Multi-GPU and distributed-memory algorithms
Project Links
Meta
Author: NVIDIA Corporation
Requires Python: >=3.11
Classifiers
Intended Audience
- Developers
Topic
- Database
- Scientific/Engineering
Programming Language
- Python
- Python :: 3.11
- Python :: 3.12
- Python :: 3.13
- Python :: 3.14
RapidsMPF
Collection of multi-GPU, distributed memory algorithms. RapidsMPF provides a unified framework for asynchronous, multi-GPU pipelines using simple streaming primitives built using NVIDIA CUDA-X components.
Documentation
Build from Source
git clone https://github.com/rapidsai/rapidsmpf.git
cd rapidsmpf
mamba env create --name rapidsmpf-dev --file conda/environments/all_cuda-133_arch-$(uname -m).yaml
./build.sh
See the Getting Started guide for debug builds, AddressSanitizer, MPI/UCX test suites, and rrun launcher details.
26.10.0
Oct 08, 2026
26.8.1
Aug 26, 2026
26.8.0
Aug 06, 2026
26.6.0
Jun 09, 2026
26.4.0
Apr 09, 2026
26.2.0
Feb 07, 2026
25.12.0
Dec 11, 2025
25.10.0
Oct 09, 2025
25.8.0
Aug 07, 2025
25.6.0
Jun 06, 2025
0.0.0a0
Apr 16, 2025
Wheel compatibility matrix
| Platform | CPython >=3.11 (abi3) |
|---|---|
| manylinux_2_24_aarch64 | |
| manylinux_2_24_x86_64 | |
| manylinux_2_28_aarch64 | |
| manylinux_2_28_x86_64 |
