NVIDIA- Past, Present & Future: An Interview with Jen Hsun Huang

Jen-Hsun Huang, Co-founder and CEO of NVIDIA is a determined person who is always ready to put up a fight for his company. Watching this energetic, enthusiastic and enigmatic CEO at the American University of Dubai was an extremely satisfying experience. He spoke passionately about GPU computing to a hall filled with young minds on how innovation rather than anything else is what drives NVIDIA- a company that went from three employees to over five thousand six hundred in sixteen years of its existence. I had the pleasure of sitting with him face to face. The following are excerpts from that interview.

“I started my career as an Engineer designing Micro Processors- building computers and workstations and designing chips. I enjoy solving problems with the same issues as other Engineers (at NVIDIA) wrap their head around and sometimes I can add value that gives me great joy.

When I was thirty years old in 1993, we saw the PC evolving- Windows 3.1 was out and the Pentium Processor was around the corner. We saw this and asked ourselves what if we could take advantage of the resources and add another processor to it and use it for gaming. Our logic was that every programmable computing platform will likely become a gaming platform. In fact, the Apple I and Apple II were enormously successful as game platforms but then people kinda went away from it. With Windows being so pervasive, we could add 3D graphics to it and bring in a level of gaming that is so compelling, so exciting and so different that people would love it.”

“When we started NVIDIA, there were three of us- Chris, Curtis and myself. Chris and I had families and Curtis was single so we’d always go to his house and that would be the NVIDIA HQ. Chris and Curtis were upstairs in their own rooms with their own workstations working on NV1 while I had the downstairs dinging room table and used to sit there and figure out our strategy, our vision. It was pretty quiet- mostly just thinking and reading and reaching out to Venture Capitalists trying to figure out how to get the company off the ground.”

I found that a bit ironic, considering he came from an Engineering background and launched NVIDIA from a business point of view. I asked him how that transformation took place.

“The first thing I did was that I bought a bunch of books- on marketing, sales, management, operations. I read every press journal. I bought a book by Gordon Bell who is very smart and methodical and in his book he shared ideas on writing good business plans for future entrepreneurs. That’s when I started writing NVIDIA’s Business Plan and after three months of rephrasing and making it sound good, I realized that if I kept working on this, the market opportunity will pass us. I knew what we believed so I just wrote the first principals and list of actions that are basically strategies on what we believe in and the resources we have. To this day, NVIDIA’s business planing is exactly like that- Short and crisp.

Things have not always been rosy and Jen-Hsun is the first to admit that not every single idea that comes out of NVIDIA is successful which is why “Tolerance to failure” is something very important at NVIDIA.

“NVIDIA was the first company in the world to have been founded with a singular purpose of creating consumer 3D graphics. About a year later 3DFX started and became an instant success because they made some really good decisions. Their architecture required low cost memory- they use a lot of memory whereas our architecture was very tolerant to high cost memory. When NVIDIA first started one Megabyte of DRAM was $50 and (3DFX) needed 4MB- that’s $200 in cost simply for memory. We could operate on just 1MB so the cost difference between our solutions was already $150 which was a big advantage for us. Then all of a sudden, memory prices collapsed and $50 became $5 and our $150 price advantage became $15. Their architecture was more elegant- it was easier to program and more compatible with OpenGL. They took off and we were flat on our back. We went from being almost successful to almost dead overnight. It became clear that we had the wrong architecture for the future where memory was going to be cheap.

We needed to figure out what to change. If we were just like them (3DFX) and already behind them then how were we gonna win in the market place. And thats when I really learned that there is a difference between strategy and technology. Two companies could use the exact same technology but have radically different strategies. For example, Yahoo and Google both provide search services at their core but the two companies obviously have very different strategies. In NVIDIA’s case these strategies are things like how do you build a product, how do you pace it, how do you work with developers, what are the things you want to be really good at and lesser-good at etc.

