Reddit user builds 448GB VRAM rig to run MiniMax M3 locally

A Reddit user assembled a local LLM rig with 12 GPUs (2x RTX Pro 6000, 8x RTX 3090, 2x RTX 5090) for 448GB VRAM. It runs MiniMax M3 in AWQ-INT4 on vLLM, achieving ~30 tokens/s. The build also uses a Threadripper 9960x and 3 PSUs.