News for hackers. Stuff that matters.

A comment on AMD May Get Across the CUDA Moat. Parent comment

Re:AMD May Get Across the CUDA Moat by JonChesterfield on Saturday October 07, 2023 @05:54AM (#37800414)
FP64 is what HPC is built on. F32 works on the cards too (same rate or faster). I don't know the status of F16 or F8.

Some architectures provide fast F16->F32 and F32->F16 conversion instructions so you can DIY the memory bandwidth saving - that always seemed reasonable to me, but I don't know if the AMD hardware people are/will go down that path.

[ Parent | on HN ]
Re:AMD May Get Across the CUDA Moat by KeplerBoy on Saturday October 07, 2023 @07:21AM (#37800790)
Sure but Radeon cards are not HPC accelerators. A modest 7800XT for example, which would be a great card for SD, has 76 TFlops@FP16, 37TF@FP32 and 1.16TF@FP64.

Keeping all those FPUs busy is another problem and not easy, but in cases where it can be done FP32 is clearly desirable.

[ Parent | on HN ]
Re:AMD May Get Across the CUDA Moat by londons_explore on Saturday October 07, 2023 @08:10AM (#37801037)
More importantly, if you specify FP16, yet the hardware only supports FP32, then the library should emit a warning but work anyway, doing transparent casts behind your back as necessary.
[ Parent | on HN ]