By the time we had our real first chip called Riva 128, they (3DFX) were already at Voodoo 3. We had to find our footing and buy enough time to come up with Riva TNT and then ultimately GeForce 256 which allowed us to really tear away. Sometimes being first is good, however being last isn’t bad either. Google was the last search engine (in the race) and being last has its strategic advantages. We tried many things and some of them worked and some of them didn’t but they all contributed towards building NVIDIA. Its our sequence of failures and break-throughs that allowed us to be what we are today. Tolerance for failure puts our employees in a position where they are willing to try new ideas.

For example, when we put GeForce 256 together, the big question was that if you put geometry on the chip and Direct X doesn’t do geometry processing- how are you going to offload the geometry? We thought that it was the right idea moving forward and if we don’t take this step, we wont be able to take the next step. So we talked to Microsoft and to OEMs- (and they said that) CPUs are fine for doing this. Of course Intel hated it. We said to ourselves that just about everybody hates the idea. We should do it. We knew we could fail but we knew this was the way forward. We reached out to all the developers around the world and they all worked with us on it and the day it came out, geometry processing was twenty times faster.

Another idea was was to put two graphics cards in one system and we could write the software to do load balancing. 3DFX had invented SLI and although our technology was completely different we still called it SLI because it was a great name. We asked our customers and Microsoft about their thoughts- Microsoft said that it was incompatible with the O/S which would make things unstable and they didn’t want us doing that. Intel said that it was a violation of PCI express. OEMs said that it would require too much power and they’d have to design new chassis and PSUs. We had all theses issues but we went ahead and did it.”

While we were on the subject of SLI, I asked Jen-Hsun why NVIDIA required identical GPUs to work under SLI while their competitor allows a more elegant “mix-and-match” solution that works across their many different models of GPUs.

“We can make it work too- its just that the scaling is nowhere near as good. Its impossible to have a infinite level of differences between the two chips. Do you render one frame here and one there, two here and one there or seven here and one there. At some point it doesn’t really deliver a very good experience. We’ve discovered that its very rare to actually deliver a better experience when the GPUs are off-ratio. It’ll speed up a tiny bit, but the latency you get in addition to that is not so great. We could make it work for the sake of competition but its not really a good experience and until we can figure out how to make it work great, we’re simply not going to support it. If something doesn’t work very well- don’t let people use it. If we’re not satisfied by it, why are we selling it?”

That prompted me to as Jen-Hsun about the NV30 which is considered as one of the worst releases by NVIDIA and many considered it as a half-baked product.

“We didn’t think it was half-baked. It wasn’t a successful product and let me tell you where it went sideways on us. GeForce FX uses an interface that was different than DX9. We chose CG as our programming interface. ATI had DX9 so the compiler generator from Microsoft could be used for ATI directly but had to be recompiled JIT for our GPU. That recompile process at that time with its JIT compilation was the reason the market hated it. And we did a really bad job at explaining it.

The reviewers felt that we were cheating and so they ran it two ways- the recompiled version and the native version. Our performance was terrible at the native version because the architecture was very different than what was implemented in DX9. That was a part of our history that we didn’t handle extremely well. There are many things we could have done differently before we announced it. I would’ve taken the architecture and taught it to the world on how GeForce FX works. That’s what we did with Fermi- taught everyone how it works because its so different and the performance is going to be so much better and faster.”

I asked him about the hostility with Intel and the cartoons.

“We just think its funny- its a nice way of letting it out. During the older times, peasants used humor against a tyrannical ruler ship. (Laughs.) It helps ease some of the frustration. But let me ask you- when was the last time you saw a company as big as Intel sue another smaller company? They’re scared and you can write this down- We will kick their ass when we go to court next year.”

Coming back to GPUs, I asked him if we have hit a brick wall with Desktop Gaming GPUs because displays havent scaled as fast as GPUs and even a mainstream GPU of today can provide a decent experience at high resolution.

“If we dont revolutionize the GPU again then we’ve already hit the wall. Thats exactly the reason why we invented PhysX and why we’re gonna bring ray-tracing to the market place. PhysX and Ray tracing even on the highest end, run really slow. You wont be able to do ray tracing across all your images but we could do things like soft shadows right away. Also, the last step of 3D graphics is a 3D to 2D projection- why render in 3D and show it on 2D- that’s weird right? If your monitor is 120Hz, you ought to be able to pop on a $100 pair of glasses and enjoy 3D but with that, your frame rate is just divided in half. So your current graphic card suddenly becomes half as good. With more Physx, Ray Tracing and some global illumination, things become more and more realistic.

