@thomrstrom@triangletoot.party
2026-08-08 00:35 UTC
On a related note, I'm upgrading our processing & training cluster to 490 cores this weekend to keep up with the ingest rate and the increasing complexity of our reinforcement learning loop. Special thanks to #Ampere and their super-efficient #arm64 chips for making sure I can do this without turning my basement into a sauna.
Replies (1)
-
@synlogic4242@social.vivaldi.net 2026-08-08 00:45
@thomrstrom@triangletoot.party I have a TODO to figure out how to buy/build the cheapest home cluster I can for massively-ish parallel test job running. core count more important than clock. want to avoid the overhead tax for spinning up ephemeral equivs in cloud. I'd love to get to 256 cores or better before year's end. both for work projects and to help me do research for my perf book