Reddit user discusses running 2T+ parameter models locally

User with 4× RTX 6000, 7× RX 7900 XTX, and modded 48GB RTX 4090s still struggles with Kimi K3. Post highlights hardware limitations for large-scale local inference.