XCP-ng Intel iGPU Passthrough

Virtualization

4 min readAdvancedLast updated Oct 29, 2026

Pass the CPU's built-in Intel graphics (iGPU) through to a VM on XCP-ng 8.2 / 8.3. The usual reason is hardware video transcoding (Intel Quick Sync) for Plex, Jellyfin, Frigate or similar in a Linux VM.

Before you start

In the BIOS/UEFI, enable VT-d and make sure the iGPU is enabled. If the server also has a discrete GPU, turn on the option that keeps the iGPU active (often called iGPU Multi-Monitor), and set the primary display to the discrete card if you want to keep a host console.
When the iGPU is hidden from the host, the host's local screen goes blank after boot. Make sure SSH or IPMI works before you start.
A VM with a passed-through device can't be live-migrated.
Use a recent Linux kernel in the VM (Ubuntu 24.04+, Debian 13, Fedora 40+). 12th-gen Intel and newer need newer i915 / xe drivers.

Find the iGPU

lspci | grep -i vga

00:02.0 VGA compatible controller: Intel Corporation AlderLake-S GT1 (rev 0c)

The Intel iGPU is almost always at 00:02.0. In full form that's 0000:00:02.0.

Hide the iGPU from the host

/opt/xensource/libexec/xen-cmdline --set-dom0 "xen-pciback.hide=(0000:00:02.0)"
reboot

After the reboot, confirm it's ready:

xl pci-assignable-list     # should list 0000:00:02.0

Attach it to the VM

Shut the VM down, then:

xe vm-list name-label="jellyfin" params=uuid
xe vm-param-set uuid=<vm-uuid> other-config:pci=0/0000:00:02.0

Keep the VM's normal virtual display. You'll still use the Xen Orchestra console to manage it, and the iGPU works alongside it as a second GPU.

Start the VM.

Set up the VM (Linux)

Check the iGPU is visible and the driver loaded:
```
lspci -nnk | grep -A3 -i vga
ls -l /dev/dri
```
You should see the Intel device with Kernel driver in use: i915 (or xe), and /dev/dri/renderD128 (sometimes renderD129).

Install the video acceleration drivers and test them:

# Debian / Ubuntu
sudo apt install intel-media-va-driver-non-free vainfo intel-gpu-tools
# Fedora (intel-media-driver comes from RPM Fusion)
sudo dnf install intel-media-driver libva-utils igt-gpu-tools

vainfo

vainfo should list H.264/HEVC profiles with VAEntrypointEncSlice.

Let your media server use the GPU:
- Native install: add the service user to the render and video groups: sudo usermod -aG render,video jellyfin
- Docker: pass the device through, e.g. in compose: devices: ["/dev/dri:/dev/dri"]
Turn on hardware transcoding in the app (Jellyfin: Dashboard → Playback → Transcoding → Intel QuickSync (QSV); Plex: Settings → Transcoder → Use hardware acceleration).
Play something that needs transcoding and watch the GPU work:
```
sudo intel_gpu_top
```

Removing passthrough

xe vm-param-remove uuid=<vm-uuid> param-name=other-config param-key=pci
/opt/xensource/libexec/xen-cmdline --delete-dom0 xen-pciback.hide
reboot

Troubleshooting

No /dev/dri in the VM – the kernel is too old for the GPU. Check sudo dmesg | grep -i -e i915 -e xe. On 12th-gen and newer with older kernels, try the kernel option i915.force_probe=<device-id> (the ID is in lspci -nn, e.g. 4680), or upgrade the kernel.
vainfo errors with "va_openDriver() returns -1" – the wrong VA driver is installed. Modern Intel iGPUs need the iHD driver (intel-media-driver), not the older i965.
VM won't start – check xl pci-assignable-list on the host and that VT-d is enabled (xl dmesg | grep -i vt-d).
The iGPU disappeared from lspci on the host – the BIOS disabled it because a discrete GPU is installed. Turn on the iGPU / multi-monitor option in the BIOS.

XCP-ng Intel iGPU Passthrough

Before you start

Find the iGPU

Hide the iGPU from the host

Attach it to the VM

Set up the VM (Linux)

Removing passthrough

Troubleshooting

Article information

Related articles