nvidia-docker required?

I’m trying to use Clara for the first time.
The directions here (https://ngc.nvidia.com/catalog/containers/nvidia:clara-train-sdk) states that I need nvidia-docker HOWEVER, the nvidia-docker page (https://github.com/NVIDIA/nvidia-docker) states “Note that with the release of Docker 19.03, usage of nvidia-docker2 packages are deprecated since NVIDIA GPUs are now natively supported as devices in the Docker runtime.”

My questions:

  1. Do I need nvidia-docker to run Clara if I have docker 19.03 installed
  2. Can someone point me to an explanation of the --runtime option and how this relates to this situation? (I tried googling but the term “runtime” is ubiquitous).

I must say that this process doesn’t seem all that straightforward…
Please help.

Thanks!

Dan

Hi Dan,

You should be able to run the container with the following command(s) with docker 19.03, there might be some additional options needed to mount directories within the docker container:

docker run --gpus all -it --rm --ipc=host --net=host --shm-size=1g nvcr.io/nvidia/clara-train-sdk:v1.0-py3 /bin/bash

docker run --gpus ‘“device=1,2”’ -it --rm --ipc=host --net=host --shm-size=1g nvcr.io/nvidia/clara-train-sdk:v1.0-py3 /bin/bash

With older version of docker we needed to use the runtime option make GPUs visible from inside containers.

We will be updating our documentation to reflect these changes.