Prepare your server's GPU so Deployrr can use it
Deployrr does not install GPU drivers or toolkits. It scans your server and reports what it finds. This page gets the server ready so System > GPU shows your GPU as Ready. For the scan screen, the result lines and installing apps with the GPU, see GPU hardware transcoding.
The steps below are general Linux, NVIDIA and Proxmox steps. Package names can differ on your system.
Have console access to the server before you start. Some steps need a reboot, and a mistake in boot settings can stop the server from starting.
What Ready means
Section titled “What Ready means”The scan reports a GPU as ready when these checks pass on the machine where Deployrr runs. In a VM or LXC, that is inside the VM or container, not on the Proxmox host.
| GPU | The scan needs |
|---|---|
| Any | The GPU shows in lspci (install pciutils if the command is missing). |
| Intel or AMD | A renderD device exists in /dev/dri. |
| NVIDIA | The nvidia-smi command is installed, and Docker lists an nvidia runtime. The scan does not run it, so check that nvidia-smi works yourself (part 2). |
Pick the part that matches your setup. Bare metal and VM share parts 1 and 2. A Proxmox VM needs part 4 first. A Proxmox LXC uses part 3.
1. Intel or AMD on bare metal or in a VM
Section titled “1. Intel or AMD on bare metal or in a VM”The kernel drivers (i915 or xe for Intel, amdgpu for AMD) usually come with your Linux distribution. Check for the device:
ls -l /dev/driYou want a name that starts with renderD, such as renderD128.
If the folder is missing or has no renderD device, install the video acceleration package, reboot, and check again. On Debian or Ubuntu that is sudo apt install intel-media-va-driver for Intel or sudo apt install mesa-va-drivers for AMD. If it is still missing, check your distribution’s documentation for your GPU model.
2. NVIDIA on bare metal or in a VM
Section titled “2. NVIDIA on bare metal or in a VM”You need two things: the NVIDIA driver, and the NVIDIA Container Toolkit so Docker can use it.
-
Install the driver. On Ubuntu:
Terminal window sudo ubuntu-drivers installOn Debian, install the
nvidia-driverpackage from the non-free repository. See the Debian wiki for your release. -
Reboot the server.
-
Check that the driver works:
Terminal window nvidia-smiYou see a table with your GPU. If you get an error instead, see If nvidia-smi shows an error.
-
Install the NVIDIA Container Toolkit. Follow NVIDIA’s install guide for your distribution. It is the same link Deployrr prints under “NVIDIA Container Toolkit: Not installed or not configured”.
-
Tell Docker to use the toolkit:
Terminal window sudo nvidia-ctk runtime configure --runtime=docker -
Restart Docker:
Terminal window sudo systemctl restart docker -
Check that Docker lists the runtime:
Terminal window sudo docker info | grep -i runtimesThe line must include
nvidia.
3. Proxmox LXC
Section titled “3. Proxmox LXC”An LXC container shares the Proxmox host’s kernel. The GPU driver lives on the Proxmox host, and you pass the GPU device into the container. Deployrr cannot do this for you.
For background, read UDMS part 11: Passing the iGPU through to a Proxmox LXC container (manual setup; it edits the container config file instead of using the menu below). The short version for Intel and AMD:
-
On the Proxmox host, check that
/dev/driexists and has arenderDdevice. If not, do part 1 on the host. -
In the Proxmox web interface, select your container, then Resources > Add > Device Passthrough.
-
Set the device path to
/dev/dri/renderD128. Use your ownrenderDname if it differs. -
Set the GID to the group ID of the
rendergroup inside the container. To find it, run this inside the container:Terminal window getent group renderThe number after the second colon is the GID.
-
Restart the container.
-
Inside the container, run
ls -l /dev/driand check therenderDdevice is there.
On older Proxmox versions without this menu, add the device lines to the container’s config file instead, as the UDMS guide and Proxmox iGPU passthrough at 2:32 show. Back up the file first, the same way as in Run Deployrr in a Proxmox LXC.
In an unprivileged container, the device can show as owned by nobody or nogroup. That is fine if the permissions let everyone read and write it (crw-rw-rw-). If they don’t, and the app can’t use the GPU, check the GID you set.
NVIDIA in an LXC needs the driver on the Proxmox host and the NVIDIA Container Toolkit inside the container. For Jellyfin, Plex and Emby, an Intel or AMD GPU is the safer choice for now.
If Docker stops starting containers inside the LXC, see Docker fails with “failed to create shim task”.
4. Proxmox VM
Section titled “4. Proxmox VM”Passing a GPU to a VM hands the whole GPU to that VM, so the host and other guests can’t use it. For an integrated GPU (iGPU), an LXC is usually easier.
- Turn on virtualization and IOMMU (often called VT-d or AMD-Vi) in the server’s BIOS.
