Official docker Jetpack image? How are you creating docker images?

Is there a docker image that emulates the run-time of an installed Jetpack?

I’d like to port a few applications to the Xavier and TX2 platforms. Both support docker so there is a some attraction of putting these applications under a standardized container.

I understand that the newest Jetpack run-time is based off an Ubuntu 18.04 image. But is there anything more official? Are all the Jetson libraries and packages available for install through a Dockerfile?

Hi,

Currently, nvidia-docker doesn’t support Jetson platform yet. You will need some hack to enable the GPU resource inside the container.
Here is a good example for your reference:

Nvidia-docker is one of our future release feature. It’s recommended to pay attention to our announcement for the update.

Thanks

Right, I understand that and exposing the nodes internally inside the container is a good temporary workaround.

However, what I’m asking is like the nvidia/cuda images sitting on Docker Hub, is there any plans to have a complete JetPack run-time image for docker based applications running on the various Jetson platforms? If not, why not? Also, can I build my own now? i.e. Are all the packages for Jetson available on some public NVidia repository?

Hi,

YES. This is one of our future feature.
Please pay attention to our announcement.

Thanks.

Can you give me a time frame @AastaLLL before I go off and redo a lot of work that has already been done?

Hi,

Sorry that we cannot share our schedule here.
The nvidia-docker will support for Jetson in our next JetPack release.

Thanks.

Alright, but will support for nvidia-docker on Jetson also include Jetson docker base images as well to work off of?

Hi AastaLLL,

Now that nvidia-docker is supported on Jetpack 4.2.1, is this base image also supported? If so, where can we find it?

Thanks,

does nvidia-docker that runs at e.g. Xavier support x86 docker containers? like can ubuntu be running as if x86 ubuntu?

That’s not how docker works. Docker is kernel dependent. It offer run-times, not complete virtualization like a VM.

@AastaLLL any update? I think folks would just like to know, what’s the best way to develop for a Jetson board from an x86 development machine that’s running the closest possible Jetson environment?

The best I can do is pulling nvidia/cudagl for cuda 10.2, which is close to what Jetpack 4.4.1 uses, but unfortunately OpenCV still needs to be upgraded to 4.1.1. and probably other libraries I’m not aware of.

Ideally the L4T ARM-based images could be ported to x86 (i.e. avoiding the need for QEMU emulation) so that I can develop with x86 tools inside the x86 container (e.g. IDEs that don’t have ARM support yet). Is that going to be supported?

Hi er2020 ,

Please help to open a new topic for your issue. Thanks