Replies (3)

Havok's avatar
Havok yesterday
So go for the 200B or Bust? I Figured Qwen 3 72B would be good. I don't thing the 3090 is a good trade due to power efficiency etc. With a cluster of GB10 a 200B is in reach, slower by a good margin but still within acceptable rates for tokens/s. No? I mean heck a 405B should be doable with somewhat acceptable lag on a cluster.
โ†‘