Proxmox GPU Passthrough (NVIDIA, AMD, Tesla)

GPU passthrough lets you dedicate a physical GPU to a virtual machine with near-native performance — useful for gaming VMs, remote workstations, or AI inference inside an isolated environment.

What you need: a CPU with IOMMU support (Intel VT-d or AMD-Vi), a motherboard BIOS that exposes it, Proxmox VE 8.x, and ideally a secondary GPU for the Proxmox host. Intel integrated graphics on the host while passing a discrete GPU to the VM is the cleanest setup.

Step 1: Enable IOMMU

In BIOS/UEFI: Intel → enable VT-d. AMD → enable AMD-Vi or IOMMU.

Edit /etc/default/grub:

Intel:

GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_iommu=on iommu=pt"

AMD:

GRUB_CMDLINE_LINUX_DEFAULT="quiet amd_iommu=on iommu=pt"

iommu=pt reduces overhead for devices not being passed through — leave it in regardless of vendor.

update-grub && reboot

Verify after reboot:

dmesg | grep -e DMAR -e IOMMU

Intel should show DMAR: IOMMU enabled, AMD shows AMD-Vi: Supported feature. Nothing means IOMMU isn’t active — go back to BIOS.

Also verify interrupt remapping is enabled — passthrough won’t work without it:

dmesg | grep 'remapping'

You should see AMD-Vi: Interrupt remapping enabled or DMAR-IR: Enabled IRQ remapping. If not, and your hardware doesn’t support it, you can allow unsafe interrupts as a workaround:

echo "options vfio_iommu_type1 allow_unsafe_interrupts=1" > /etc/modprobe.d/iommu_unsafe_interrupts.conf

Check IOMMU Groups

for d in /sys/kernel/iommu_groups/*/devices/*; do
    n=${d#*/iommu_groups/*}; n=${n%%/*}
    printf 'IOMMU Group %s ' "$n"
    lspci -nns "${d##*/}"
done

Your GPU and its audio device need to be in the same IOMMU group, ideally isolated. If they share a group with a SATA controller or USB hub, see ACS override in the troubleshooting section.

Step 2: Load VFIO Modules

Add to /etc/modules:

1
vfio
2
vfio_iommu_type1
3
vfio_pci
4
vfio_virqfd

update-initramfs -u -k all && reboot

Step 3: Bind the GPU to VFIO

Find your GPU’s PCI IDs:

lspci -nn | grep -iE "nvidia|amd|vga|3d"

Output example:

1
01:00.0 VGA compatible controller [0300]: NVIDIA GeForce RTX 3080 [10de:2206]
2
01:00.1 Audio device [0403]: NVIDIA HD Audio [10de:1aef]

Bind both the GPU and its audio device to VFIO:

echo "options vfio-pci ids=10de:2206,10de:1aef disable_vga=1" > /etc/modprobe.d/vfio.conf
update-initramfs -u -k all && reboot

Verify:

lspci -nnk -d 10de:2206

Look for Kernel driver in use: vfio-pci. If you see nvidia or nouveau instead, blacklist them:

echo "blacklist nouveau" >> /etc/modprobe.d/blacklist.conf
echo "blacklist nvidia*" >> /etc/modprobe.d/blacklist.conf
update-initramfs -u -k all && reboot

Step 4: Create the VM

In the Proxmox web UI, these settings matter:

• Machine: q35 — required for PCIe passthrough
• BIOS: OVMF (UEFI) — add an EFI disk when prompted
• CPU type: host — don’t use kvm64 or default
• Memory: disable ballooning, set a fixed amount
• OS: for Windows, add the VirtIO drivers ISO as a second CD-ROM

Add the GPU

Hardware → Add → PCI Device, then check:

• All Functions — passes GPU + audio together
• Primary GPU — if this GPU handles the VM’s display output
• PCI-Express — enables PCIe bandwidth
• ROM-Bar — required for NVIDIA cards

Step 5: NVIDIA — Fix Error 43

Older consumer NVIDIA drivers detect virtualization and refuse to initialize. Hide the hypervisor by editing /etc/pve/qemu-server/[VMID].conf:

1
cpu: host,hidden=1
2
args: -cpu 'host,hv_vendor_id=NV43FIX,kvm=off'

hv_vendor_id can be any 12-character string. kvm=off hides the KVM signature from the driver.

For Windows VMs with NVIDIA GeForce Experience or other GPU software that crashes the VM, add:

echo "options kvm ignore_msrs=1 report_ignored_msrs=0" > /etc/modprobe.d/kvm.conf

Step 6: Tesla P40 for AI Workloads

The P40 is a datacenter card — 24GB GDDR5, no display output, pure compute. Around €100–150 used, it’s the best value-per-GB option for homelab AI workloads.

Key differences from gaming GPU passthrough:

• No display output — don’t check “Primary GPU”; the VM uses VNC for management
• No Error 43 workaround needed — datacenter drivers load cleanly in VMs
• Passive cooling — no fan, relies on chassis airflow; in a desktop case point active airflow at the card
• 250W TDP — full-length, full-height; verify your slot and PSU

Install CUDA drivers inside the VM (Ubuntu 22.04):

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update && sudo apt install -y cuda-toolkit-12-4 nvidia-driver-550-server

Reboot and verify:

nvidia-smi
# Tesla P40 | 24576MiB | ...

Troubleshooting

Black Screen After VM Boot

1. GPU not bound to vfio-pci — confirm Kernel driver in use: vfio-pci with lspci -nnk
2. Machine type isn’t q35
3. “Primary GPU” checked but vga: none not set (or vice versa)
4. Wrong PCIe slot — some boards only expose IOMMU correctly on specific slots

Error 43 Still Showing

If the fix in Step 5 doesn’t work, check that you’re not mixing cpu: and args: across multiple lines — Proxmox only supports one args entry. Also try changing the hv_vendor_id string.

GPU Hangs After VM Shutdown (Reset Bug)

Some NVIDIA cards can’t complete a PCIe reset — the GPU hangs and the host needs a full reboot. Fix with vendor-reset:

apt install pve-headers-$(uname -r) build-essential git
git clone https://github.com/gnif/vendor-reset
cd vendor-reset && make && make install
echo "vendor-reset" >> /etc/modules
update-initramfs -u -k all && reboot

Bad IOMMU Groups

Add to GRUB_CMDLINE_LINUX_DEFAULT in /etc/default/grub:

1
pcie_acs_override=downstream,multifunction

update-grub && reboot

Resources

• Proxmox PCI Passthrough — Official Wiki
• vendor-reset — gnif/GitHub