This repository contains source code for an auxiliary snap that assists in assembling and loading NVIDIA drivers.
Installing this snap implies consent with NVIDIA license.
When installed together with a compatible kernel snap, an attempt will be made to assemble, load and setup NVIDIA graphics drivers on an Ubuntu Core system.
# snapd2.59 is required
$ snap refresh snapd
$ snap install nvidia-assemble --channel 22/stable
$ sudo ls -latr /dev/nvidia*
$ sudo lsmod | grep nvidia
Expectation is that /dev/nvidia devices are available on the system,
and that nvidia drivers are loaded into the kernel.
Install content provider with userspace libraries, install sample demo apps, and connect the content provider.
$ snap install nvidia-core22
$ snap install graphics-core22-samples
$ snap connect graphics-core22-samples:graphics-core22 nvidia-core22:graphics-core22
$ snap connect graphics-core22-samples:hardware-observe
$ snap connect graphics-core22-samples:kernel-module-observe
$ snap connect graphics-core22-samples:network-bind
$ snap connect graphics-core22-samples:opengl
After above setup is done one can query EGL card information, query CUDA device information, and execute a sample CUDA application:
$ graphics-core22-samples.eglinfo
$ graphics-core22-samples.deviceQuery
$ graphics-core22-samples.bandwidthTest
Once the drivers are available on the system, one can use nvidia-core22 graphics-core22 content interface provider in your snap to provide userspace libraries to access NVIDIA graphics card functionality. This content interface can be used if stock userspace libraries are sufficient for a given snap, or if one wants to share userspace libraries across multiple snaps. Alternatively, one can simply stage one's own userspace dependencies as required.
Samples of GL & CUDA CLI applications are available from graphics-core22-samples. This is an example snap that vendors existing GL binaries, and complies a CUDA binary from source. The snapcraft.yaml for this sample removes staged libraries from the snap, such that at runtime all libraries are used from the nvidia-core22 graphics-core22 content provider.
First is to check that kernel contains nvidia driver components:
$ sudo ls /lib/modules/*/kernel | grep nvidia
To troubleshoot further collect dmesg output and output from snap installations:
$ sudo journalctl -b -k
$ sudo journalctl -b -u snap.nvidia-assemble.service