- 2x RTX Pro 6000 Max-Q (96GB) - 8x RTX 3090 (24GB) - 2x RTX 5090 (32GB) - 3 PSUs - 128GB DDR5 SDIMM RAM (4-channel) - Threadripper 9960x - 1x Ryobi Portable Fan - 1x large Uber Eats bill 448GB VRAM Running MiniMax M3 in AWQ-INT4 on VLLM via PP over TP groups of 2. ~30 tp/s per single stream ~960 tp