are there limitations on using nvidia-docker

Can i install nvidia docker to pull some prebuilt containers from the nvidia gpu cloud? Can i install nvidia-docker as described here: https://github.com/nvidia/nvidia-docker/wiki/Installation-(version-2.0)?

Problem is that I cannot figure out which nvidia driver i am using or which version of CUDA.

$nvidia-smi
bash: nvidia-smi: command not found

$nvcc -V
gives nothing

$cat /proc/driver/nvidia/version
bash: nvidia-smi: command not found

I would say no on using the prebuilt containers. These are for x86_64 images. You can tell by the first item in the “Prerequisites” section. The Nano uses Arm.

there are some containers prebuilt for arm64. Docker hub allows filtering by architecture. And I CAN run THOSE containers - for example

docker run -it openhorizon/aarch64-tx2-nvidia-caffe:jetpack3.3 bash

however even tho this runs on the nano and gives me a shell, the command nvidia-smi or nvcc do not work for me.

Hi,

nvidia-docker and nvidia-smi doesn’t support Jetson platform.

But you still can use the official docker on Nano:
https://github.com/Technica-Corporation/Tegra-Docker

Thanks.