Feature Request: Parallel Inference Support to Mitigate VRAM Limitations on RTX 4090

### Feature Idea

I use a single RTX 4090 for text-to-image generation. When the FPS and clarity increase, it often leads to VRAM overload. Apart from upgrading to a card with higher VRAM, such as the H100, is it possible to support parallel inference capabilities?

### Existing Solutions

_No response_

### Other

_No response_