Re:I'm making LLMs work on small and old devices
by rpcope1 on Friday November 13, 2026 @11:18AM
(#50485713)
How about weird devices? In the race to pick esoteric devices that can do matmul, I've been wondering if the Xeon Phi cards that are floating around for scrap prices can be made to do something. I also wonder if you couldn't canabilize a slightly aged IBM i, especially if it had one of the newer power architectures with better matrix support as a sort of LLM accelerator.
[ Parent
| on HN ]
There are no comments on this comment yet.