Day 2 Keynote
The main event started off with a bang. NVIDIA’s CEO, Jen-Hsun Huang took the stage and discussed many of the exciting technologies being shown and discussed at this years GTC including some of the emerging companies being shown that I’ll cover in another article. The first demo JHH displayed was of the upcoming HAWK2 game being developed by Besot. HAWK was a flight combat game using DirectX 10.1. HAWK2 is the sequel using features like Tessellation and Shader Model 5.
Jen-Hsun Huang’s speech also covered computing as the third pillar of science with applications in Drug Design, Seismic Imaging, Automotive design, Medical Imaging, Astrophysics, dress design and much more. Tony Tamasi, NVIDIA’sSenior VP or Product Development was next on the stage talking about NVIDIA’s momentum with CUDA with over 668000 cumulative downloads and 9 Tesla OEMs with over 334 GTC submissions all double the previous year.
The next announcement brought an audible gasp out of the audience, the announcement of CUDA x86. Now I know what you must be thinking but no it doesn’t mean NVIDIA has an x86 license and developing an x86 CPU. It is in fact an x86 compiler to allow CUDA applications to be run on Intel or AMD CPUs in conjunction with PGI.
Jen-Hsun and the other NVIDIA presenters at the Keynote showed off some impressive looking demos including a Tessellation demo with over a billion polygons and a Lighthouse demo showing realistic movement of waves against a shore. From the perspective of a gamer, the ability of games to have millions or more polygons means better graphics.
The iRay and Mental Images demos were very impressive. They showed off how the GPU could render a realistic scene in a matter of seconds and change the scene. There were also demos of software that used physics and the power of the NVIDIA GPU to render cloth to allow the company to design dresses and change it on the fly. Not only can you view the dress in 3D, you can see how it would move on the body once it’s put on.
The big announcement was of NVIDIA’s roadmap. I’m sure if you follow NVIDIA’s past you know they release a new architecture every year or two and have for a long time with a “kicker in the middle.” NVIDIA’s next generation architecture is called Kepler, following the tradition of the Fermi in naming architecture after a famous scientist. This will be NVIDIA’s first 28nm product and should launch in the 2H of 2011. NVIDIA says it should have 4x the performance per watt of Fermi.The generation after Kepler is called Maxwell and should be on the next process node reduction, likely 22nm. Each new architecture should have a kicker in between at a yearly interval with more performance. Jen-Hsun Huang stated that there will be new features including virtual memory with page fault correction.
So NVIDIA confirms they will have next generation architecture next year on the 28nm process node, which should mean lower power consumption and less heat for the chips. As you know, the current high end of the NVIDIA graphics chips is the GeForce GTX 480 and they are notoriously hot and draw a lot of power compared to their competition from AMD. One could surmise that the next graphics chip architecture will have improved performance compared to the current generation, but using performance per watt as a metric during the keynote likely means they are targeting the same 2x performance increase as they had from GeForce 5 to GeForce 6 from 6 to the 8 series from 8 series to GeForce GT200 and from GT200 to G100 which is now the Fermi architecture while drawing less power than the current generation and or improving DP FP power.It should be interesting to see if NVIDIA delivers on time and with the improved performance per Watt. The next part of this article covers some of the companies I met with during the Emerging Companies Summit with more coverage in the next part of my GTC coverage.