Re:Triton: Open-Source GPU Programming for Neural Networks
by croes on Wednesday July 28, 2021 @01:59PM
(#27987104)
It's not about performance but open standards.
Remember Oracle vs. Google, at some time in the future NVidia could decide to get money out of CUDA.
[ Parent
| on HN ]
Re:Triton: Open-Source GPU Programming for Neural Networks
by einpoklum on Wednesday July 28, 2021 @04:15PM
(#27989150)
Well, yes and no. Neither AMD nor NVIDIA are committed to open, libre, standards - they've both been "betraying" OpenCL (just like Apple and Intel, I should mention). So we have AMD's not-open (?) ROCm and NVIDIA's not-open CUDA.
Also, if we only cared about open standard we'd simply not use a GPU at all and settle RISC-V chips :-)