So it's obvious that a big Tesla and CUDA target is oil and gas, or computational finance or other big industry that's already invested in HPC. But we still think there's a market that'll grow to use CUDA on the desktop with the end-user, doing consumer apps. We're just wondering how you'll manage supporting the big guys doing the big stuff, and guys writing apps for single end users. Will you support all CUDA ISVs equally?

That's a good question, and the answer is inside the business model at NVIDIA. The reason Jen-Hsun and I showed there's a GeForce and Quadro and Tesla, is that each business has its own value chain and boards and customers and channels and ISV relations. So the GeForce business unit has a set of ISVs that run applications on the GeForce, and they're using that computing product within the set. The business that I run, the Tesla business, is entirely focused on the ISVs and the application sets that run on those products. So how we divide it up is by business unit.

So there'll be some CUDA ISV overlap between all three business units, because you can all run CUDA apps on your products, but you'll primarily just take care of your business section?

Right. Let's take video as an example. So as a common technology, it's exploited in different ways, on different GPU brands. Quadro uses on the professional side and there's different attributes the GPU has to have there to work with it, and there's a different value chain to plug in to. They'll use the computing nature of the GPU in the right way to suit the Quadro business. And each business gets to decide how they support those ISV endeavours.

So say a market comes completely out of left field that uses CUDA that you haven't seen coming, would you put every effort in to supporting that regardless of it being an established market or business model?

Exactly, and we'd just sit in the customers seat there. If it turned out to be a great computing application but something that created output pixels for design, it would be mapped to the Quadro business, the Quadro people would go and care about it. And you could ask, "would the other business units help out there with a new market" and the answer is of course, that just naturally happens at NVIDIA where we leverage the expertise of a big company. We just sit in the customer's seat and see where it maps into our business.

We're basically just fishing to see if you'll support GPU computing equally across the entire business, not just with the big guys using Tesla. Are you going to look after everybody.

Absolutely, yes, and that decision is made by business unit and how they want to invest in their customer base and their ISV base to allow that to happen and opportunities to happen.

So let me answer part of the underlying question directly. The single-precision capability for example is available across all of the CUDA-supporting product lines, and will continue to be available everywhere, and that's the model that goes forward. So take the G84 and G86, we support CUDA there too. Now there's a separate roadmap that the computing products will follow, so double precision is a feature that clearly maps into HPC and GPU computing, but we can't see much use for it in the consumer space, so it's something that'll be available with Tesla and the high-end of the Quadro product line, but below that it'll only be single precision and DP will not be on the die.