boltbeam.fyi

H200 for local LLMs

3 measured runs. 140 GB of VRAM; memory measured at 4662 GB/s, which sets the roofline for every model below.

ModelQuantEnginetok/sRooflineShare
Qwen3-14BAWQvLLM166.9547.330%
Qwen3.8-27BW4A16-AWQvLLM100.8299.134%
Qwen3.8-Flash-NextUD-Q4_K_XLllama.cpp100.4709.114%

Share = measured tok/s over the roofline: the bytes a token reads over this GPU's measured memory speed. Open a quant for the test machine and the per-role and per-kernel breakdown.

Open the explorer on this GPU →