I’m not cynical about the market but If you give them the same thing over and over again, they’ll get tired. Look at iPod- there’s iPod fatigue. If you were to have an iPod, well, yeah- who doesn’t? If you have a Zune HD now, everybody wants to look at it- people need a new thing. It is possible to build a new thing. Thats the reason why we didn’t wanna take the GT200 and add DX11. Its possible. Its as easy as peach.

Do you think the PC will steal some thunder back from the consoles with your new technologies?

Yes. Batman was recently released and extremely popular and everybody said that Batman on PC with Physx was the best. Reviews were fabulous and people who had played it on consoles, played it again on the PC. I also think 3D vision will cause people to play their games all over again.

Will GPUs become obsolete in the not-so-distant future when everything is rendered in the cloud?

“In twenty years, it will be Tegra on your handheld device and Tesla on a server. There will be millions and millions of Teslas all over the world rendering for everybody. Its gonna be used for rendering HDTVs, Cellphones and displays in your cars. The question is that is this the right thing to build? If it is and if I can deliver an extra-ordinary experience building it, then I don’t care if it cannibalizes our current products. If I put Tesla in the cloud, I know that every Intel integrated graphics user can have a good experience.”

Where are Tegra and ION headed?

“I believe they’re casual computers that people carry with them for the primary purpose to have a convenient access to the web. Some people have a PC sensibility so they get a netbook which is like a PC- Other people start from a cell phone sensibility- an iPhone sensibility. Its just that the iPhone is not big enough so they get themselves a bigger smartbook or smartpad. People that want the same thing but they come from two different places. Netbooks and Smartbooks are gonna exist because of people coming from these two different sensibilities. I have a smartphone but I wish I had a better Internet access. For the guy with a laptop- the laptop is too heavy and the battery life is not good enough. We would like to give you a great computing experience irrespective of your perspective. We’re a visual computing company.

With netbooks, our strategy is to bring a great experience to someone who has an Atom PC. An ION is an energized Atom so we’re gonna have all kinds of Energizers in the future- many many more Ions to come. Atom is a good but its a little limiting- not because the processor is slow but because the cache is not enough. Intel shrunk the cache too much. If you’re a little patient, dual core is gonna come and chips will get better and better. ”

Can you talk about Tegra 2?

“We’ll probably announce something at CES and while other people are still showing their generation catching up with Tegra, we’re gonna leap forward ourselves. NVIDIA’s rhythm is once a year and its been a year. We wanna keep the market moving very very fast so Tegra 2 is coming.”

Any Tegra based Smartphones coming up?

“No. Smartphones take a lot longer to get done because they have to go through a nine month period of career certification and we can’t announce our customers product. They have their own marketing department. ”

That was about as much time as I had with him and in conlusion I asked him where he sees technology going.

“Things will tend to change slower than you think in five years but faster in ten years. Reason for that is that it takes time for the industry to cultivate and organize. Once you organize, the rate of change is very fast. I believe that the personal computers as we know it, will become a “work” station. For example, our PC will become our work station- we’ll be working at that station. Most families will have a multi-media workstation.

But our “personal” computers for accessing the internet or casual computing will likely just be a web based device- 4G with 20mb/s internet connectivity will allow you to do everything with it. It could either be a large or small tablet- just a sheet of glass so thin and possibly translucent with a camera to augment the image in front of you with computer graphics. Basically Iron Man. That kind of future is not only a certainty, its practically here.”

I would very much like to thank Jen Hsun and wish NVIDIA the best of luck with their upcoming Fermi . Here is a picture that we took before the interview, one that I will certainly cherish.

About Abbas Jaffar Ali

Started t-break a good ten years back and the fun factor graph has risen tremendously at the expense of the inversely proportionate income level graph