Enabling GPU Passthrough in Proxmox (Step-by-Step Guide + Verification)

Introduction

GPU passthrough allows a virtual machine (VM) to directly use a physical GPU. This is especially useful for:

AI / Machine Learning workloads
Video processing
CUDA-based applications
High-performance computing

In this guide, we will walk through:

How to enable GPU passthrough in Proxmox
How to configure the system
How to verify that the GPU is successfully enabled

System Overview

In our setup:

Host: Proxmox (Debian-based)
GPUs:
- RTX 3060 → assigned to VM
- GT 1030 → used by host
VM OS: Ubuntu

Step 1: Enable IOMMU (VT-d)

First, enable IOMMU in GRUB.

Edit:

vi /etc/default/grub

Update:

GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_iommu=on iommu=pt"

Apply changes:

update-grub

Step 2: Enable VFIO Modules

Edit:

vi /etc/modules

Add:

vfio
vfio_iommu_type1
vfio_pci
vfio_virqfd

Step 3: Blacklist GPU Drivers

Prevent the host from using the GPU.

vi /etc/modprobe.d/blacklist.conf

Add:

blacklist nouveau
blacklist nvidia
blacklist nvidiafb
blacklist rivafb

Step 4: Bind GPU to VFIO

Identify GPU IDs:

lspci -nn

Then configure:

vi /etc/modprobe.d/vfio.conf

Example:

options vfio-pci ids=10de:2504,10de:228e

Step 5: Update Initramfs

update-initramfs -u -k all

Step 6: Reboot

reboot

Step 7: Verify GPU Binding

Run:

lspci -nnk | grep -i nvidia -A 3

Expected output:

Kernel driver in use: vfio-pci

This confirms the GPU is detached from host and ready for passthrough.

Step 8: Attach GPU to VM

In Proxmox UI:

Go to VM → Hardware
Click “Add → PCI Device”
Select GPU (e.g., 01:00.0)

Enable:

All Functions
PCI-Express
ROM-Bar

Step 9: VM Configuration

Update VM settings:

qm set <VMID> -machine q35
qm set <VMID> -cpu host
qm set <VMID> -hostpci0 0000:01:00,pcie=1

(Optional fix for NVIDIA):

args: -cpu host,kvm=off

Step 10: Start VM

qm start <VMID>

Step 11: Verify Inside VM

Check GPU detection:

lspci | grep -i nvidia

Install driver:

apt update
apt install nvidia-driver -y
reboot

Final Verification

Run:

nvidia-smi

Expected Output:

GPU name (RTX 3060)
Driver version
CUDA version
Memory usage

Example:

NVIDIA-SMI 535.xx
GPU: RTX 3060
Memory: 12GB

This confirms GPU passthrough is fully working.

Result

You now have:

GPU successfully passed to VM
CUDA enabled
Ready for AI / ML / compute workloads

Key Takeaways

IOMMU enables hardware-level isolation
VFIO binds GPU for passthrough
Proxmox assigns GPU to VM
nvidia-smi confirms success

Conclusion

GPU passthrough in Proxmox unlocks powerful capabilities such as AI model execution, high-speed rendering, and GPU-accelerated workloads — all within a virtualized environment.

With proper configuration and validation, you can achieve near bare-metal GPU performance inside your VM.

Monitor GPU

Monitor GPU usage in real time:

watch -n 1 nvidia-smi

Your GPU is now fully enabled and production-ready