libvirt 10.0 Released With QEMU VM Migration Improvements

[DumbFsck]

Does GPU pass-through require a dedicated GPU to the guest VM, so in my case the iGPU for the host and dGPU for the guest? My understanding is that is indeed the case, but since still learning asking here for confirmation (or to stand to be corrected!) Of course, I have both the iGPU and dGPU in this particular case, just looking to confirm this is a hard requirement for pass-through (and I assume yes on what I have read, and also logically makes sense if you passing through the whole GPU to the guest - but then again, still learning!)

Thanks!

It does not require, per se, you can search for libvf io, sriov, gvt-g and lots of other GPU paravirtualization related terms and technologies.

I tried braving those grounds, but the amount of thorns I got cut by put me off - if you, like me, had an AMD gpu and a CPU with no integrated graphics, I'd offer my thoughs and prayers and all the best.

Since you DO have integrated graphics, I'd say the bother of splitting one gpu between systems (a host and vm client) is not worth it. Do it with the whole gpu.

There will still be some sharp edges, like if your gpu complains about resetting, but with the Arch wiki on your hands and granite resolve on your mind, it honestly shouldn't be too hard.

Maybe one day AMD will realise these features are not only for "pro" cards, and honestly there is nothing tech-wise to prevent consumer GPUs from being shared by multiple OSs at the same time, but from a recent video from the level1linux YouTube channel, titled 'Adventures at AMD: AMD and the Linux Kernel' - this 'one day' is a future still very far from the here and now. They talk about it at the end, can't remember the exact time.

[zexelon]

Since I am still learning here, and looking for something similar possibly, asking... I have a desktop PC at work where I am using the iGPU on the Intel CPU, but there is also an nVidia graphics card.

Does GPU pass-through require a dedicated GPU to the guest VM, so in my case the iGPU for the host and dGPU for the guest? My understanding is that is indeed the case, but since still learning asking here for confirmation (or to stand to be corrected!) Of course, I have both the iGPU and dGPU in this particular case, just looking to confirm this is a hard requirement for pass-through (and I assume yes on what I have read, and also logically makes sense if you passing through the whole GPU to the guest - but then again, still learning!)

Thanks!

The short answer is yes, you have to dedicate the GPU to the VM if you are using Nvidia or AMD. I have had tons of success passing through Nvidia to both windows 10 & 11, and have never tried AMD so cant speak to that one. A little while ago MS issued an update to 11 that did tank performance, so I am back to windows 10 and it is operating well.

There is tons of really good information in the Arch docs as mentioned that applies to a number of different distros. If you are using Pop_OS though there are a couple of key differences in the way Pop_os boots using systemd so you may have to look for some distro specific information there.

On an older system I used the Intel iGPU for Linux and passed through the Nvidia to windows and this worked the best (was also the simplest). On the current rig I dont have an iGPU anymore, just two Nvidia cards, so its a bit different.

[qarium]

You'd need gpu pass through or use virgl, though I'm unsure of the virgl support in windows.

on my system i write right now threadripper1920X+Vega64 PCIe gpu passthrough is not possible because the host OS need graphic card to

on my AMD PRO w7900 something like PCIe gpu passthrough would be possible because the ryzen 7950X3D has a iGPU for linux

but of course something like this is not favored because the best case would be to use the big GPU in linux and windows.

isn't virgl only for openGL ? i think there is a vulkan project to with similar goals. and directX12?

[Anux]

libvirt QEMU VM
it would be nice if AMD could make it possible to use such a GPU also in a VM in Windows 11

For Win guest you only have 2 options:

• GPU pass through with 2 GPUs (integrated and discrete)
• GPU pass through while the host is running headless, needs only one GPU but more hassle

Sadly no virGL guest driver for Windows so that option is Linux only.

If everything works somewhat ok in wine I wouldn't use GPU pass through, there are many quirks to get latency to acceptable levels while running in VM, and the next big update could screw you.

[ehansin]

Thanks everyone for the additional feedback here, this has been helpful and informative!

[qarium]

For Win guest you only have 2 options:

• GPU pass through with 2 GPUs (integrated and discrete)
• GPU pass through while the host is running headless, needs only one GPU but more hassle

Sadly no virGL guest driver for Windows so that option is Linux only.

If everything works somewhat ok in wine I wouldn't use GPU pass through, there are many quirks to get latency to acceptable levels while running in VM, and the next big update could screw you.

yes thank you this ryzen 7950X3D+PRO w7900 system has 2 gpus so the first pass through option would be possible but then i would have 2 monitor outputs. and would need 2 monitors or i would need to switch the first monitor to the second output.

yes to be honest this is not the solution i search for what i really want is fully integration means with full supported OpenGL/vulkan/directX Guest driver fully integrated into the linux system and not a linux only solution i mean a solution to make windows 11 run in linux without trouble.

[S.Pam]

Seems the developer interest for virtual gpu support outside of enterprise solutions like Citrix HD is rather limited. Hopefully we will see improvements now when Microsoft is developing support via WSL.

There seems to be several other projects looking to enable acceleration in Windows, but status is unknown. At least opengl seems somewhat possible, but vulkan or directx is not near ready I think.

https://github.com/tenclass/mvisor-win-vgpu-driver

https://gist.github.com/Keenuts/199184f9a6d7a68d9a62cf0011147c0b

https://docs.mesa3d.org/drivers/venus.html

[qarium]

Seems the developer interest for virtual gpu support outside of enterprise solutions like Citrix HD is rather limited. Hopefully we will see improvements now when Microsoft is developing support via WSL.
There seems to be several other projects looking to enable acceleration in Windows, but status is unknown. At least opengl seems somewhat possible, but vulkan or directx is not near ready I think.

https://github.com/tenclass/mvisor-win-vgpu-driver

https://gist.github.com/Keenuts/199184f9a6d7a68d9a62cf0011147c0b

https://docs.mesa3d.org/drivers/venus.html

sadly the use of OpenGL in windows11 is very limited.

with vulkan means venus people could run directx11/12 to vulkan layers like proton do ?