boltbeam.fyi

RTX 4090 for local LLMs

4 measured runs. 24 GB of VRAM; memory measured at 962 GB/s, which sets the roofline for every model below.

ModelQuantEnginetok/sRooflineShare
Qwen3-14BAWQvLLM97.7113.086%
Qwen3-14BQ4_K_Mllama.cpp94.4111.285%
Qwen3.8-27BW4A16-AWQvLLM54.461.788%
Qwen3.8-27BUD-Q4_K_Mllama.cpp50.461.182%

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 →