Power-limit vs TG/s for 2x3090
r/LocalLLaMA
•
Generative AI
Trying to find the sweet-spot to tradeoff between power and tg/s. 250W seems to be a sweet spot for Qwen3.6-27B.