boltbeam.fyi

Llama-3.3-70B-Instruct speed on every GPU

2 measured runs across 1 GPUs, fastest first.

GPUQuantEnginetok/sRooflineShare
2× RTX 3090AWQvLLM35.242.184%
2× RTX 3090Q4_K_Mllama.cpp19.221.390%

Download