Dynamic GPU sharing

[Jabali]

I am trying to build a system for multiple people to remotely access for Data Modeling, CAD modeling & gaming.

I would like to use 2 3080's and maybe an Intel Z series motherboard?
My goal is to have a system that can be used remotely by 3-5 people for work and for gaming.

 

The challenge I face is how do set up windows to enable some form of Dynamic GPU sharing?

I would like to have it such that I can use part of my GPU capacity for my work and someone else can use more of it.

If no one else is using the GPU, then I can use all of it and vice versa.

 

I got inspired by Linus's One CPU 6 workstations, but would prefer to not have a hard definition of the resource sharing

 

Thanks

[Uptivuptiz]

I think this calls for a machine that can run VMs for the people that needs its resources. The thing with Nvidia GPUs, though is that only (as far as I'm aware) their Quadro cards support GPU sharing (also called GPU slicing, Nvidia calls its Multi-Instance GPU for their Quadro cards). So sharing one GPU between multible VMs is not possible with consumer for the Geforce cards.

 

Also I'm kinda wondering how you are imagining this will work with Windows as the host OS. How are you going to have multible people use the same Windows maschine? Am I missing something?

Edited October 19, 2022 by Uptivuptiz
Added link for MIG

[Jabali]

My mistake. Each user will be running their own Windows VM. I trying for a solution compatible with windows VMs.

One solution I was thinking of was to have something like a docker layer running and then a user can request a specific amount of GPU as required for that session.