Proxmox GPU Passthrough

[carminefox]

Hello everyone. I have configured GPU Passthrough on Proxmox with my Nvidia Quadro 4000. The video card is connected to the server, and an HDMI monitor is attached. When I create a Windows VM, for example, and add a PCI hardware device (i.e., my Nvidia Quadro 4000 GPU) in the settings, Windows detects the GPU but it seems that it doesn't use it as the primary GPU (see screenshot). Additionally, the Windows VM is very, very slow. I believe this is because all the video processing is handled by the CPU and not the GPU, making it extremely slow. What could be the issue? The Driver is installed



This is my /etc/modprobe.d/iommu_unsafe_interrupts.conf file:



/etc/default/grub file



/etc/modules file



/etc/modprobe.d/kvm.conf file



/etc/modprobe.d/vfio.conf file



this is the setting of my VM

 

[jeenam]

Try changing your CPU type from host to x86-64-v2-aes. There have been reports of Windows running slow when CPU Type is set to host. Basically test by using any CPU Type except host.

https://forum.proxmox.com/threads/windows-10-vm-so-slow-when-cpu-type-is-host.123186/

 

[carminefox]

Try changing your CPU type from host to x86-64-v2-aes. There have been reports of Windows running slow when CPU Type is set to host. Basically test by using any CPU Type except host.

https://forum.proxmox.com/threads/windows-10-vm-so-slow-when-cpu-type-is-host.123186/

The problem is that I need to get GPU Passthrough working

 

[jeenam]

The problem is that I need to get GPU Passthrough working

You mentioned the system is running slow. Change the CPU Type to try and fix that. Re GPU, the GPU is showing up without any error messages in Device Manager. It would seem to be working properly.

Can't comment on why the host won't boot when the GPU is configured as the Primary. However, there are plenty of reports of the A4000 working just fine for GPU passthrough so it'll be interesting to hear what the issue is.

 

[carminefox]

You mentioned the system is running slow. Change the CPU Type to try and fix that. Re GPU, the GPU is showing up without any error messages in Device Manager. It would seem to be working properly.

Can't comment on why the host won't boot when the GPU is configured as the Primary. However, there are plenty of reports of the A4000 working just fine for GPU passthrough so it'll be interesting to hear what the issue is.

Thank you very much. Yes it is really interesting to understand why. I really need that windows VM to work with the GPU, I need it for small renderings.

 

[carminefox]

Jul 31 16:47:54 animalspbx pvedaemon[1327]: <root@pam> end task UPID:animalspbx:000010AE:00005478:66AA4E97:qmstart:107:root@pam: OK


Jul 31 16:49:25 animalspbx pvedaemon[1326]: VM 107 qmp command failed - VM 107 qmp command 'guest-ping' failed - got timeout


Jul 31 16:49:44 animalspbx pvedaemon[1326]: VM 107 qmp command failed - VM 107 qmp command 'guest-ping' failed - got timeout


Jul 31 16:50:04 animalspbx pvedaemon[1325]: VM 107 qmp command failed - VM 107 qmp command 'guest-ping' failed - got timeout


Jul 31 16:50:13 animalspbx pvedaemon[1326]: <root@pam> successful auth for user 'root@pam'


Jul 31 16:50:17 animalspbx pvedaemon[1326]: VM 107 qmp command failed - VM 107 qmp command 'guest-ping' failed - got timeout


Jul 31 16:50:35 animalspbx kernel: tap107i0: left allmulticast mode


Jul 31 16:50:35 animalspbx kernel: fwbr107i0: port 2(tap107i0) entered disabled state


Jul 31 16:50:35 animalspbx kernel: fwbr107i0: port 1(fwln107i0) entered disabled state


Jul 31 16:50:35 animalspbx kernel: vmbr0: port 5(fwpr107p0) entered disabled state


Jul 31 16:50:35 animalspbx kernel: fwln107i0 (unregistering): left allmulticast mode


Jul 31 16:50:35 animalspbx kernel: fwln107i0 (unregistering): left promiscuous mode


Jul 31 16:50:35 animalspbx kernel: fwbr107i0: port 1(fwln107i0) entered disabled state


Jul 31 16:50:35 animalspbx kernel: fwpr107p0 (unregistering): left allmulticast mode


Jul 31 16:50:35 animalspbx kernel: fwpr107p0 (unregistering): left promiscuous mode


Jul 31 16:50:35 animalspbx kernel: vmbr0: port 5(fwpr107p0) entered disabled state


Jul 31 16:50:35 animalspbx qmeventd[1098]: read: Connection reset by peer


Jul 31 16:50:35 animalspbx pvedaemon[1327]: VM 107 qmp command failed - VM 107 qmp command 'guest-ping' failed - client closed connection


Jul 31 16:50:35 animalspbx pvedaemon[1325]: VM 107 qmp command failed - VM 107 not running


Jul 31 16:50:35 animalspbx pvestatd[1274]: VM 107 qmp command failed - VM 107 not running


Jul 31 16:50:35 animalspbx systemd[1]: 107.scope: Deactivated successfully.


Jul 31 16:50:35 animalspbx systemd[1]: 107.scope: Consumed 7min 47.147s CPU time.


Jul 31 16:50:37 animalspbx qmeventd[5841]: Starting cleanup for 107


Jul 31 16:50:37 animalspbx qmeventd[5841]: Finished cleanup for 107

These are the journalctl logs when I try to start the VM with Primary GPU enabled