I’m calling this the Monstrosity. 5 ex mining BC-250 boards Qwen3-Coder-Next Q4 at 40 tok/s

Using an asrock 12 unit case running one board as the main with the rest of them headless. About 71GB of vram exposed. So far 40 tok/s is with 30k context and it dips to around 30 tok/s at 100k context. This is all over the 1gb Ethernet that is on the boards already. I have 2 more of these and will probably get them running to see if 3.8 flash next runs at usable speeds. This setup is wildly inefficient with power but cost me less than $800.