nVidia CEO Sees Tegra as New Growth Engine - Page 2

Page 2 of 2

Those challenges include getting its graphics business moving. ATI, AMD's graphics unit, is well-ahead of nVidia in coming to market with a 40 nanometer part as well as having product ready for the DirectX 11 graphics library in Windows 7.

But by the second half of 2010, nVidia could have made the shift from a graphics processor to a powerful SoC design that becomes the bulk of its business, and Feeney thinks Rayfield is the one to do it. "He spent 16 years at Texas Instruments, and they are the leaders in integration on silicon," she said. "I think Mike is on to something but it will take time."

Hundreds of cores

Tegra didn't get all of the attention, though. Huang talked at length about transitioning his GPU business from just game acceleration to co-processing with a CPU.

"There are so many new apps that now require us to add something to it, whereas the architecture of the microprocessor is no longer adequate. The answer is a parallel architecture that doesn't just have two or four cores but hundreds, moving to thousands of cores, not fetching data through the cache but streaming through the processor," he said.

"If you make the bet that you don't want to replace the microprocessor, that the microprocessor is an extraordinary piece of hardware that should be there, but you want to add multiprocessing to it, then you will make a unique architecture good at one thing: parallel processing," Huang added.

nVidia projects the Quadro and Tesla market, which use the graphics processors as computation accelerators, will be a $5 billion business in time and that nVidia "can make a far greater impact on the industry through GPU computing."

What non-x86 processors lacked was scale. There were no Cell processors in the field, he said by way of example, so no one wrote apps for it, and therefore, no one used it. The GeForce graphics processor, on the other hand, is in millions of computers. By writing apps in nVidia's CUDA language, developers could take advantage of their existing video card in a C-like language and gain tremendous horsepower.

"By putting CUDA on GeForce, we eliminated a chicken and egg problem. The volume was there. Only x86 has more developers. No other computing architecture I know has shipped 120 million units and it's one coherent architecture," he said, in reference to the multiple generations of GeForce.

nVidia currently has 5,000 enterprise customers using CUDA applications in things like energy exploration, drug research and medical imaging, and more than 200 universities around the world teaching CUDA to its students.