PROXMOX 9 GPU PASS THROUGH

[HambergerDaddy]

I was also able to passthrough using just the options in the UI. That was so much better than the old way.

Question, I'm not able currently able to test this myself for reasons.

When using this new method to pass through a GPU, is the GPU still usable (for lack of a better word) by the pve host until the VM which has pass through setup start? or is it like the old way where the gpu becomes unavailable to pve during bootup?

 

[patch]

iGPU is available to the host PVE until a VM is started which uses iGPU pass through

 

[stn]

I'm running a debian 13 vm with a mapped PCIe device (Nvidia RTX2000) and everything (just video encoding so far no output) works so far.

UNTIL

When I try to to increase the ram allocation of the vm it refuses to start. The vm is shown as running, but the vnc console doesn't connect and the vm is not reachable via network.

In the pve webinterface it is not possible to do anything to control this vm. I have to use the command line on the pve to "stop" it successfully.

When I decrease the memory setting again to the initial value (16GB) OR I remove the PCI mapping it starts and works again.

Quite silly - any ideas?

details:

• pve 9.1.7
• Linux host-bla 6.17.13-2-pve #1 SMP PREEMPT_DYNAMIC PMX 6.17.13-2 (2026-03-13T08:06Z) x86_64 GNU/Linux
• 2 node cluster setup with quorum device and ZFS replication
• a dozen other vms doing other stuff
• mapped pci hardware: NVIDIA Corporation AD107GL [RTX 2000 / 2000E Ada Generation] (rev a1)
• no vGPU setup just one vm using that card

vm setup:

agent: 1
boot: order=virtio0
cores: 8
cpu: host
description: debian trixie video backend
hostpci0: mapping=nvgpu,pcie=1
machine: q35
memory: 16384
meta: creation-qemu=10.1.2,ctime=1773331842
name: tvbackend
net0: virtio=BC:24:11:3D:22:22,bridge=vmbr0,firewall=1
numa: 0
ostype: l26
scsihw: virtio-scsi-single
smbios1: uuid=697bc201-30e5-4831-b5b6-111111111111
sockets: 1
tags: trixie
virtio0: prodz:vm-180-disk-0,iothread=1,size=32G
virtio1: rotating-rust:vm-180-disk-0,iothread=1,size=12T
vmgenid: 15c33964-7aac-451d-98a7-111111111111

device mapping on host:

root@host-bla:~# cat /etc/pve/mapping/pci.cfg
nvgpu
map id=10de:28b0,iommugroup=13,node=host-bla,path=0000:b1:00,subsystem-id=10de:1870
map id=10de:28b0,iommugroup=13,node=host-foo,path=0000:b1:00,subsystem-id=10de:1870
description RTX2000