[FEA] Improve support for machines with mixed GPU types

[will-hill]

For systems with different types of GPUs, it would be nice to know what GPU is which in the dashboard.
I assume this is not a problem when all your GPUs are the same type, but if you have different types, than it is difficult to know which device I am profiling.

Also, I think it would help the experience to draw an outline indicating the max VRAM in the 'GPU Memory' widget.

[jacobtomlinson]

I think overall this issue covers adding quality of life improvements for mixed-SKU systems. It's unusual to see a mix of different GPU types in the same machine. Perhaps home brewed dev machines would have this configuration?

We use the indices to tell the GPUs apart. GPU 0 in the dashboard will match with GPU 0 in nvidia-smi. Are you suggesting we add the SKU (e.g L4, B100, A10g, etc)?

Adding a max memory line to the graph makes sense if you have a mix. Currently the assumption is they are all the same and the graph just shows min and max.

[will-hill]

Yes, adding the SKU (e.g L4, B100, A10g, etc) would improve the DX for mixed device folks like myself. And would just look cool, too.

Max memory line would also improve the DX. I use nvdashboard for both Data Science (cuDF, cuML) and LLM deployments.

I actually use nvdashboard for tuning and checking memory usage, even when I am not in a Jupyter environment.