Wondering if anyone could provide accurate supportive or negating comments re: this blog speculation which popped up in daily Google watch??? → http://www.kitguru.net/components/graphic-cards/zardon/nvidia-to-drop-cuda-on-gtx465/
Total hogwash?
Only for low end cards?
Longer term shift to OpenCL?
Other?
Would appreciate any perspective esp. as many are dedicating considerable time & effort to CUDA apps.
TIA, V.
That piece was written by someone with about as much understanding of GPU hardware and software as the sandwhich I ate for lunch.
There is no attribution except “Asia” and “rumours”. KitGuru is interviewing a sock puppet, except that the sock is missing and he’s just mouthing the words with his fingers.
This rumor is ridiculous, and dropping CUDA for OpenCL would save NVIDIA very little money, except perhaps the ability to fire some Open64 compiler developers. OpenCL and CUDA are two different high level languages that compile down to PTX and run on the same hardware.
By redesigning their chips from the ground up to be more general purpose computing devices, NVIDIA has basically committed themselves to supporting CUDA/OpenCL/whatever GPGPU programming language comes next. There is no part of the GPU die that you could label as “CUDA-specific” anymore, and therefore save money by removing. (It sounds like there are a few parts of the chip that you could label as “graphics only,” supporting features not yet exposed to CUDA.)
Really dropping CUDA (and reducing production costs) would require a total GPU redesign, which makes no sense for the GTX 465.
The GTX465 is about to be released and will just be another GF100 card.