For me, I would love to have a single GPU in my server that I can split up for use in transcoding videos for Plex in one VM, and another VM running something like Blue Iris with AI video analysis.
The potential use cases are many and varied, including some gaming use cases. You could have a single GPU in your Linux desktop, and be able to pass that through to a Windows VM to get native performance gaming in a VM. This is technically already possible, but you need two GPUs. With SR-IOV you could get away with only having one
Not really ai specifically but VMs. Maybe you want a Windows vm for gaming with a gpu, just give it a slice and it’s fine. Maybe you want lots of VMs for all various different office clients, split off sections of the gpu and you can have a bunch of hardware accelerated thin clients
If it has SR-IOV in the consumer version then yes I need it. Otherwise f*ck both AMD & NVIDIA.
I’ve been out of the loop for some years, could you eli5 for me please?
SR-IOV allows you to share your GPU among many virtual machines in much the same way that you are able to share a single CPU among many VMs
Thanks for the info. I guess that’s not that essential for gaming but more for AI specific tasks?
For me, I would love to have a single GPU in my server that I can split up for use in transcoding videos for Plex in one VM, and another VM running something like Blue Iris with AI video analysis.
The potential use cases are many and varied, including some gaming use cases. You could have a single GPU in your Linux desktop, and be able to pass that through to a Windows VM to get native performance gaming in a VM. This is technically already possible, but you need two GPUs. With SR-IOV you could get away with only having one
Ok, now I want it too. Thanks for the explanation!
Not really ai specifically but VMs. Maybe you want a Windows vm for gaming with a gpu, just give it a slice and it’s fine. Maybe you want lots of VMs for all various different office clients, split off sections of the gpu and you can have a bunch of hardware accelerated thin clients
I’ve never thought about that, but it makes sense. Thanks for the explanation!
For real
Okay a quick search tells me this is short for Single-root input/output virtualization right? Can you explain why that would be advantageous in a GPU?
See here