I am using up-to-date Proxmox VE 8.3 (pve-manager/8.3.4/65224a0f9cd294a3 (running kernel: 6.8.12-8-pve)). I am trying to pass through two Nvidia Geforce 3060 cards to be used for ML, not as primary display.
I've passed one 3060 as primary display for a Win11 VM before and passing both as non-primary displays works fine on Debian Bookworm. But it fails with any newer Linux distro like Manjaro or Debian Trixie Testing which both use kernel 6.12.x. When starting those VMs the process will take very long before eventually failing with a timeout or sometimes with a "qemu exited with code 1" message. For some reason, the Manjaro VM managed to start once, but this, too, took a long time. Both cards were available then within the VM. Ever since, it keeps failing on start. I'm at a loss here.
On the host system, I am seeing ever increasing RAM usage for the 6.12 VMs with 1 core at full load while starting. I assume the "code 1" happens when all RAM is used up and the timeout happens if the timeout is reached before running out of RAM.
Is this in any way a known issue? Let me know which specs you need for troubleshooting. Thank you!
I've passed one 3060 as primary display for a Win11 VM before and passing both as non-primary displays works fine on Debian Bookworm. But it fails with any newer Linux distro like Manjaro or Debian Trixie Testing which both use kernel 6.12.x. When starting those VMs the process will take very long before eventually failing with a timeout or sometimes with a "qemu exited with code 1" message. For some reason, the Manjaro VM managed to start once, but this, too, took a long time. Both cards were available then within the VM. Ever since, it keeps failing on start. I'm at a loss here.
On the host system, I am seeing ever increasing RAM usage for the 6.12 VMs with 1 core at full load while starting. I assume the "code 1" happens when all RAM is used up and the timeout happens if the timeout is reached before running out of RAM.
Is this in any way a known issue? Let me know which specs you need for troubleshooting. Thank you!
Last edited: