Authors: fire1ce | Created: 2021-08-27 | Last update: 2026-03-06

Proxmox GPU Passthrough to VM¶

Introduction¶

GPU passthrough is a technology that allows the Linux kernel to present the internal PCI GPU directly to the virtual machine. The device behaves as if it were powered directly by the virtual machine, and the virtual machine detects the PCI device as if it were physically connected. We will cover how to enable GPU passthrough to a virtual machine in Proxmox VE.

Your mileage may vary depending on your hardware.

Prerequisites

Before starting, make sure IOMMU / VT-d (Intel) or AMD-Vi (AMD) is enabled in your BIOS/UEFI.

Proxmox Configuration for GPU Passthrough¶

The following examples use an SSH connection to the Proxmox server.

Find the GPU PCI Address¶

lspci -nnv | grep VGA

Output will look similar to:

0b:00.0 VGA compatible controller [0300]: NVIDIA Corporation TU104 [GeForce RTX 2080 SUPER] [10de:1e81] (rev a1)

The PCI address is 0b:00.0. List all devices in the same PCI group:

lspci -s 0b:00

A multi-function GPU typically includes a GPU, Audio, USB and UCSI controller:

0b:00.0 VGA compatible controller: NVIDIA Corporation TU104 [GeForce RTX 2080 SUPER] (rev a1)
0b:00.1 Audio device: NVIDIA Corporation TU104 HD Audio Controller (rev a1)
0b:00.2 USB controller: NVIDIA Corporation TU104 USB 3.1 Host Controller (rev a1)
0b:00.3 Serial bus controller: NVIDIA Corporation TU104 USB Type-C UCSI Controller (rev a1)

Get the vendor:device IDs for the VFIO configuration:

lspci -s 0b:00 -n

0b:00.0 0300: 10de:1e81 (rev a1)
0b:00.1 0403: 10de:10f8 (rev a1)
0b:00.2 0c03: 10de:1ad8 (rev a1)
0b:00.3 0c80: 10de:1ad9 (rev a1)

Note the IDs — you'll need them in the next steps:

10de:1e81,10de:10f8,10de:1ad8,10de:1ad9

Configure GRUB¶

Edit the GRUB configuration file:

nano /etc/default/grub

Find the GRUB_CMDLINE_LINUX_DEFAULT line and update it:

For Intel CPUFor AMD CPU

GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_iommu=on iommu=pt video=efifb:off video=vesa:off kvm.ignore_msrs=1 modprobe.blacklist=radeon,nouveau,nvidia,nvidiafb,nvidia-gpu"

GRUB_CMDLINE_LINUX_DEFAULT="quiet amd_iommu=on iommu=pt video=efifb:off video=vesa:off kvm.ignore_msrs=1 modprobe.blacklist=radeon,nouveau,nvidia,nvidiafb,nvidia-gpu"

What each flag does

iommu=pt — passthrough mode, improves performance
video=efifb:off video=vesa:off — prevents the host from using the GPU framebuffer
kvm.ignore_msrs=1 — suppresses NVIDIA MSR access errors in KVM
modprobe.blacklist=... — prevents host GPU drivers from loading

Save and update GRUB:

update-grub

Configure VFIO¶

Create the VFIO configuration file. Replace the IDs with the ones from your GPU:

echo "options vfio-pci ids=10de:1e81,10de:10f8,10de:1ad8,10de:1ad9 disable_vga=1" > /etc/modprobe.d/vfio.conf

disable_vga=1 prevents the host from trying to use the GPU as a VGA device.

Add the required VFIO kernel modules to /etc/modules:

nano /etc/modules

# Modules required for PCI passthrough
vfio
vfio_iommu_type1
vfio_pci
vfio_virqfd

Install vendor-reset (NVIDIA RTX Cards)¶

NVIDIA RTX (Turing/Ampere/Ada) GPUs have a hardware reset bug that prevents proper VM passthrough without a software fix. Install the vendor-reset kernel module:

apt install -y dkms git build-essential pve-headers-$(uname -r)
git clone https://github.com/gnif/vendor-reset.git /tmp/vendor-reset
cd /tmp/vendor-reset && dkms install .
echo "vendor_reset" >> /etc/modules

Apply Changes¶

update-initramfs -u -k all

Reboot Proxmox to apply the changes.

Verify IOMMU is Enabled¶

dmesg | grep -e DMAR -e IOMMU -e AMD-Vi

Intel — expected outputAMD — expected output

[0.067203] DMAR: IOMMU enabled

[2.311473] pci 0000:00:00.2: AMD-Vi: IOMMU performance counters supported
[2.318109] perf/amd_iommu: Detected AMD IOMMU #0 (2 banks, 4 counters/bank)

Verify VFIO Binding¶

Confirm all GPU functions are bound to vfio-pci:

lspci -k | grep -A3 "0b:00"

You should see Kernel driver in use: vfio-pci for each function.

Verify IOMMU Groups¶

Check that the GPU is isolated in its own IOMMU group:

for g in $(find /sys/kernel/iommu_groups/* -maxdepth 0 -type d | sort -V); do
  echo "IOMMU Group ${g##*/}:"
  for d in $g/devices/*; do
    echo -e "\t$(lspci -nns ${d##*/})"
  done
done

The GPU and all its functions should appear together in a single group, with no other unrelated devices sharing that group.

