JetPack and deepstream-l4t docker images compatibility

Till recently when we run a Jetson deepstream-l4t container with runtime="nvidia" a substantial amount of shared libraries and symbolic links present in the host are loaded in the container.

The full list can be found on /etc/nvidia-container-runtime/host-files-for-container.d/l4t.csv

This kinda ties a certain JetPack version to specific deepstream-l4t version(s), as there might exist incompatibilities caused by “breaking changes” in the shared libraries.

I think that changed since deepstream-l4t 6.1 as we can find the following information in the documentation:

Since Jetpack 5.0.1 DP, NVIDIA Container Runtime no longer mounts user level libraries like CUDA, cuDNN and TensorRT from the host. These will instead be installed inside the containers.

Is it safe to assume a JetPack 4.6 installation in the host will be able to handle the following containers without issues?

  • deepstream-l4t:6.0.1-triton
  • deepstream-l4t:6.1-triton
  • deepstream-l4t:6.* (future versions)

Or in order to run a deepstream-l4t:6.1-triton container is required update the JetPack to 5.0.1?

1 Like


Waiting some support from NVIDIA on this to figure out if is possible to choose some “stable” JetPack so in the future the host OS does not need to be updated when we install newer versions of deepstream-l4t docker image

Thanks in advance

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.

Sorry for the late.
I do not think it’s possible to have one stable JP version, just need to update the docker image. the bsp image includes low level libraries except user level libraries like CUDA, cuDNN and TensorRT. for the deepstream samples run success in the container, you need to add --runtime=nvidia which will mount some nvidia low level libraries into container.

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