Single-user llama.cpp tensor-parallel benchmark of Qwen3.6-27B on 4x modded 20GB RTX 3080 (in-China box), with a self-contained dashboard.
By chatting or signing in you agree to the Terms and chat-message logging (revocable in History).