[Feature] Select GPU for Image generation / TTS

[MerijnHendriks]

Describe the Issue

I have a system with multiple GPUs.
Two blackwell-based NVIDIA GPU's run a text model with no capacity to spare, and a third RDNA2-based AMD iGPU is free to use.
But I can't assign the image generation and TTS specifically to the AMD iGPU, resulting in crashes for lack of VRAM available on the NVIDIA GPUs.

Request

I want to be able to select on which GPU everything image-related (model, VAE, Clip, T5) and TTS-related will run.

[henk717]

This is currently possible by running the image gen in a separate instance, using the network tab you can change the port the image gen instance runs on.

Even if we add this its unlikely that we will allow this multi vendor outside of the vulkan option as they aren't cross compatible otherwise. So doing this with separate instances should be the best way to do it long term.

If you want to do this inside KoboldAI Lite we have an option to provide an A1111 API server, for the image generation, your second KoboldCpp can be used with this option.