AMD is held back by their interconnect and firmware disadvantage compared to nvidia. They’ve been trying really hard to create their own cuda, but rocM and HIP still aren’t very popular especially for research.
replyAnd their repeated refusal to either implement CUDA or reimplement everyone's CUDA libraries on their own platform. They say that AMD never misses a chance to miss a chance.
replyHave you ever actually had anyone work with these chips? Developer ux on amd is terrible.
replyYes. We have a quad MI300A server and run several inference models on it. For $107k it has saved us so much money on tokens already and it's a heck of a lot faster than cloud services.
reply> Have you ever actually had anyone work with these chips? Developer ux on amd is terrible.
replyJust how much of dev ux do you need? A foundational library, of course, but as the AI companies keep saying, their models can vibe-code what's needed for those chips anyway.