Fixing GPU issues after installing nvidia-driver packages

Reviewed on 13 December 2023 • Published on 01 December 2022

All Scaleway GPU Instances come with a Nvidia driver preinstalled. This means you do not have to install them by yourself.

The drivers installed on each Instance are from the nividia-headless-<driver version> packages. These are intended for use on Instances without a Windows Desktop or display server.

Important

The manual installation of a nvidia-driver package forces the installation of the xorg display server and other Gnome packages. As there is no physical display connected to the GPU of your Instance, this may cause side effects such as your Instance entering into sleep mode after a while. The user will get disconnected as a result.

If you have nvidia-driver-* packages installed on your GPU Instance, remove them from the Instance using the apt purge command and make sure to install and use nvidia-headless-* packages instead.

Tip

If you want to use ffmpeg in a Docker container on your GPU-Instance, make sure to install the libnvidia-encode library corresponding to the driver version installed on your Instance. For example: apt install -y libnvidia-encode-525 (with GPU OS12). You also need to specify the driver capabilities to launch your containers (By default graphics and video are not enabled):

docker run --rm --runtime=nvidia -e NVIDIA_DRIVER_CAPABILITIES=graphics,video