GPU Passthrough in Proxmox VE: IOMMU and VFIO Setup

Introduction

GPU passthrough (PCI passthrough) lets you assign a physical GPU directly to a VM, giving it near-native performance. This is essential for machine learning workloads, video transcoding, gaming VMs, and GPU-accelerated applications. Proxmox supports this via IOMMU/VT-d.

Requirements

CPU: Intel VT-d or AMD-Vi (IOMMU support in BIOS)
Motherboard: IOMMU enabled in BIOS
GPU: Secondary GPU (can't pass through the GPU used for Proxmox display)
Linux kernel 5.x+ (Proxmox 7+ uses kernel 5.15)

Step 1: Enable IOMMU in GRUB

BASH

# Edit GRUB config
nano /etc/default/grub

# For Intel CPU:
GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_iommu=on iommu=pt"

# For AMD CPU:
GRUB_CMDLINE_LINUX_DEFAULT="quiet amd_iommu=on iommu=pt"

# Update GRUB
update-grub

# Reboot
reboot

Step 2: Verify IOMMU is Working

BASH

# Should show IOMMU enabled
dmesg | grep -e IOMMU -e iommu

# List IOMMU groups (GPU must be in its own group)
for d in /sys/kernel/iommu_groups/*/devices/*; do
    n=${d#*/iommu_groups/*}; n=${n%%/*}
    printf 'IOMMU Group %s ' "$n"
    lspci -nns "${d##*/}"
done

If your GPU shares an IOMMU group with other devices, you'll need the ACS Override Patch (or a different slot/motherboard).

Step 3: Load VFIO Modules

BASH

# Add modules
echo "vfio
vfio_iommu_type1
vfio_pci
vfio_virqfd" >> /etc/modules

# Blacklist GPU drivers (so host doesn't grab the GPU)
echo "blacklist nouveau
blacklist nvidia
blacklist radeon
blacklist amdgpu" >> /etc/modprobe.d/blacklist.conf

# Apply
update-initramfs -u -k all
reboot

Step 4: Find GPU PCI IDs

BASH

# Find your GPU
lspci -nn | grep -E "VGA|Audio" | grep -i "nvidia|amd|radeon"
# Example output:
# 01:00.0 VGA compatible controller [0300]: NVIDIA Corporation RTX 3090 [10de:2204]
# 01:00.1 Audio device [0403]: NVIDIA Corporation GA102 HD Audio [10de:1aef]

Note the IDs: 10de:2204 and 10de:1aef

Step 5: Bind GPU to VFIO

BASH

# Create VFIO conf
echo "options vfio-pci ids=10de:2204,10de:1aef" > /etc/modprobe.d/vfio.conf

# Update initramfs
update-initramfs -u
reboot

Step 6: Verify GPU is Bound to VFIO

BASH

lspci -k | grep -A3 "01:00"
# Should show:
# Kernel driver in use: vfio-pci
# NOT: nvidia or nouveau

Step 7: Configure VM for GPU Passthrough

In Proxmox UI → VM → Hardware → Add → PCI Device:

Select your GPU (01:00.0)
Check: All Functions (includes audio)
Check: ROM-Bar
Check: PCI-Express
Primary GPU: Yes (if no other display adapter)

Or via CLI:

BASH

# Add to VM config
echo "hostpci0: 01:00,pcie=1,rombar=1,x-vga=1" >> /etc/pve/qemu-server/VM_ID.conf

# Complete VM config example
nano /etc/pve/qemu-server/200.conf

INI

name: gpu-vm
memory: 32768
sockets: 1
cores: 16
cpu: host
machine: q35
bios: ovmf        # Required for UEFI + GPU
efidisk0: local-zfs:vm-200-disk-0,efitype=4m
scsi0: local-zfs:vm-200-disk-1,size=200G
net0: virtio=AA:BB:CC:DD:EE:FF,bridge=vmbr0
hostpci0: 01:00,pcie=1,rombar=1,x-vga=1  # GPU
args: -cpu host,hidden=1,hv_vendor_id=NvidiaHack  # Hide VM from NVIDIA driver

NVIDIA Driver Notes

NVIDIA drivers detect if they're running in a VM (error 43 in Windows). To fix:

BASH

# Add to VM config
args: -cpu host,hidden=1,hv_vendor_id=NvidiaFTW

Verifying in the VM

After booting the Windows/Linux VM:

BASH

# Linux: verify GPU is visible
nvidia-smi  # Should show GPU name and memory

# Run GPU benchmark
glxgears
nvidia-smi dmon -s u  # Monitor GPU utilization

Troubleshooting

BASH

# GPU not appearing in VM
dmesg | grep vfio  # Check VFIO bound properly
dmesg | grep "IOMMU group"

# NVIDIA error 43 in Windows
# Add hv_vendor_id to args in VM config

# VM won't start with GPU
# Check IOMMU group — no other devices should share the group
# or use ACS override

# Performance not native
# Ensure CPU type is "host" not "kvm64"
# Enable hugepages for the VM