Adding the GPU to the Kubernetes Nodes

Time to add the GPU! For this your Kubernetes node VM needs to be powered off first. Hot-plugging isn’t an option here. If you disabled the autostart option, it should still be powered off after the last host reboot.

Step 1: Configure GPU Passthrough in Proxmox

#

Add the GPU via a mediated device (MDev Type) to your Kubernetes node VM using either a resource mapping or as a raw device:

Important: Ensure that vIOMMU in the Machine configuration is set to Default (None).

Step 2: Disable Secure Boot

#

Two ways to do this.

Method 1: EFI Disk Replacement

#

Simply remove the current EFI disk and add a new one, ensuring Pre-Enroll keys is disabled:

Method 2: BIOS Configuration

#

Alternatively, disable Secure Boot through the VM’s BIOS:

1. Start the VM and quickly open the console
2. Press Esc repeatedly during boot to enter the BIOS
3. Navigate to Device Manager → Secure Boot and disable it (multiple confirmations required)
4. Save settings and reboot

Step 3: Install GPU Drivers

#

Your approach depends on your OS.

For Talos Systens

#

Add the official i915 system extension to your installation. Extension management is beyond this guide’s scope, but here’s what you need:

customization:
  systemExtensions:
    officialExtensions:
      - siderolabs/i915

For Minimal or Cloud-Based Installations

#

SSH into your node and install the GPU drivers, reboot afterwards:

sudo apt install linux-generic -y
sudo reboot

For Other Distributions

#

Most full distributions include these drivers by default. Verify that linux-generic is installed.

Step 4: Verify GPU Passthrough

#

Let’s confirm your GPU made it through.

Verification on Talos

#

talosctl -n <node-name> ls /dev/dri		# Use your actual node name

Expected output:

NODE          NAME
<node-name>   .
<node-name>   by-path
<node-name>   card0						# Your GPU!
<node-name>   renderD128				# The render device

Verification on Other Systems

#

Log into your node and execute:

ls /dev/dri

Should show:

by-path  card0  renderD128				# card0 is your GPU!

Step 5: Configure Additional Nodes

#

Rinse and repeat! Apply these steps to every Kubernetes node that needs GPU support.