logoalt Hacker News

fwiptoday at 1:48 AM1 replyview on HN

Doesn't seem especially relevant for LLMs. Especially not for training, but even for inference.


Replies

nomeltoday at 2:11 AM

Sure, but it seems very reasonable to expect some co-hosting of latency sensitive tasks, especially with some low hanging thin client fruit, like cloud gaming and the like. I've been very happy not burning a thousand watts of $0.45/kWh power locally, for my AC to force out of the house, to run blender.