Now your Proxmox host is ready for GPU passthrough!

Windows Virtual Machine GPU Passthrough Configuration¶

For best results use the Windows 10/11 Virtual Machine configuration for Proxmox.

Limitations & Workarounds

In order for the GPU to function properly in the VM, you must disable Proxmox's Virtual Display — set it to none.
You will lose the ability to connect to the VM via Proxmox's Console.
Display must be connected to the physical output of the GPU for Windows to initialize the GPU properly.
You can use a HDMI Dummy Plug as a workaround.
Make sure you have an alternative way to connect to the VM, for example via Remote Desktop (RDP).

In the Proxmox web UI, navigate to the VM's Hardware tab → Add → PCI Device.

Select the GPU by its PCI address (0000:0b:00.0) and enable:

✅ All Functions
✅ ROM-Bar
✅ Primary GPU
✅ PCI-Express

Power on the VM, connect via RDP, and install the latest NVIDIA driver from the NVIDIA website.

Verify with GPU-Z or Device Manager.

Linux VM GPU Passthrough Configuration (Headless)¶

This section covers a headless Ubuntu Server setup for GPU compute workloads (AI, CUDA, etc.).

VM Configuration in Proxmox¶

Machine type: q35
BIOS: SeaBIOS (default)
CPU type: host
VGA: serial0 (no display needed)

In the Proxmox web UI, navigate to the VM's Hardware tab → Add → PCI Device.

Select the GPU (0000:0b:00.0) and enable:

✅ All Functions
✅ PCI-Express
❌ ROM-Bar — must be disabled for headless Linux VMs
❌ Primary GPU

ROM-Bar must be disabled for headless Linux VMs

With SeaBIOS, enabling ROM-Bar causes the firmware to execute the GPU's VBIOS during POST. On NVIDIA RTX cards this hangs the VM at boot. Disabling ROM-Bar skips GPU firmware initialization — safe for compute-only use where you don't need display output.

Boot the VM¶

Boot the VM and verify the GPU is visible:

lspci | grep -i nvidia

Expected output:

01:00.0 VGA compatible controller: NVIDIA Corporation TU104 [GeForce RTX 2080 SUPER] (rev a1)
01:00.1 Audio device: NVIDIA Corporation TU104 HD Audio Controller (rev a1)
01:00.2 USB controller: NVIDIA Corporation TU104 USB 3.1 Host Controller (rev a1)
01:00.3 Serial bus controller: NVIDIA Corporation TU104 USB Type-C UCSI Controller (rev a1)

Install NVIDIA Drivers¶

Check which driver versions are available:

apt search nvidia-driver | grep "^nvidia-driver-[0-9]"

Install the latest production driver (headless — no GUI packages):

apt install --no-install-recommends -y nvidia-driver-570 nvidia-utils-570

Reboot the VM:

reboot

Verify driver initialization:

nvidia-smi

Expected output:

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.x.xx    Driver Version: 570.x.xx    CUDA Version: 12.x     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
|   0  NVIDIA GeForce RTX 2080 SUPER  Off |   00000000:01:00.0 Off |                  N/A |
+-----------------------------------------------------------------------------------------+

Install NVIDIA Container Toolkit (Docker GPU Support)¶

To use the GPU inside Docker containers:

curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | \
  gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg

curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | \
  sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' \
  > /etc/apt/sources.list.d/nvidia-container-toolkit.list

apt update && apt install -y nvidia-container-toolkit
nvidia-ctk runtime configure --runtime=docker
systemctl restart docker

Test GPU access inside a container:

docker run --rm --gpus all nvidia/cuda:12.0-base-ubuntu22.04 nvidia-smi

Debug¶

Show hardware initialization messages and errors:

dmesg -w

Display all PCI devices:

lspci

Show kernel driver in use for each PCI device:

lspci -k

Check VFIO binding for a specific device:

lspci -k -s 0b:00

Display IOMMU groups:

for g in $(find /sys/kernel/iommu_groups/* -maxdepth 0 -type d | sort -V); do
  echo "IOMMU Group ${g##*/}:"
  for d in $g/devices/*; do
    echo -e "\t$(lspci -nns ${d##*/})"
  done
done

Support us at

Proxmox GPU Passthrough to VM¶

Introduction¶

Proxmox Configuration for GPU Passthrough¶

Find the GPU PCI Address¶

Configure GRUB¶

Configure VFIO¶

Install vendor-reset (NVIDIA RTX Cards)¶

Apply Changes¶

Verify IOMMU is Enabled¶

Verify VFIO Binding¶

Verify IOMMU Groups¶

Windows Virtual Machine GPU Passthrough Configuration¶

Linux VM GPU Passthrough Configuration (Headless)¶

VM Configuration in Proxmox¶

Boot the VM¶

Install NVIDIA Drivers¶

Install NVIDIA Container Toolkit (Docker GPU Support)¶

Debug¶

Comments