Proxmox GPU, USB and PCI Passthrough: The Complete IOMMU / VFIO Guide

Hardware Passthrough: Giving Direct Access to Hardware

By default, Proxmox virtualizes hardware: each VM receives emulated devices (virtio network card, virtual GPU, virtual SCSI controller). This is sufficient for most workloads — but not all.

Hardware passthrough (or PCI passthrough) allows you to assign a physical device directly to a VM, bypassing emulation. The VM sees the hardware as if it were running bare-metal, with native drivers and native performance.

Use cases are numerous: GPU passthrough for gaming, AI/ML compute or VDI, USB passthrough for home automation dongles (Zigbee, Z-Wave) or license keys, PCI passthrough for high-performance network cards, RAID controllers or capture cards.

Hardware and BIOS Prerequisites

Passthrough does not work on just any hardware. Here is what you need to check before getting started:

Step-by-Step IOMMU Configuration

The first step, common to all types of passthrough (GPU, USB, PCI), is to enable IOMMU at the Linux kernel level on the Proxmox host.

Step 1: Bootloader Parameters

The iommu=pt (passthrough mode) parameter tells the kernel not to translate DMA addresses for host devices — which improves host network and disk performance while enabling IOMMU for VMs.

Step 2: Load VFIO Modules

The principle: you blacklist GPU drivers on the host so the GPU is not initialized at boot, then reserve it via VFIO so it is available exclusively for passthrough to VMs.

Step 3: Verify IOMMU

# Verify that IOMMU is active
dmesg | grep -e DMAR -e IOMMU
# Should display: "DMAR: IOMMU enabled" or "AMD-Vi: IOMMU enabled"

# List IOMMU groups and their devices
find /sys/kernel/iommu_groups/ -type l | sort -V

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

GPU Passthrough: NVIDIA and AMD

GPU passthrough is the most sought-after use case. It delivers native graphics performance inside a VM — gaming, 3D rendering, video transcoding, AI/ML compute (CUDA, ROCm).

VM Configuration for GPU Passthrough

# Create a VM with UEFI (OVMF) — required for GPU passthrough
qm create 100 --name gpu-vm --memory 16384 --cores 8 \
  --bios ovmf --efidisk0 local-lvm:1,format=raw,efitype=4m,pre-enrolled-keys=1 \
  --machine q35 --cpu host,hidden=1 \
  --net0 virtio,bridge=vmbr0

# Attach the GPU (replace 01:00 with the actual PCI address)
qm set 100 --hostpci0 01:00,pcie=1,rombar=1,x-vga=1

# For NVIDIA (avoid error code 43):
qm set 100 --args '-cpu host,kvm=off'

# Attach GPU + audio from the same IOMMU group:
# qm set 100 --hostpci0 01:00,pcie=1,rombar=1,x-vga=1

USB Passthrough

USB passthrough is simpler than GPU passthrough and supports hot-plug — you can attach and detach USB devices to a running VM.

Generic PCI Passthrough

Beyond GPU and USB, any PCI/PCIe device can be passed through to a VM. The most common use cases:

vGPU and SR-IOV: Sharing a GPU Between VMs

Standard passthrough assigns an entire GPU to a single VM. To share a GPU between multiple VMs, two technologies exist:

Troubleshooting: Common Issues

Limitations and Trade-offs

Our Approach at RDEM Systems

At RDEM Systems, we use passthrough selectively:

For a turnkey deployment, discover our dedicated Proxmox server plans with GPU passthrough : we configure IOMMU, VFIO and drivers for your specialized workloads.

Best Practices

Frequently Asked Questions

Official Documentation

To dive deeper into the concepts covered in this article:

Related Articles