vgpu-proxmox VS LookingGlass

This is a good guide on setting up vGPU. https://gitlab.com/polloloco/vgpu-proxmox

A better way [for me, anyways] is getting GRID drivers running. However, this only works with the 9xx cards up to the 2080.

https://gitlab.com/polloloco/vgpu-proxmox

We've followed PolloLoco's guide and confirmed we had vGPU functioning on the host and could pass vGPU with an A profile to a Windows 10 VM. Profile was overridden to the following (NOTE: nvidia-49 is a Q profile):

Look into proxmox and vgpu, craft has a couple videos on it itself. If you have an ampere + (rtx 3k,rtx 4k series)card, you're out of luck. At least for the "free" version. Also, don't use his videos as step by step guides. They're outdated but good for a general "road map" of his it'll be to set up. I followed this gitlab to setup a test server with a 1660 ti. In theory you should be able to do a windows and Linux VM off the same card. I say in theory because the 5 minutes I tried to set it up it didn't work. I was in a hurry and just haven't gotten back to trying it out.

Although not quite what you want, the T1000, as a Turing-based card, is one of the ones that can support the vGPU hack: https://gitlab.com/polloloco/vgpu-proxmox

For the GPU portion, you can unlock vGPU support in proxmox for free with this script

Most AI projects are going to need a GPU to work. If you really want to get into AI I recommend moving to proxmox and use a Vgpu unlock script to split your GPUs between VMs instead of being limited to one pass through GPU per VM.

Unfortunately AMD cards suffer from a reset bug, still.

The reset bug being that you can pass through the card fine, once. But if you try to pass it through again (or the card experiences an issue and needs to reset), they get caught in some kind of bad state and won’t work until power is removed and restored. Which requires a reboot or a only slightly less disruptive dance with system power states.

For vega and 5000 series gpu’s, there’s https://github.com/gnif/vendor-reset

Incidentally, nvidia gpus are so good at resetting, they’ve probably done so without you noticing. If the screen ever goes black for a fraction of a second and returns in normal usage, it was probably because it reset itself.

The lower 6000 series lower than the 6800’s for example may or may not have the issue. It seems most “reference” cards are fine, but custom vendor cards often but not always have issues. My reference 6700 works fine, but a sapphire 6700 probably won’t.

And the 7000 series is also fucky in a new way somehow. Gnif knows far more about this than me, and has basically thrown up his hands at how AMD doesn’t care. He’s made occasional posts about it on https://forum.level1techs.com/

Gnif is also responsible for Looking glass: https://github.com/gnif/LookingGlass

You could try LookingGlass. That may require two GPUs.

It's a pair of apps, one runs on a Windows virtual machine, the other on the host OS, that uses shared memory to copy a passed-through GPUs frame buffer. Runs fast enough to get 4k/120fps very low latency, so if you have a spare GPU you can game on it in Windows, from a Linux desktop. https://github.com/gnif/LookingGlass

Check out https://github.com/gnif/LookingGlass if you can accept a separate window.

then everything looks normal. It's all up to EGL though, as looking-glass really doesn't have any code that handles image scaling (see here).

https://github.com/Arc-Compute/LibVF.IO/tree/master/ plus https://github.com/gnif/LookingGlass works pretty well. If you use an Intel GPU, particularly one of their new Arc dedicated GPUs, it supports the functionality on the consumer grade hardware without any trickery and you just need Looking Glass to map the outputs.

I'd like to introduce you to Looking Glass as a much higher quality and lower latency alternative to NDI when you are trying to relay frames between virtual machines within the same computer. With a patched nvidia driver to enable nvfbc on the gaming VM there is almost zero performance penalty when capturing gameplay. It also has options for audio, which I don't use because they didn't exist when I was configuring my system. I have a similar use case (gaming and streaming) and hardware (5950X + two gpus). I am not using proxmox, however. My configuration using QEMU and libvirt has this XML:

What stopping from tagging a stable release? The milestone has no open issues.