I can run an image in Docker of Ubuntu 18.04, but jetson-containers suggests I have to match release versions in order for CUDA etc to work. Correct me if I am wrong but the release ties in tightly with the operating system.
Is there a way to run an Ubuntu 18.04 image - from jetson-containers or vanilla 18.04 image, where cuda is enabled and the right version to work with the Xavier NX on 5.x and Ubuntu 20.04?
I feel there should be a way to make it work in Docker once I let the container access the camera ports.
Hi @brianlmerritt, it would appear that this user was able to compile librealsense 2.50 from source for JetPack 5.0:
The jetson containers that were released prior to JetPack 5.0 / L4T R34.1 would not work on JetPack 5.0, because those previous containers (the r32.x ones) relied on mounting in CUDA/cuDNN/ect from the device, whereas the JetPack 5.0 containers have CUDA/ect installed directly into the container.
There are still some device files and drivers that get mounted into the JetPack 5.0 containers by the NVIDIA Container Runtime when --runtime nvidia is used (you can see these mounted devices/files in /etc/nvidia-container-runtime/host-files-for-container.d/l4t.csv) These are meant to be used against l4t-base:r34.x or a container derived from l4t-base:r34.x - which use 20.04. Hence it is untested if you can get GPU working from a totally different base like ubuntu:18.04