- On an Intel CPU, add
intel_iommu=onto the kernel boot line of the Proxmox host. AMD CPUs usually have it on already. Proxmox’s PCI passthrough documentation shows where this setting lives for your boot loader. - Reboot the Proxmox host.
- In the Proxmox web interface, select the VM, then Hardware > Add > PCI Device.
- Pick your GPU and save. For a modern GPU, tick PCI-Express if the option shows.
- Start the VM.
- Inside the VM, do part 1 (Intel or AMD) or part 2 (NVIDIA).
Set System Type to Virtual Machine in Prerequisites. See Run Deployrr in a Proxmox LXC.
If nvidia-smi shows an error
Section titled “If nvidia-smi shows an error”Run these checks on the machine where the driver is installed: the server or VM, or the Proxmox host for an LXC. Start with what the error says.
| Error | Most likely cause |
|---|---|
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver |
The driver’s kernel module isn’t loaded. Secure Boot blocked it, or a kernel update left it unbuilt. |
Failed to initialize NVML: Driver/library version mismatch |
The driver was updated but the old one is still loaded. Reboot. In an LXC, the driver inside the container doesn’t match the one on the Proxmox host. |
nvidia-smi: command not found |
The driver isn’t installed, or on Debian the nvidia-smi package is missing. |
Check that the module is loaded
Section titled “Check that the module is loaded”lsmod | grep nvidiaNo output means the module isn’t loaded. Try to load it and read the error:
sudo modprobe nvidiaSecure Boot
Section titled “Secure Boot”Check whether Secure Boot is on:
mokutil --sb-stateIf it says SecureBoot enabled and modprobe printed Key was rejected by service, the kernel refused the driver because it isn’t signed with a key your system trusts. You have two ways out:
-
Enroll the signing key. A driver built on your server (DKMS) is signed with a local key that you enroll once. Import it and set a one-time password:
Terminal window sudo mokutil --import /var/lib/dkms/mok.pubReboot. A blue MOK Manager screen opens: choose Enroll MOK and type that password. On Ubuntu, the driver install may ask for the password itself, and
sudo ubuntu-drivers installusually installs a module that is already signed, so this comes up less often there. The Debian wiki’s Secure Boot page has the details for Debian. -
Turn off Secure Boot in the BIOS or UEFI settings.
You need a screen and keyboard or the server’s remote console for both. In a VM, Secure Boot is a setting of the VM, not the host.
A kernel update broke the driver
Section titled “A kernel update broke the driver”The driver’s kernel module must match the running kernel. When the driver is built on your server (DKMS, as Debian’s nvidia-driver package does), it is rebuilt for each new kernel. If that build failed, the new kernel has no NVIDIA module. Check:
uname -rdkms statusIf dkms status doesn’t list nvidia as installed for the kernel uname -r shows, install the headers for that kernel and rebuild:
sudo apt install linux-headers-$(uname -r)sudo dkms autoinstallOn a Proxmox host, the headers package is proxmox-headers-$(uname -r) (pve-headers-$(uname -r) on older releases) instead of linux-headers-$(uname -r).
If dkms status lists nothing, the driver came prebuilt (common on Ubuntu) or from NVIDIA’s own installer. Run sudo ubuntu-drivers install or NVIDIA’s installer again so it installs the module for the new kernel.
Reboot and run nvidia-smi again. If it still fails, pick the previous kernel in the boot menu to get the GPU back while you sort it out.
Driver versions on Proxmox and in an LXC
Section titled “Driver versions on Proxmox and in an LXC”An LXC uses the Proxmox host’s kernel module, and the driver files inside the container must be the same version. After you update the driver on the host, install the same version inside the container. On the host, show the loaded version:
cat /proc/driver/nvidia/versionDon’t run that inside the container: it shows the host’s module there too. Inside the container, list the installed driver files instead:
ls /usr/lib/x86_64-linux-gnu/libnvidia-ml.so.*The number at the end, such as 550.120, must match the host’s version. If the host itself shows the error, fix the host first with the checks above.
Apps say the driver is too old
Section titled “Apps say the driver is too old”If an app container won’t start and the error mentions unsatisfied condition: cuda>=, the app’s image needs a newer NVIDIA driver than the one installed. Update the driver on the server (or the Proxmox host), reboot, and start the app again.
Check it worked
Section titled “Check it worked”- Open System > GPU in Deployrr and run the scan again. It does not refresh on its own.
- You want “Intel Iris: Ready”, “AMD ROCm: Ready” or “NVIDIA Container Toolkit: Ready”. GPU hardware transcoding explains every line. To see the scan, watch Deployrr 6 overview: Intelligent GPU Detection at 5:37.
- Install the app with the GPU. To add the GPU to an app you installed earlier, reinstall it. See If the app is already installed.

