Docker container Jetpack 7.x?

Robotics & Edge Computing Jetson Systems Jetson Thor
,
1

Hello,

I wanted to ask if the NVIDIA L4T JetPack container repo is still being maintained, i.e. if a docker image for Jetpack 7.x will be provided here?

Also, is the Nvdia Pytorch for Jetson site still being maintained? Will there be an upload for Jetpack 7.x and Python 3.12?

Thanks in advance!

4

l4t jetpack container was last published 11/7/24. I’ve seen other posts from Nvidia that stated it is not being updated.

pytorch.org is currently building torch with cuda enabled. Most recent releases can be installed on Jetpack 7.2 with:

pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu132

Torch 2.12.0 and 2.13.0 are wheels are published with cuda. For a cuda version earlier than cuda132 browse to https://download.pytorch.org/whl/torch/ and then find / ctrl-f cu13 and pick wheel with aarch64, your cuda and python versions.

5

Hi,

We don’t have JetPack 7.x container.
But for PyTorch, you can use the NGC container directly:

Thanks.

6

So does that mean there will not be realsed any more l4t-jetpack or l4t-base containers for 7.x in the future? Because in this thread it was said that there will be a l4t-jetpack container for 7.1

7

Hi,

We don’t have a plan for this.

Thanks

8

So is the conclusion that the intention expressed in this post: Docker Image for JetPack 7.0 - #9 by kayccc

Targeting to have it (Docker Image) at JP7.1 release.

has been shelved and the recommended way to leverage a containerized L4T38/39 (Ubuntu 24.04-based) with CUDA and JetPack 7 on Jetson Orin (NX) is to use nvcr.io/nvidia/cuda:13.0.0-devel-ubuntu24.04 as mentioned on CUDA Setup — Jetson Orin Nano Developer Kit - User Guide ?

The container tag in the User Guide is from December of 2025: should now one use the more recent tags like nvcr.io/nvidia/cuda:13.3.1-cudnn-runtime-ubuntu24.04 / nvcr.io/nvidia/cuda:13.3.1-tensorrt-devel-ubuntu24.04 instead?

Is there Jetson documentation listing the most recent NVCR tags compatible/leveraging the specific hardware of each board (Orin Nano/NX/AGX, Thor Thor T2/3/4/5000) similarly to what shown here for DeepStream? Installation — DeepStream documentation

Is DeepStream 9.1 deepstream:9.1-triton-multiarch the “go-to”/most comprehensive base image for both desktop/amd64 and Jetson/arm64 containerized development now?

9

Hi,

Yes, the nvcr.io/nvidia/cuda:13.0.0-devel-ubuntu24.04 as base image.
The newer CUDA/TensorRT container can also work on Thor.

From JetPack 7.2, both Thor and Orin can use the SBSA driver, so the container on our NGC server with aarch64 support can work.

Thanks.