After our recent series of articles about the state of stream computing today, from this general introduction to the many CUDA-related developments spearheaded by NVIDIA, we now turn our attention to the other GPU company, ATI.
In fact, ATI is one of the pioneers of this emerging field. Some of you may remember that the first demonstration of Stanford's Folding project on a GPU happened to use the ATI Radeon X1000. Havok FX physics and other high performance applications that used the GPU for number crunching were also demoed. That was in 2006. A year before, ATI showed off physics simulations using the GPU. Those were the days when Ageia was touting its PhysX physics processing unit (PPU) and ATI was responding to this with talk of possible multi-GPU configurations where one GPU could be tasked to handle the physics calculations.
With such a bright start, ATI should be disappointed that its rival, NVIDIA has grabbed most of the subsequent headlines. Since CUDA made its debut, NVIDIA has been pushing its GPU computing initiative aggressively. In terms of marketing dollars and publicity, NVIDIA has definitely made its presence felt, from university courses on CUDA to GPU acceleration on Adobe CS4 to GPU-powered Tesla supercomputers. Folding too has been ported to the NVIDIA CUDA platform.
It has been a non-stop barrage of CUDA related news from NVIDIA, especially after the introduction of NVIDIA's GeForce 8 series of GPUs that support CUDA fully. Additionally, the acquisition of Ageia by NVIDIA has brought PhysX to CUDA, increasing its scope to gaming. Quite a few developers like EA and 2k Games have already started using PhysX in their latest games. It's apparent that NVIDIA is building its GPU products around its approach of Graphics Plus, where it's not just about the graphics performance and games. There's little doubt which GPU computing platform has the mind and market share at the moment (though this could be moot with OpenCL support from both parties).
ATI's response has been rather muted by comparison. We rarely hear about its stream computing initiatives besides the occasional update on Close to Metal and later, the all-in-one Stream SDK package. It certainly has no GPU physics to tout of, with Havok now in Intel hands (a potential rival now in this area with Larrabee). Many of the promising technologies that we saw a couple of years earlier have also not played out as planned. Perhaps it was the acquisition of ATI by AMD that put a kink in their plans but late last year, we finally saw something new from ATI regarding stream computing.
Coinciding with the increased interest in stream computing, due to the ratification of OpenCL 1.0, ATI made an update to its stream computing initiative recently. Chief among the announcements from the company include the inevitable pledging of support for OpenCL (they worked on the open standard after all) and a promise to make it work with it. We paid a visit to ATI's local offices in December last year to get the news firsthand.