Check this blog if you want a Docker Container with Nvidia GPU Support.

1. About Image

1.1. Check Existing Images and Statistics

1
2
3
4

# List all images
docker images
# Check disk usage of docker (images, conatiners)
docker system df

To list all images not being used by any container, run:

1
2
3
4
5

docker images --filter "dangling=false" -q | xargs -r docker inspect --format '{{ .Id }} {{ .RepoTags }}' | grep -v '<none>' | while read id tags; do
  if ! docker ps -a --format '{{.Image}}' | grep -q "$(echo $tags | cut -d'[' -f2 | cut -d']' -f1)"; then
    echo "$id $tags"
  fi
done

1.2. Rename an Image

1

docker tag <image-id> <image-name>:<image-tag>

1.3. Pull an Image

Search for an image in Docker Hub.

Pull the image by running:

1

docker pull <image-name>:<image-tag>

Or download the layers to <dir-path> by using this script:

1
2
3
4
5
6

mkdir -p <dir-path>  # Directory to store the layers

bash download-frozen-image-v2.sh <dir-path> <image-name>:<image-tag>

# Load the image
tar -cC <dir-path> . | docker load

1.4. Build an Image

Use a Dockerfile to build a new image based on an existing image.

Check this example project: jamesnulliu/deeplearning-docker-build.

The following command in ./scripts/build.sh would build image <image-name>:<image-tag>:

1
2
3
4

docker build \
    -f Dockerfile \
    -t jamesnulliu/deeplearning:torch2.6-cuda12.6-ubuntu24.04 \
    .

Note that . specifiles the build context, where docker can reference (e.g., COPY and ADD) the containing files during the build process. You can change it to other directory path according to your needs.

You will find the new image after building by runing docker images.

1.5. Save an Image to a tar File

To export an image to a tar file with reusable layer imformations, run the following command:

1
2
3

docker save -o <some-name>.tar <image-name>:<image-tag>
# Or
docker save -o <some-name>.tar <image-id>

See 1.6. Load an Image to load the generated tar file to an image.

1.6. Load an Image

Suppose you have a tar file <some-name>.tar, if the file is generated by docker save (see 1.5. Save an Image to a tar File), load the image by running:

1

docker load -i <some-name>.tar

Or if the file is generated by docker export (see 2.4 Export a container to a tar File), load the container to an image by running:

1

docker import <some-name>.tar <image-name>:<image-tag>

1.7. Remove an Image

Before removing an image, you need to:

  • Stop containers using the image: docker stop <container-id>.
  • Remove containers using the image: docker rm <container-id>.

Then remove the image:

1
2
3

docker rmi <image-name>:<image-tag>
# Or
docker rmi <image-id>

2. About Container

2.1. Check Existing Containers and Statistics

1
2
3
4
5
6

# List running/[all] containers
docker ps [-a]
# Check statistics of all/[some] containers
docker stats [<container-id>]
# Check disk usage of docker (images, conatiners)
docker system df

2.2. Create a Container

Basic Parameters:

Click to expand
  • -it: Interactive mode with pseudo-TTY terminal
  • --name <container-name>: Assign a name to the container for easier reference
  • -p <host-port>:<container-port>: Map host port to container port (e.g. 8888:22)
  • --entrypoint /bin/bash: Override default entrypoint with bash shell
  • <image-name>: Name of the Docker image to use (e.g. ubuntu, nvidia/cuda)
  • <image-tag>: Version/tag of the image (e.g. latest, 20.04)

Additional Options:

Click to expand
  • --shm-size <shm-size>G: Set size of /dev/shm (shared memory) in GB (e.g. 2G)
  • --gpus all: [Optional] Give container access to all GPUs (requires nvidia-docker)
  • -v <host-path>:<container-path>: Mount host directory into container
  • -e KEY=VALUE: Set environment variables
  • --network=host: Use host network stack; If set, -p is not needed, and you can access host services directly with localhost
  • --ipc=host: Use host IPC namespace
  • --privileged: Give extended privileges to container
  • -u $(id -u):$(id -g): Run as current user instead of root
  • --rm: Automatically remove container when it exits

Example:

1
2
3
4
5
6
7
8
9

# Create and run a container from <image-name>:<image-tag>
docker run -it \
    --gpus all \
    --name <container-name> \
    -p <host-port>:<container-port> \
    --entrypoint /bin/bash \
    --shm-size <shm-size>G \
    <image-name>:<image-tag>
# Now you are in the container.

To exit container and return to host, run exit, which will stop the containter; Or type Ctrl+P+Q, which will detach from container without stopping it.

2.3. Start, Stop, Remove and Rename a Container

Start a container:

1

docker start <container-name>

Attach to a container:

1

docker exec -it <container-name> bash

Stop a container:

1

docker stop <contianer-name>

Remove a container:

1

docker rm <container-name>

Rename a container:

1
2
3

docker rename <container-id> <new-name>
# or
docker rename <old-name> <new-name>

2.4. Export a Container to a tar File

Note that docker export only exports the container’s filesystem without layer information. It is more recommended to use docker save to save a image to a tar file (see 1.5. Save an Image to a tar File).

1

docker export <container-id> > <some-name>.tar

See 1.6. Load an Image to load the generated tar file to an image.

3. Change Docker Root Directory to a Different Location

Check current docker root directory:

1

docker info | grep "Docker Root Dir"

Stop docker daemon:

1

systemctl stop docker

Edit (or create) the docker configuration file:

1

vim /etc/docker/daemon.json

Add or modify the configuration:

1
2
3

{
    "data-root": "<target-docker-root>"
}

Copy the existing docker data to target location:

1

rsync -aP <current-docker-root> <target-docker-root>

Restart docker daemon:

1
2

systemctl daemon-reload
systemctl start docker

Verify the new location:

1

docker info | grep "Docker Root Dir"