[mycall]

DGX does parallel inference with llamacpp/vLLM much better than AMD GPU at 128GB VRAM.