docker in jetpack: are we also able to switch cuda/cudnn version?


when I bough jetson nano I was glad to discover it now supports nvidia-docker2 and read official page on it along with few examples…and then tried few things. Nvidia did a really great job with cuda/cudnn docker images and my question is: are we also able to switch cuda/cudnn version in containers like on regular GPUs, in other words: jetpack 4.2 supports CUDA10, but I also have my home made app dependent on CUDA 9, can I just run 2 separate docker containers and use 2 CUDA version at the same time? I know official answer for installing CUDA 9 on latest jetpack is that you can’t do it, but is it also impossible with dockers?

I can’t seem to find this answer anywhere, so any information is more than welcome!


Unfortunately, the answer is no.
That’s because the GPU driver is integrated in the L4T image so you cannot update the container with a compatible driver for CUDA 9.0.