Upgrading nvidia-container-toolkit to 1.16 wants to remove nvidia-jetpack

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Xavier
1

I have an AGX Xavier running Jetpack 5.1.2 and I’ve just upgraded CUDA to 12.2.2 following the instructions at https://developer.nvidia.com/blog/simplifying-cuda-upgrades-for-nvidia-jetson-users/

That all seemed to okay but I now need an updated nvidia-container-toolkit to support the later CUDA version. The nvidia-container-toolkit version currently installed is 1.11.0~rc.1-1 and I’d like 1.16.1-1, which is a candidate according to apt-cache policy.

However, if I run sudo apt install nvidia-container-toolkit I see:

The following additional packages will be installed:
  libnvidia-container-tools libnvidia-container1 nvidia-container-toolkit-base
The following packages will be REMOVED:
  nvidia-container nvidia-jetpack nvidia-jetpack-dev nvidia-jetpack-runtime
The following NEW packages will be installed:
  nvidia-container-toolkit-base
The following packages will be upgraded:
  libnvidia-container-tools libnvidia-container1 nvidia-container-toolkit
3 upgraded, 1 newly installed, 4 to remove and 2 not upgraded.

nvidia-jetpack seems to depend on most of the nvidia software, so it seems a bad idea to uninstall that.

Have I missed something?

3

Hi,

To run newer CUDA library on previous driver is supported by the compat feature.
So you don’t need to upgrade nvidia-container-toolkit.

Do you meet any issue when using 1.11.0~rc.1-1?

Thanks.

4

Hi AastaLLL,

I’m trying to run an Ubuntu 22.04 docker image on my Jetpack 5.1.2 system. I create the image based on nvidia/cuda:12.2.2-devel-ubuntu22.04 (although I could use any cuda version after 11.7, which is the earliest I could find).

When I run the image with the command:

docker run -it --runtime nvidia --gpus all <image>

I get the error:

docker: Error response from daemon: failed to create task for container: 
failed to create shim task: OCI runtime create failed: 
nvidia-container-runtime did not terminate successfully: exit status 1: 
time="2024-08-27T12:01:33Z" level=error 
msg="failed to create NVIDIA Container Runtime: failed to construct OCI spec modifier: 
requirements not met: unsatisfied condition: cuda>=12.2 (cuda=11.4)"

I noticed that the release notes for NVIDIA Container Runtime mentioned updating the base cuda image which is why I thought that might be the solution.

Should I take a different approach?

Thanks, Dave.

5

A slight simpler version showing the error is:

$ docker run -it --runtime nvidia --gpus all nvidia/cuda:11.7.1-devel-ubuntu22.04
docker: Error response from daemon: failed to create task for container: 
failed to create shim task: OCI runtime create 
failed: nvidia-container-runtime did not terminate successfully: 
exit status 1: time="2024-08-27T23:57:51Z" level=error 
msg="failed to create NVIDIA Container Runtime: failed to construct OCI spec modifier: 
requirements not met: unsatisfied condition: cuda>=11.7 (cuda=11.4)"
: unknown.
6

Is it expected that the above command should work without error on an AGX Xavier running Jetpack 5.1.2?

7

I think I might have solved it. When I use the nvcr.io/nvidia/l4t-jetpack images it seems to work.

9

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.