Hacker News Reader

I would use one. The ISPC compiler recently got support for this new generation of GPUs and in principle it’s exactly the niche they want to hit. The code I write is then used by my research group colleagues.

The same goes for AVX512. It seems niche but then you use it and it’s great, but only a weird subset of laptops have it for now (but MBA 2020 and MBP 13 2020 do have it, MBP 16 doesn’t)

View on HN

How is ISPC's support for AMD MI supercomputing accelerators, or Nvidia's A100s ?

I feel like Hip, OneAPI, ISPC, etc. haven't improved anything over CUDA.

With CUDA we had one proprietary API. Now we have 3-4.

Ispc used to support NVPTX (since it’s based on LLVM) but for obvious reasons they dropped it.

I agree that OpenCL should have won but OTOH most of us have AVX2 or NEON to work with at least and things like ISPC make that trivial to write for, for the workloads where it matters.

Also I think if you have access to an A100 (which is 6k+ eu before tax academic price), you aren’t really worried about using a proprietary API.

Reply on news.ycombinator.com