Jupyter Notebook Containers with CUDA Support

This repository contains a Makefile that can be used to build and push Jupyter notebook containers that include support for NVIDIA CUDA, using the Jupyter Docker Stacks project.

Download from Dockerhub

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

The following image and tags are currently available from Dockerhub.

CUDA 12.0.0 runtime image

• base-notebook-cuda:cuda12.0.0-runtime-ubuntu22.04-py3.10
• minimal-notebook-cuda:cuda12.0.0-runtime-ubuntu22.04-py3.10
• r-notebook-cuda:cuda12.0.0-runtime-ubuntu22.04-py3.10
• scipy-notebook-cuda:cuda12.0.0-runtime-ubuntu22.04-py3.10
• datascience-notebook-cuda:cuda12.0.0-runtime-ubuntu22.04-py3.10
• tensorflow-notebook-cuda:cuda12.0.0-runtime-ubuntu22.04-py3.10
• pytorch-notebook-cuda:cuda12.0.0-runtime-ubuntu22.04-py3.10
• pyspark-notebook-cuda:cuda12.0.0-runtime-ubuntu22.04-py3.10
• all-spark-notebook-cuda:cuda12.0.0-runtime-ubuntu22.04-py3.10

CUDA 11.7.1 cudnn8-runtime image

• base-notebook-cuda:cuda11.7.1-cudnn8-runtime-ubuntu22.04-py3.10
• minimal-notebook-cuda:cuda11.7.1-cudnn8-runtime-ubuntu22.04-py3.10
• r-notebook-cuda:cuda11.7.1-cudnn8-runtime-ubuntu22.04-py3.10
• scipy-notebook-cuda:cuda11.7.1-cudnn8-runtime-ubuntu22.04-py3.10
• datascience-notebook-cuda:cuda11.7.1-cudnn8-runtime-ubuntu22.04-py3.10
• tensorflow-notebook-cuda:cuda11.7.1-cudnn8-runtime-ubuntu22.04-py3.10
• pytorch-notebook-cuda:cuda11.7.1-cudnn8-runtime-ubuntu22.04-py3.10
• pyspark-notebook-cuda:cuda11.7.1-cudnn8-runtime-ubuntu22.04-py3.10
• all-spark-notebook-cuda:cuda11.7.1-cudnn8-runtime-ubuntu22.04-py3.10

For other combinations of Jupyter, CUDA, Ubuntu and Python, please build your own using the included Makefile.

Build your own

Prerequisites

• Docker
• Make
• Access to a Dockerhub account for pushing the images

Getting Started

1. Clone this repository and the docker-stacks submodule

git clone --recurse-submodules https://github.com/kentwait/docker-jupyter-cuda.git

2. Open the Makefile and edit the following variables at the top of the Makefile to match your desired configuration:

CUDA_VERSION := 11.7.1
CUDA_FLAVOR := cudnn8-runtime
CUDA_OS := ubuntu22.04
PYTHON_VERSION := 3.10

3. Run make to build the images

To build all the images from the most basic docker-stacks-foundation-cuda to the all-pyspark-notebook-cuda, run the following command:

make build-all

To build only a specific image and its prerequisite images, run the following command:

make build/<image-name>

Available image names are listed below

• docker-stacks-foundation-cuda
• base-notebook-cuda
• minimal-notebook-cuda
• r-notebook-cuda
• scipy-notebook-cuda
• datascience-notebook-cuda
• tensorflow-notebook-cuda
• pytorch-notebook-cuda
• pyspark-notebook-cuda
• all-spark-notebook-cuda

4. Run make to push the images to your Dockerhub account

make push-all

Targets

The following make targets are available:

• docker-stacks: clones the Jupyter Docker Stacks repository from GitHub
• build/docker-stacks-foundation-cuda: builds the foundation image with CUDA support
• build/%-cuda: builds the specified image with CUDA support
• build-all: builds all images with CUDA support
• push/%: pushes the specified image to Dockerhub
• push-all: pushes all images to Dockerhub