Shared dynamic GPU usage across multiple tasks
Shared dynamic GPU usage across multiple tasks
I'm designing a setup for several users to connect remotely for data modeling, CAD work, and gaming. You're considering using two 3080 graphics cards and possibly an Intel Z-series motherboard. Your aim is to manage Windows environments that allow flexible GPU usage—someone can share their GPU capacity while others have more available. I'm drawn to ideas from Linus's One CPU 6 project, but prefer a more fluid approach without strict resource definitions.
I believe we need a system that can manage VMs and allocate resources as required. With Nvidia GPUs, only Quadro cards offer GPU sharing capabilities, which means multiple VMs can use the same GPU if they meet the requirements. For consumer GeForce cards, this isn't feasible. I'm also curious about how this will function under Windows as the host operating system—how will multiple users access the same machine? Did you overlook something?
I realize each user will have their own Windows VM. I'm looking for a fix that works within Windows VMs. One idea I had was to use a Docker layer, allowing users to ask for the GPU they need during a session.