Help patching pve-qemu

[esp3r]

Hello Everyone:

I'm using proxmox latest release, i've create a w10 VM with intel gvt-g vga (mediated device), my goal is to have a headless game machine for casual gaming on this little box. I want to patch qemu pve according this reference:

https://adamgradzki.com/2020/04/06/faster-virtual-machines-linux/#fix-qemu-graphics-refresh-rate

I couldn't find any reference to the GUI_REFRESH_INTERVAL_DEFAULT param (to change from 30 to 16) and also any console.h file over pve-qemu repository, so I'm stuck there.

git://git.proxmox.com/git/pve-qemu.git

The general context, I'm getting really low /"laggy" experience on desktop and gaming when i use parsec app for remote access to the vm. (this doesn't happen when running over the same hardware but non virtualized).

I've found that using other tools like steam remote play doesn't have this issue, but this is limited to steam games and it's really tricky jump from steam stream to desktop, thinking in this method as a workarround.

Ther other strange behabiour, i'm seeing "double mouse pointer", also laggy on console, and parsec too. (doen't happen on steam remote play) , i've tried switching "Use tablet for pointer" without success, also installed latest guest agent tools.

Any hint would be helpfull.

Thanks!

 

[H4R0]

If i understand correctly its a vt-d pci passtrough ?

If yes, your qemu patch wont help you at all.

Im also running pci passtrough with a gtx 1080 and a 170hertz monitor and i reach 170fps with gsync.

I never tried parsec but steam play works with full performance no delay for me and you can setup other games not bought in steam you just have to import them.

For desktop use rdp/vnc. Steam play and rdp work completely different, bot are optimized for their work.

 

[esp3r]

If i understand correctly its a vt-d pci passtrough ?

If yes, your qemu patch wont help you at all.

Im also running pci passtrough with a gtx 1080 and a 170hertz monitor and i reach 170fps with gsync.

I never tried parsec but steam play works with full performance no delay for me and you can setup other games not bought in steam.

Hi, It is not a pci passthrough, it's a virtual gpu, with same hardware acceleration specs that the host has, you install the official drivers on the guest, but you don't have a physical output like hdmi/dp/dvi.

https://pve.proxmox.com/pve-docs/chapter-qm.html#_mediated_devices_vgpu_gvt_g

So, the qemu patch should help.

Regarding steam,I couldn't run Legue of legends as non steam title, it fails, but i guess it's related to the path that it uses when adding to the library.

 

[BobMccapherey]

Are you sure that's your problem? Install the intel-gpu-tools package on your host and run "intel_gpu_frequency" while your CPU is under load to check what frequency your Intel GPU is running at. You can also use "intel_gpu_top" to see how much load you're placing on the Intel GPU.

I'm running into a situation where a large amount of load isn't causing the GPU to properly increase its frequency to handle the workload. As a result, I'm getting laggy performance. Using "intel_gpu_frequency -m" is a temporary fix to lock the frequency to max - however, randomly the GPU will override this and stay locked at the minimum frequency. I'm not sure if this is a heat thing (not sure how it could be if this server is in a datacenter), or some sort of BIOS setting that needs to be overridden.

 

[esp3r]

Are you sure that's your problem? Install the intel-gpu-tools package on your host and run "intel_gpu_frequency" while your CPU is under load to check what frequency your Intel GPU is running at. You can also use "intel_gpu_top" to see how much load you're placing on the Intel GPU.

I'm running into a situation where a large amount of load isn't causing the GPU to properly increase its frequency to handle the workload. As a result, I'm getting laggy performance. Using "intel_gpu_frequency -m" is a temporary fix to lock the frequency to max - however, randomly the GPU will override this and stay locked at the minimum frequency. I'm not sure if this is a heat thing (not sure how it could be if this server is in a datacenter), or some sort of BIOS setting that needs to be overridden.

I've noticed the patch applied to the kernel fixed laggy console experience (when you use spice for example to access the windows guest), that was the first symptom.

Related to the gpu frequency issue that you wrote, i tested today again using a mediated device (gvt-g), and noticed the next behaviour:

poweredoff vm: gpufrequency shows speed, 300Mhz.
poweredon vm: gpufrequency show speed, 300Mhz (with or without load)

force -m flag with powered off vm, gpu frequency shows max speed always, once i start the vm, frequency change randomly, I didn't find any related documentatión to gpu frequency and gvtg devices, but i don't think it's related with a heating problem, since when i passthrough the whole igpu it works as expected, no lag.