To maximise the VRAM available for an LLM on the same machine. That's why I asked myself the same question, anyway.