Feature Request - Automatically use all GPU's available for generation

### Feature Idea

I think it would be great if the comfyui interface could automatically use more then 1 GPU when generating allowing for larger models. Similar to how now llama.cpp in text-generation-webui now does.

If the model can be split across and work on say half each if two GPU's speeding up the generation time and creating a larger VRAM space to hold bigger models. Is this possible?

### Existing Solutions

None really.

### Other

_No response_