Skip to main content

Install MAGE × NVIDIA cuGraph

Follow this guide to install Memgraph with NVIDIA cuGraph GPU-powered graph algorithms.



To be able to run cuGraph analytics, make sure you have compatible infrastructure first. The exact system requirements are available at the NVIDIA RAPIDS site, and include an NVIDIA Pascal (or better) GPU and up-to-date CUDA & NVIDIA drivers.

Docker requirements 🐳

If running MAGE × NVIDIA cuGraph in Docker, the following applies:

Local build requirements:

If building MAGE × NVIDIA cuGraph locally, these requirements apply (tested on Ubuntu):

Installing the Docker image from Docker Hub

The simplest way of starting Memgraph with cuGraph GPU analytics is to download the image from Docker Hub. Just pull the image, and get it running with these simple commands:

docker run --rm --gpus all -p 7687:7687 -p 7444:7444 memgraph/memgraph-mage:1.3-cugraph-22.02-cuda-11.5

Depending on your environment, different versions of MAGE/cuGraph/CUDA can be installed:

docker run --gpus all -p 7687:7687 -p 7444:7444 memgraph/memgraph-mage:${MAGE_VERSION}-cugraph-${CUGRAPH_VERSION}-cuda-${CUDA_VERSION}

To see the available versions, explore our Docker Hub organization and look for the images tagged memgraph-mage.


The development image with cuGraph support is not available yet. If you want to develop cuGraph-powered query modules in Docker, do not hesitate to contact us about it.

Building MAGE with NVIDIA cuGraph locally with Docker

  1. Download the MAGE source code from GitHub:

    git clone && cd mage
  2. Build the MAGE × cuGraph-tagged Docker image:

    docker build -f Dockerfile.cugraph -t memgraph-mage .
  3. Start Memgraph-MAGE with the following command:

    docker run --rm --gpus all -p 7687:7687 -p 7444:7444 --name mage memgraph-mage

You can now query Memgraph from querying platforms such as Memgraph Lab or mgconsole.

If you made any changes while the Docker container was running, you need to stop the container and rebuild the image. For a workaround, check Development process for MAGE with Docker.

Installing MAGE natively from the source


Make sure you have installed all prerequisites and dependencies before building the MAGE × NVIDIA cuGraph from source.

  1. Download the MAGE source code from GitHub and run the setup script. It will generate a dist directory with all the needed files:

    python3 setup build --gpu

    The --gpu flag enables building the cuGraph dependencies and creating the shared library with cuGraph algorithms that are loaded into Memgraph.

  2. Copy the contents of the newly created dist directory to /usr/lib/memgraph/query_modules:


    To speed the installation up, you can specify a path for the setup script to copy the built executables:

    python3 setup build -p /usr/lib/memgraph/query_modules --gpu
  3. Start Memgraph and enjoy MAGE × cuGraph!


    If your Memgraph instance was already running, execute the following query inside one of the querying platforms to reload the modules:

    CALL mg.load_all();

    If the modules are still missing, restart the instance by running systemctl stop memgraph and then systemctl start memgraph.

    For more about loading query modules, consult this guide.