your philosophy on M&A? Thank you.
Our U.S. cash balance, yes, is at $1.7 billion. It's down from where it was in the prior year as we focused on a good portion of that being returned to shareholders. We receive cash flow both from our known royalty stream that you have. Other of our U.S. businesses also produce U.S. cash flow as well as also our stock exercises as well help us with our cash balance. We work to make sure we think about what we'll need for our overall operations, but making sure we're not carrying more than that, and we can return as best as we can to overall shareholders. You'll see that be a focus for us as we go forward. But that's where we stand today.
M&A. We work with startup companies, we work with other companies all over the world. We have no trouble working with them without owning them. Number one, we work with a lot of people, and you guys can see a lot of startups. We've made investments in a lot of very innovative companies. Sometimes the world is just better if you just left those companies alone. Sometimes the world's better if you made an investment in them and turbocharged them. Sometimes it's better for the world if we were to own them. We're thoughtful about all of that. We're thoughtful about that entire spectrum. Now, in terms of semiconductor acquisitions. We're very different than a normal semiconductor company. We're not exactly like Broadcom. For example, we don't have a large portfolio of things.
I think SGS-Thomson, excuse me, STMicro once told me they have 500,000 things on their catalog. I can't even fathom that. We have, like, four things on our catalog. Acquiring semiconductor companies, the first question is, semiconductor company, is it possible to integrate that semiconductor company into my substrate that we today call Pascal? It is incredibly hard to do that. The nature of our company's business is just very, very different. We're not like Broadcom, where you can have a large portfolio of components. We're not like Avago. We're not like STMicro. It doesn't work quite like. We're not like, we're not like TI. I mean, they have wonderful businesses, but we're not like that. We're not a catalog of products company. We are a platform company. We really only have one or two platforms.
That's also the reason why it's possible we can understand the markets that we serve so well. We're really only doing one or two things. It's just that we apply it deeply for four applications four large applications. Okay, is that helpful to you? Yes, ma'am.
You mentioned that the PC gaming industry market has been growing at about an 8% CAGR and that you've grown at about 17% over the same timeframe. Are you still trying to grow it twice as fast as the PC gaming market? If so, what are the main drivers that would enable you to grow twice as fast?
I'm not gonna let Fish answer that. The reason for that is because I insist that he grows faster than even that. See, if he were to answer that, he would have gone on record and said, "That sounds pretty good." I don't know about you guys, but that would take a whole week of correction, okay? On first principles, I actually don't know. Growth isn't something you can control. That's the score, if you will, of the game. The game you can play, and here's how the game is played. This is basic recipe.
When the production value of games go up, production value meaning the richness of the beauty, the richness of the graphics, the number of characters, the people that you can connect into it just the incredible lushness, if you will, of the world. When the production value goes up, the GPUs needed goes up, and the ASPs go up. Largely, that's the reason why we're growing faster than the overall game market. That's probably the most significant bit. The most significant bit is that the production value of games have gone up quite dramatically, even though it's just a game, another $60 game. Even though the game industry didn't grow, we grew faster than that because the production value of the game that was sold is so much higher. Does that make sense? Okay. That's number one.
The production value of games have gone up. Of course, there are other factors that have caused the GPU ASPs to have gone up. Resolutions have gone up. 4K. Oh my gosh, that's a lot more pixels. It's true that there's a replacement cycle, but I actually think the replacement cycle has now shortened. The reason for that is because our expectations have gone up. We see beautiful things among our friends. We know what they can do with their computers, and we don't know why our computers are so lousy. There's the number of games that benefits from higher, you know, higher, richer graphics and better GPUs has really gone up.
Even if the overall market is only growing at a certain level, I do believe that it's possible for us to continue to grow faster than that, but we'll see. Our job. How do I translate that to action? Instead of hoping, the strategy is this: We will go and invest in things like GameWorks that naturally, through our abilities and our understanding of science and math and computer graphics, by creating these modules called GameWorks that Tony was talking about earlier, and being the Industrial Light & Magic of the whole industry, we can lift the production value of the whole industry. If it wasn't because of Industrial Light & Magic, how would we have Star Trek? How would we have Star Wars? How would we have? They lifted the production value of CGI.
We are lifting the production value of the whole game industry, first of all, because it's great for games, second of all, because it's great for our ASPs, okay? Lastly, I can't wait till VR goes out. I'm excited about Crescent Bay. I'm really super excited about Crescent Bay, DK3, the new Oculus display. Hopefully, they go to market this year. When that goes to market this year, I think they sold something like 250,000 DK2s, the last dev kit. If they were to sell, you know, in production, call it just 1 million enthusiasts with VR glasses, that alone would turbocharge our gaming business in a way that hasn't been felt recently, okay? There's a lot of different ways that I think we can outgrow the market.
Now he knows what to do.
He knows what he's doing.
Thanks for a great day. Really appreciate it. You talked about GRID being at liftoff. With VMware, when do you think we're gonna see this inflection point where we'll really see GRID become a meaningful percentage of the overall revenue?
In terms of absolute numbers, it's in the, you know, $10-ish million. It grew in terms of percentage faster than any other business in the company last year. It was approximately zero the year before that, and now call it $10s of millions. I think it has the opportunity to grow quite fast, but it genuinely is hard to guess. It genuinely is hard to guess. If I could have guessed, I would have. Here's what we know we're tracking. We track three things. We track basically, how many are in trials, how many have gone to pilot, and how many that have been deployed. We track that.
We track the number of engagements that we are partnered directly with VMware, we call direct access. Where the enterprise has direct access to both of our teams, engineering teams, marketing teams, so on and so forth, so that we can accelerate. They wanna be direct access customers because they wanna accelerate the deployment. We've analyzed the situation, analyzed the account, and we think that they should be a perfect customer for rapid adoption. As Brown was saying earlier, we were expecting about 100, and we have 500. Anybody who could miss a forecast like that doesn't deserve to do the next forecast. Okay? I stopped asking these guys. Hey, what do you think it's gonna be like next year?
200.
Yeah, 200.
200. I think it's good.
You know, he's gonna double every year, and it makes no sense. I stopped asking him. On first principles, it actually makes sense. On first principles, he has no clue. He has no clue. The only thing that he knows is that the customers have real suffering. They do have real suffering. They have real needs. We have a real solution. There's real success. Everybody around us are equally excited.
It's not a fluke.
That's right. Okay. That's really all he knows. I think on first principles, that's actually, for me, enough. Now I know we have something of great value. We have partners that are aligned, and we're operationalizing their go-to-market, and we're excellent at that. On the other hand, I know that the market opportunity is very large. Very, very large. Now in the middle, I can't tell whether it's a straight line or an exponential or probably some kind of an S curve, but I just don't know where it's gonna take off. Okay.
Okay. Automotive, you guys did a phenomenal job, $183 million from $99. Thank you very much for the transparency on the model. Greatly appreciate that. You've talked in the past, Jensen, about how much backlog you have in automotive. It's the clearest line of sight that you have. When can we expect that inflection quarter in automotive?
We have to ship these 25 million cars, say in the next three to four years. Most of our design pipeline is about that long. Most of the things that Rob is winning now is being layered onto year, you know, three and four year out. People are trying to accelerate their product development. People are trying to accelerate because there's so much changing. It's still gonna take longer than two years to build, with the exception of some really fast engineering companies. It'll still take longer than two years to build, but it won't take longer than three or four years to build. Most of the design wins that he has now would layer on top of that.
For the next couple of three or four, I think he's going to have to ship 25 million cars, whereas since the beginning of, I think Rob's been working on it since the Jurassic Age. All of that, he's added up to how many cars so far? 5 million on the road, right? 5- 8 million on the road. He's got to ship three times that. He's cumulatively shipped eight in the last, call it six, seven years. Now he has to ship another 25 million more in the next three to four years. That kind of gives you a. Okay?
Yeah. Hi. Question about video. We haven't had an Instagram of video yet, meaning that normal people can make good quality video with whatever device they are on, you know, as they go and, you know, like kind of a user-generated video scene. Video seems. I mean, YouTube videos still look like YouTube videos, not very good. It seems like it's a harder problem than photos. Is that something that interests you, that you think NVIDIA could bring solutions to the market that could make that easier, accelerate that process?
There are many interesting video applications. Let me give you one. Suppose an internet service provider who provides a lot of video were to sign on a new advertiser, and the new advertiser says, "I'm willing to advertise on your platform. However, I wanna know that all of the previous content that has been streamed using our mark, okay, is made known to me so that I know that it's not being used improperly." Well, you would have to go back to the history of time to look for every single video, look through every single video to find a particular mark or a particular movie or a particular character or something. That particular application is very sensible because these companies don't want their proprietary content, okay, to be used in an improper way.
