CUDA-X (Formerly RAPIDS)

CUDA-X (Formerly RAPIDS)#

NVIDIA CUDA-X allows researchers to adapt existing CPU-based Python data analytics and machine learning workflows for GPU acceleration with relatively small code changes.

Note

Beginning August 11, 2026, the NVIDIA RAPIDS brand will transition to NVIDIA CUDA-X. All library functionality remains the same. However, during this transition, much of the documentation may continue to reference RAPIDS and/or CUDA-X. Please note that both refer to the same set of NVIDIA supported tools and libraries.

Using our CUDA-X Environment#

Note

This environment contains only the basic packages required to run CUDA-X and Python 3.13. If you’d like to install additional packages for use alongside CUDA-X, please follow the instructions at the bottom to create your own custom environment.

  1. Start an interactive session on an NVIDIA GPU compute node, or create a batch script.

sinteractive --partition=aa100 --qos=gpu-testing --ntasks=10 --gres=gpu:a100_3g.20gb:1 --nodes=1 --time=01:00:00

  1. Load the miniforge module:

module load miniforge

  1. Activate the environment:

mamba activate /curc/sw/conda_env/rapids-25.10

  1. Start using CUDA-X in your Python code! The CUDA-X user guide has some great examples.

Creating a Custom CUDA-X Environment#

Note

The example below is for installing CUDA-X version 26.08. Information on installing the most recent version can be found in the CUDA-X installation guide.

  1. Start an interactive session on a GPU node:

sinteractive --partition=aa100 --qos=gpu-testing --ntasks=10 --gres=gpu:a100_3g.20gb:1 --nodes=1 --time=01:00:00

  1. Load the miniforge module to use Mamba:

module load miniforge

  1. Install CUDA-X:

mamba create -n rapids-26.08 -c rapidsai -c conda-forge rapids=26.08 python=3.14 'cuda-version>=13.0,<=13.3'

  1. Activate the environment:

mamba activate rapids-26.08

  1. Start using CUDA-X in your Python code! The CUDA-X user guide has some great examples.