AnalysisAI ModelsSeptember 24, 2026

Two BC-250 mining APUs run Qwen3.6-35B-A3B at 60 tok/s locally

Read original source →reddit.com

Two $115 BC-250 ex-mining APUs with ~27GB combined GPU memory run Qwen3.6-35B-A3B Q4_K_M at 60 tok/s with 64k context via llama.cpp Vulkan and RPC on Bazzite. Total build cost around $300 including PSU; boards link over 1Gb Ethernet.

1 source

More stories today

Open the live feed