They wanna know that people are not taking their IP and mashing it up with their video and turned it into something else, okay. That's another way of infringing someone's IP. They wanna go through the entire library and do that if they were gonna become an advertiser. Well, the internet service provider would be more than happy to do that. Well, it turns out it's rather computationally intensive in searching through all of those videos to find those marks. That would be a perfect application. For example, of deep learning today. Now, on the streaming side, I could imagine that by the year's end, you will be enjoying on the SHIELD console 4K 60 hertz video from some very prominent service providers.
Looking at 4K 60 Hz is a lot different than 1080p 30Hz. It's just shockingly different. Okay, it's twice the frame rate and four times the pixels. Yeah, eight times the fidelity, and that will happen by the end of the year.
If I could follow up on that, with big data analytics and deep learning. One thing you mentioned was surveillance, or maybe kind of quickly mentioned surveillance. Can you talk about that potential there as a long-term business opportunity in terms of governments as a vertical, looking at not just images, but obviously signal and sound and, any other sorts of processing which we could use as a security application for example?
We're engaged with a lot of municipals on what is known as intelligent video analysis. We engage them on a platform that includes two parts. A little Tegra chip that takes the video in and is doing pre-processing of your deep neural network analysis of things. They don't know exactly what things they're looking for all the time. They download software running on a deep neural network, and it will find things. It's connected to a data center with a bunch of Tesla cards. The Tesla cards are now doing intelligent video analysis on a very large number of video feeds. Intelligent video analysis is part of an overall umbrella initiative in nearly every major country. It's called Smart City.
Smart City, Safe City is an initiative in nearly every single country. Somebody mentioned earlier about London. I think it was Shanker was talking about London earlier. The model of London is about to be replicated, except with high definition, except with more cameras, except with fewer people sitting there looking at all the cameras. It's going to be done on a large scale and done in an intelligent way using computational methods. How long would this take? It's hard to say, but you can just imagine the potential of the applications. It'll be in airports, it'll be in train stations, it'll be in bus stops, it'll be, you know, all over the place.
My sense is that it'll be done in a way that doesn't cross that line. I mean, they'll be smart about these, about the way to do it. One of our partners has won the contract for, you know, crowd behavior analytics at the Super Bowl, which will be held in Santa Clara next year
Perfect. Okay, let's take one more give me guys a chance. Colette, you have FinFET , transition coming in down the line. Is that a meaningful impact on your OpEx? Also, could you tell us your OpEx expectation for this year and next year?
It's our transitions are part of every day in terms of what we've been working on, multiple year discussion in terms of those transitions. I wouldn't consider it to be any type of a one-time event as we think about the future and what we may offer. When we think about our overall OpEx level for this year and what we think about next year, we'll continue the same focus of what we've done in the last year, assuring that we make the right investments in the expansion of the TAMs that my peers absolutely talked about, that are so important and vital for the overall future.
At the same time, the look in terms of efficiencies that we can find in how we work every single day will be top of on our list to do. Every single quarter and predicting what that exact number is gonna be one I'm probably not gonna be perfectly accurate, but I'm gonna work the hardest that we can, just as we've guided, to do that throughout the full fiscal year. As I look past that, I think that's just too far out to really think about. Let us get through fiscal year 2016 and go through. No, I'm not answering the question specifically with an exact number. Again, we're gonna continue to follow what we did in terms of last year. Again, focus really on these business models and growing their TAMs. Okay.
Thank you.
I think that's Colette's way of saying, "I guide once a quarter." With that, let me thank all of you for the great day. It's a privilege to spend the day with you guys. I hope that you guys take away several things. One, this is a company that has more capabilities in the field of visual computing than any company on the planet. We have reshaped our company so that we focus on four large vertical markets, and we offer it very valuable platforms and products. These four vertical markets has tremendously expanded our TAM, and that TAM is the reason why we're able to grow these growth initiatives, these four growth initiatives, despite the fact that the overall computing market is largely flat.
As a result of the high-value products, our gross margins are improving. Okay. I hope that those were the takeaways you had as well, and look forward to seeing you guys again soon. Thank you very much. Good job, guys.