Peter Zhang
Apr 23, 2025 13:20
NVIDIA has open-sourced cuPyNumeric 25.03, enhancing its accessibility with PIP set up and native HDF5 assist, fostering transparency and collaboration in multi-GPU computing.
NVIDIA has introduced that its cuPyNumeric 25.03 library is now fully open supply, marking a major milestone in its improvement. The replace introduces highly effective new capabilities, together with assist for PIP set up and native HDF5 IO, in line with NVIDIA.
Full Open Supply Transition
With this newest launch, NVIDIA has open-sourced the complete stack of cuPyNumeric, together with the Legate framework and runtime layer that powers it, underneath the Apache 2 license. This transition underscores NVIDIA’s dedication to fostering transparency, reproducibility, and collaboration within the improvement group. The open-source nature permits contributors to discover, audit, and improve the system with out obstacles.
PIP Set up Assist
Beforehand installable solely by way of conda, cuPyNumeric can now be put in utilizing PIP, simplifying the setup course of considerably. This enhancement facilitates simpler integration into workflows, digital environments, and CI pipelines. The cuPyNumeric bundle on PyPI is multinode and multirank succesful, supporting each single-node a number of GPUs and multi-GPU multinode clusters.
Native HDF5 IO Assist
One other notable function of cuPyNumeric 25.03 is its native assist for HDF5 by GPU Direct Storage, which optimizes the dealing with of huge datasets. This function ensures environment friendly IO operations, vital for high-performance computing and data-intensive functions. Customers can now handle complicated knowledge constructions with improved efficiency and portability.
Set up and Utilization
The set up course of has been streamlined to incorporate a easy PIP command: pip set up nvidia-cupynumeric. This replace bundles all main dependencies besides MPI, that are in any other case simply resolvable by PyPI. NVIDIA gives detailed steering on establishing and working cuPyNumeric on SLURM clusters, emphasizing the convenience of use in multinode and multirank environments.
For additional particulars and to discover the great capabilities of cuPyNumeric 25.03, NVIDIA encourages customers to evaluation the official launch notes and contribute to the continued improvement by way of the GitHub repository.
Picture supply: Shutterstock
