dl-env-docker

Dockerfiles for building a deep learning development environment, which is based on CUDA & cuDNN images provided by the nvidia/cuda repository on Docker Hub.
For more details, read my blog posts.

Prerequisites

• Systems with NVIDIA GPUs that support CUDA
  • You can check the GPU model name installed in your current computer with the lspci | grep -i nvidia command in the terminal.
  • On the https://docs.nvidia.com/deeplearning/cudnn/latest/reference/support-matrix.html page, check the supported NVIDIA graphics driver versions, required CUDA Compute Capability conditions, and the list of supported NVIDIA hardware for each cuDNN version.
  • Find the corresponding model name in the GPU list at https://developer.nvidia.com/cuda-gpus, and check the Compute Capability value. This value must meet the CUDA Compute Capability condition checked earlier to use CUDA and cuDNN without issues.
• Host OS with the appropriate version of the graphics driver installed
• NVIDIA Container Toolkit
• Docker Engine

Overview of Dockerfiles

Three flavors of images are provided:

• base: Includes CUDA & cuDNN, setuptools, and pip.
• torch: Builds on the base and includes the JupyterLab, Numpy, Scipy, pandas, Matplotlib, Seaborn, scikit-learn, tqdm, and PyTorch.
• rapids: Builds on the torch and includes the Cupy, cuDF, cuML, and DALI.

How to use

Using pre-built images

Pre-built images are available in the yunseokim/dl-env repository on Docker Hub.
You can pull the images with the following command.

# Tag(base)
docker pull yunseokim/dl-env:base-cuda12.4.1-cudnn9.1.0-ubuntu22.04

# Tag(torch)
docker pull yunseokim/dl-env:torch-cuda12.4.1-cudnn9.1.0-ubuntu22.04

# Tag(rapids)
docker pull yunseokim/dl-env:rapids-cuda12.4.1-cudnn9.1.0-ubuntu22.04

When you run a container from this prebuilt image, the account password for the 'remote' user is initialized to 000000. It is extremely vulnerable, so reset the login password immediately after the first run of the container.

Build manually

After downloading the desired Dockerfile from this repository, you can build the image yourself by running the following command from the path where the Dockerfile is located.

docker build -t dl-env:<tag> -f ./Dockerfile . \
--build-arg USER_PASSWORD=<password>

Enter the tag of your choice in the <tag> field, and in the <password> field, enter the login password to be used when remotely login to the 'remote' account with SSH.

Docker run example

• Make sure to add the --gpus all argument to use GPUs in the container runtime.
• Below is an example format for a Docker run. You can customize it to your needs.

docker run -itd --name dl-env --gpus all \
-p <server-port>:<container-jupyter-port> \
-p <server-port>:<container-ssh-port> \
--restart <always> \
-v <volume-name>:<workspace> \
-w <workspace> \
<image-name>:<tag>