https://jasongrout.github.io/medium-archive/pelican/2024-04-17T09:01:00+00:00The Project Jupyter blog: news, releases, and community stories, archived from blog.jupyter.org.2024-04-16T22:08:00+00:002024-04-17T09:01:00+00:00Ayaz Salikhovtag:jasongrout.github.io,2024-04-16:/medium-archive/pelican/posts/2024/cuda-enabled-jupyter-docker-images/

<p>I am happy to announce that Jupyter Docker Stacks project now provides GPU accelerated Docker images. This allows PyTorch or TensorFlow…</p>

<p>I am happy to announce that <a href="https://jupyter-docker-stacks.readthedocs.io/en/latest/index.html">Jupyter Docker Stacks project</a> now provides GPU accelerated Docker images. This allows PyTorch or TensorFlow operations to use compatible NVIDIA GPUs for accelerated computation.</p> <h3 id="how-to-find-these-images">How to find these images</h3> <p>All our images are available on <a href="https://quay.io/organization/jupyter">Quay.io registry</a>. We provide CUDA accelerated versions of images are available for <code>tensorflow-notebook</code> and <code>pytorch-notebook</code>.</p> <p>To use such an image, you have to specify a special prefix tag to the image: versioned CUDA prefix like <code>cuda11-</code> or <code>cuda12-</code> for <code>pytorch-notebook</code> or just <code>cuda-</code> for <code>tensorflow-notebook</code>.</p> <blockquote> <p>We build <code>pytorch-notebook</code> only for 2 last major versions of CUDA, <code>tensorflow-notebook</code> image supports only the latest CUDA version listed in the <a href="https://www.tensorflow.org/install/source#gpu">officially tested build configurations</a> list. CUDA-enabled images are available on x86_64 platform.</p> </blockquote> <p>For example, you can use an image <code>quay.io/jupyter/pytorch-notebook:cuda12-python-3.11.8</code> or <code>quay.io/jupyter/tensorflow-notebook:cuda-latest</code>.</p> <h3 id="running-the-images">Running the images</h3> <p>To enable CUDA support on <strong>Linux</strong> host machine, you need to:</p> <ul> <li>have compatible Nvidia GPU</li> <li>have NVIDIA Linux driver installed</li> <li>add <code>--gpus all</code> (or <code>--gpus '&quot;device=all&quot;'</code>) flag to if you’re using <code>Docker</code></li> <li>add <code>--device 'nvidia.com/gpu=all'</code> flag if you’re using <code>Podman</code></li> </ul> <p>For example, you can run the image like this: <code>docker run -it --rm -p 8888:8888 --gpus all quay.io/jupyter/tensorflow-notebook:cuda-latest</code>.</p> <p>You can also enable GPU support on <strong>Windows</strong> using <a href="https://docs.docker.com/desktop/gpu/">Docker</a> or <a href="https://github.com/containers/podman/issues/19005">Podman</a>.</p> <h3 id="reaching-out">Reaching out</h3> <p>You will find our documentation <a href="http://jupyter-docker-stacks.readthedocs.io">here</a>.<br> If you encounter any problems, please, take a look at <a href="https://github.com/jupyter/docker-stacks">GitHub issues</a>.<br> If you would like to improve GPU support in our images, please, reach out <a href="https://github.com/jupyter/docker-stacks/issues/1557">here</a>.</p> <h3 id="acknowledgments">Acknowledgments</h3> <p>I would like to thank:</p> <ul> <li><a href="https://github.com/johanna-reiml-hpi">Johanna Reiml</a>, for implementing general <code>variant</code> concept for our images and making it work for <code>pytorch-notebook</code> (PR <a href="https://github.com/jupyter/docker-stacks/pull/2091">#2091</a>)</li> <li><a href="https://github.com/ChristofKaufmann">Christof Kaufmann</a>, for making it work for <code>tensorflow-notebook</code> (PR <a href="https://github.com/jupyter/docker-stacks/pull/2100">#2100</a>)</li> <li><a href="https://github.com/benz0li">Olivier Benz</a>, for helping me out to write this blog post</li> </ul>