So what exactly is the TDP of the high end Fermi based GeForce? And what about 3D performance? Because without that information you can not logically make the claims you are trying to assert.
OpenCL implementation on ATI is garbage. Talk to any programmer who has done work with ATI and OpenCL and Nvidia CUDA. CUDA is much more straightforward and intuitive and supported, and better performing as well (more direct access to the GPU). And it's about to get a lot better with Fermi. Plus Fermi has ECC which will be the deciding factor for any organizations and institutions who are serious about GPGPU.






Bookmarks