3 ms·
lol this is the wackiest non-news; every serious project has at least some parts of their kernels implemented in PTX/AMDGCN.
by almostgotcaught 2y ago
lol this is the wackiest non-news; every serious project has at least some parts of their kernels implemented in PTX/AMDGCN.
- snovv_crash 2y agoML engineers are finally finding out that their abstraction layers are actually slowing them down.
- almostgotcaught 2y ago> finally finding out there's no finally - perf people have always written kernels in asm.