What are the correct docker base images to use?

Robotics & Edge Computing Jetson & Embedded Systems Jetson Orin NX
1

We are coming from TX2 / TX2NX where it was pretty clear that we had to use a docker image corresponding to the underlying l4t version and the cuda/tensorrt libs are mounted from the host into the container.

Now we are migrating our applications to Orin NX and the available information is rather confusing:

  1. The latest available l4t image is r36.2.0 from July 2023 (corresponding ot jetpack 6.0 DP) without an automatic mounting of libraries from the host system.
  2. The issue No l4t-base:r36.3.0 docker image mentions that one should use ubuntu | NVIDIA NGC) for jetpack 6.0 (last update April 24).
  3. In the same issue, it is mentioned that users should try l4t-cuda. This is here NVIDIA L4T CUDA | NVIDIA NGC and it seems that these images are the most active (latest version August 2024). Now I see that there is a l4t-cuda:12.6.11-runtime, but the -devel image is missing.

What is the nvidia recommendation for docker base images suited for different versions of jetpack? Do we have to choose l4t-cuda with a suitable cuda version? Is the l4t-cuda:12.6.11-devel missing by intention? How are we supposed to include development support in the container?

3

Hi,

If you need a CUDA library, you can use l4t-cuda since it has CUDA preinstalled already.
If you need the dev packages, please install them with the apt command directly.

Here are some examples that build the container for Jetson for your reference:

Thanks.

4

Switching to l4t-cuda images seems like the safest bet especially when working with projects like llamacpp needing CUDA support. But yeah the missing -devel tag is confusing for development tasks.

5

Hi,

Please find the devel image below:

We will check with our internal team to see why the devel image is separated from the l4t-cuda.

Thanks.

7

Hej,

quick question, what Docker base image to choose for a GStreamer Pipeline running on the Jetson Orin NX 16GB inside a docker container using hardware acceleration. Anything to consider when running the container?

8

Hi,

From JetPack 6.1, the base image is l4t-cuda.

Thanks.

10

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