Proxmox GPU Passthrough to a VM¶

Full PCIe passthrough gives one VM direct control of a physical GPU. The Proxmox host and other VMs cannot use that device while it is assigned.

Requirements¶

• CPU and motherboard support for IOMMU interrupt remapping
• Intel VT-d or AMD-Vi enabled in BIOS/UEFI
• A GPU in a usable IOMMU group
• A second display or remote-management method when this is the host's only GPU

This guide uses the current Proxmox VE baseline. It does not add ACS overrides, unsafe interrupts, generic framebuffer flags or vendor-specific reset modules.

Find the GPU and Its Functions¶

lspci -nnk | grep -A3 -E "VGA|3D|Audio"

A GPU can have separate graphics, audio, USB and USB-C functions. Note the PCI addresses and the vendor/device IDs in brackets:

0b:00.0 VGA compatible controller [0300]: NVIDIA Corporation GPU [10de:1e81]
0b:00.1 Audio device [0403]: NVIDIA Corporation Audio [10de:10f8]

Do not copy these example addresses or IDs.

Load VFIO¶

Edit /etc/modules:

nano /etc/modules

Add:

vfio
vfio_iommu_type1
vfio_pci

Refresh the initramfs and reboot:

update-initramfs -u -k all
reboot

Recent Proxmox kernels enable IOMMU by default on AMD and on Intel with kernel 6.8 or newer after it is enabled in firmware. Older Intel kernels may require intel_iommu=on; follow the bootloader instructions for the installed Proxmox version instead of assuming the host uses GRUB.

Verify IOMMU¶

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

The exact output varies. It must show that IOMMU, Directed I/O or interrupt remapping is enabled.

Check Proxmox's PCI and IOMMU information. Replace NODE with the node name:

pvesh get /nodes/NODE/hardware/pci --pci-class-blacklist ""

It is normal for a GPU to share a group with its own functions or root port. Do not pass an unrelated host device only to work around a poor group.

Bind the GPU to VFIO When Needed¶

Proxmox VE tries to make an assigned PCI device unavailable to the host automatically. Normally, add the GPU to the VM in the next section first and return here only if assignment fails. After assigning the GPU and rebooting, check each function:

lspci -nnk -s 0b:00

The driver should be vfio-pci, or the Kernel driver in use line may be absent.

If automatic binding fails, add only the IDs verified on this host:

echo "options vfio-pci ids=10de:1e81,10de:10f8" > /etc/modprobe.d/vfio.conf
update-initramfs -u -k all
reboot

Replace both example IDs. If the host needs the same driver for another GPU, do not blacklist that driver globally.

Add the GPU to the VM¶

1. Shut down the VM.
2. Open Hardware > Add > PCI Device.
3. Select the GPU by its PCI address.
4. Use All Functions when the related GPU functions must be passed together.
5. Use PCI-Express with the q35 machine type.
6. Enable Primary GPU only when the guest should use the physical GPU as its main display.

Proxmox recommends q35, OVMF and PCIe for the best GPU-passthrough compatibility. OVMF requires a UEFI-capable GPU ROM; otherwise use SeaBIOS.

Install and Verify the Guest Driver¶

Install the driver supported by the guest operating system and GPU vendor. Avoid copying a fixed driver version from an old guide.

On Linux:

lspci -nnk | grep -A3 -E "VGA|3D|Audio"

For NVIDIA compute workloads, verify the installed NVIDIA driver with:

nvidia-smi

On Windows, check Device Manager and the GPU vendor's control or diagnostic tool.

Troubleshooting¶

dmesg -w
lspci -nnk
lspci -nnk -s 0b:00

Use the real PCI address from your host. If the device does not reset cleanly, check the GPU model, firmware and vendor documentation before installing any out-of-tree reset module. The vendor-reset project supports only the AMD device families listed by that project; it is not an NVIDIA RTX fix.

Sources¶

• Proxmox VE Administration Guide: PCI(e) Passthrough
• vendor-reset supported devices
• NVIDIA driver downloads

Comments