How can I run GPU codes on tx2 docker with Jetpack 4.3 (Nvidia-container-runtime)?


I’m using jetson tx2 boards for an autonomous self-driving car (for research).
For self-driving, we are using TensorFlow and these codes are using GPUs for high performance.
Recently, we want to run these codes on the container(docker) for some reason.
So we installed the jetpack 4.3 version for using “Nvidia-container-runtime”.
However, the problem is that our container is using an arm64 container(using CUDA 11.0 version) and Jetpack supports CUDA 10~10.2 version, so it occurs errors.
In addition, the arm64 containers which support CUDA are only supporting the newest CUDA version like 11.x.

So, my question is
Can I run the GPU codes on the container (tx2 board) by using normal arm64-ubuntu docker images? (without these docker images (Docker Hub))

If it is not, how can I run my TensorFlow codes on a container? (I already installed Jetpack 4.3 version).

Thank you for taking your time.


You will need to use Jetson based docker for the hardware support.
Please check if the container listed in our NGC page can meet your requirement or not:

There are two image for JetPack 4.3: