Cheap $20 USB Ethernet multi-node GPU runs 39.7GB LLM at 30 t/s

A Reddit user demonstrates multi-node inference on a 39.7GB laguna Q2_K_XL model using two RTX 4060s and one RTX 4060 connected via a $20 USB-to-Ethernet direct link, achieving ~30 tokens/s. Inter-GPU traffic peaks at 30-70 MB/s, showing expensive networking is not required for modest multi-node setups.