boltbeam.fyi

Qwen3-8B speed on every GPU

6 measured runs across 3 GPUs, fastest first.

GPUQuantEnginetok/sRooflineShare
H100 80GBQ4_K_Mllama.cpp227.0684.433%
RTX 3090Q4_K_Mllama.cpp124.5188.966%
RTX 5090BF16TensorRT-LLM95.9112.485%
RTX 3090BF16llama.cpp54.058.992%
RTX 3090BF16TensorRT-LLM51.458.987%
RTX 3090BF16vLLM50.258.985%

Download