Things could have gone out of control yesterday during a fireside chat with Nvidia's chief, - and moderated by analyst Jon Peddie - when it was interrupted by an activist handing out anti-Nvidia flyers about the chip recall issue. Fortunately, the chip maker CEO managed to keep his cool and even asked for one of the pink pamphlets!

"Could I get one of those [flyers]", said Nvidia CEO Jen-Hsun Huang to the activist. "Are you guys still a lot boycotting us?".

This incident aside, Huang covered a broad set of topics including his experience as an entrepreneur when he co-founded Nvidia 15 years ago or his role models as well as more technical questions like the future competition with Intel's Larabee, the integration of CPU and GPU and the recent licensing agreement with Transmeta.

Huang's advice to entrepreneurs: don't try to know everything or you might never start! And put your ego aside and put yourself in a perpetual learning mode.

"Not knowing everything was what saved us. If you know now what you what you will know someday, you will not do this. There's no question about it. Because it's too hard. There's too much to know. You have to know so many things to make everything come together. There are too many infinite number of details in order to be successful. Not to mention a lot of serendipity. So if you know now what it will take you to be successful you wouldn't do it because the odds are just impossible... My ignorance protected me from being overwhelmed by what you guys [entrepreneurs] are up to".

Nvidia lost the leadership in the high-end graphics market to ATI by misjudging its competitor' strategy/motives. Calls for a need to "simulate" the competition.

"We learn something of our recent experience [with ATI], in that to make sure we understand the competition... Nvidia's market-share in the high-end is probably something in the likes of 97%. So in order for us to grow we have to grow the TAM [total available market]... But if you're a competitor you don't need to grow the TAM, you just have to grow into that 97%. So here's a perfect example, where 2 companies [ATI and Nvidia] have built the same products but have radically different strategies. Our focus is growing the TAM, notice Nvision. There [ATI] strategy is taking my share, notice don't focus on the TAM, focus on building the fastest little widget at the lowest possible price... we have to make sure we simulate our competitors... Knowing our competitors is not enough... That analysis process is something we really need to internalise and do much much better... At the scale of business that we have today, all it takes is one chip to disrupt 100 million chips we have in the marketplace".

On integrating a CPU and graphics processor (GPU), Nvidia's boss stressed two very different types of integration: a physical integration and an architectural integration, with the latter - what AMD and Intel have been suggesting - making no sense.

"Physical integration which is taking object A and object B and adding them together. And there's architectural integration".

First, Physical integration of CPU and GPU: Yes!

Physical integration has been happening for some time. What's the difference to me wether Intel integrates the GPU or the graphics chip into their chipset or their CPU when the 2 are bundle together? The answer is no difference. They have been bundling and integrating forever.

Second, Architectural integration of CPU and GPU: Nonsense!

My belief is that architectural integration is actually likely to be working on the opposite direction and the reason for that is that we're starting to develop this understanding that there's a new class of processors that would be beneficial to mankind. The task parallel sequential processing instruction set architecture that makes it a CPU is a work of art, it's wonderful but it's not enough to solve a lot of the problems that we have today which are data parallel in nature. And if you have a data parallel type of problem there's no reason not to have a data parallel type of processor because we have an infinite number of transistors anyways. So why do all the work in the work with a screwdriver [the CPU] when you can have a screwdriver, a plier, a hammer... So there's no reason to have 2 architectural processors. It turns out that it's much easier to program... Heterogeneous [a CPU and a GPU] computing is the best way to go".

Is Nvidia building a CPU? Nope.

"I don't think we need to own our own CPU... I don't want to build what other people has built already. If Intel has built a wonderful CPU and AMD has built a wonderful CPU, why built it?".

What about Intel Larabee? It's a Powerpoint deck!

"First of all we don't know what Larabee is. It has to ship... I never competed against a Powerpoint deck for this long. So let's ask ourselves why that is the case... Imagine you're a start-up company and you have a great idea. This is going to be a killer, it's going to be the best thing in the world, it's going to change everything. And I'm going to tell my competitors right away. Is there something wrong with that? What would you do that? Well because the competitors are already there! We've already built Larabee, it's call CUDA... It's not x86 binary compatible but neither is Larabee. When you'll put Larabee on your PC do you think it'll boot Windows? I don't think so. Can you run Word? No... So what can you do with it? Write a C program and compile to it, just like CUDA... They [Intel] know that we've built Larabee 2 years ago. They know that we have 100 million Larabees in the world today. And they know that by the time they show up, there will be 200 million Larabees in the world. And if the case, they have a talk about Larabee and they have to talk about it as often as they can".

And Intel building a GPU? Great!

"I think it's fabulous because it's going to make the GPU market bigger... GPUs are not only for graphics... When you have another player comes in the market place and talks about the general purpose nature of this processor and start highlighting design wins that are very general purpose in nature... It's going to expand people's understanding and it's going to eliminate this barrier of 'Oh, GPUs are only for graphics'. It's going to open up the field incredibly".

Why did you license Transmeta's technologies? For power and dynamic translation.

"Transmeta innovated in 2 areas that are very important. One area is low power technology and building processors that are low power. And it's not that we can't reinvent it all. But we have to be respectful to the intellectual properties of other companies who have invented it. And so we're using a lot of very similar techniques as they have invented over the years as we built more and more processors that needs to be low power. The second is that they created a dynamic translation software that frankly most processors someday will use. CUDA is dynamic translation. We have an instruction set that is general purpose but that instruction set is a little different inside the chip and so hardware instruction set is changing all the time but the instruction set that is presented to the user never changes. That technology is something that we've using for several years. And again these are kind of grounds that Transmeta has walked through before".