Docker: Error response from daemon: Unknown runtime specified nvidia

Robotics & Edge Computing Jetson Systems Jetson Nano
1

Just flashed NVIDIA Jetson Linux L4T 32.7.1 to my Jetson Nano, pulled docker with sudo docker pull nvcr.io/nvidia/l4t-ml:r32.7.1-py3.

Running docker
sudo docker run -it --rm --runtime nvidia --network host nvcr.io/nvidia/l4t-ml:r32.7.1-py3

leads to: docker: Error response from daemon: Unknown runtime specified nvidia.

Running docker without --runtime nvidia works, but seems to have no access to GPU (python3 ; import tensorflow leads to Could not load dynamic library ‘libcudart.so.10.2’)

Thanks for help

3

Hi,
It looks similar to:
Docker isn't working after apt upgrade - #7 by harrison-matt

Please take a look and give it a try.

May also try docker for deepstream:
Install nvidia container runtime with docker intergration in jetson nano emmc (16gb) - #7 by AastaLLL

4

Hello,
thanks a lot for your help. I tried it, and

  • the first hint did not help
  • the second (changing path in /etc/docker/daemon.json) helped finding –runtime nvidia

so
sudo docker run -it --rm --runtime nvidia --network host nvcr.io/nvidia/l4t-ml:r32.7.1-py3
started without problems. But tensorflow could still not be imported due to the missing library libcudart.so.10.2. I searched for it manually in the container, could not locate it either.

6

Hi,

The container of JetPack 4 is mounted CUDA-related library from the Jetson device.
So you will need to install the JetPack components to get the CUDA library.

Thanks.