Preferred Networks Releases CuPy v13
Over 140 signal processing functions added through support of Chan Zuckerberg Initiative
TOKYO – January 18, 2024 – Preferred Networks, Inc. (PFN) today released CuPy™ v13, the new major update to the open-source library for general-purpose matrix calculation.
CuPy v13 provides the following new features:
- Enhanced Signal Processing Module
Over 140 functions have been added to the signal processing module (cupyx.scipy.signal). These APIs provide signal processing algorithms such as B-Spline interpolation and filtering with the same API as SciPy, making it easy to utilize GPUs for such tasks. The development of this module was supported by Essential Open Source Software for Science program of the Chan Zuckerberg Initiative.
- Multi-GPU Distributed Array (Preview)
A sharded array that distributes processing across multiple GPUs to perform operations has been added. This feature was initially developed during the PFN Summer Internship Program 2023.
- Support Import without an installed GPU Toolkit
It is now possible to import CuPy in environments where NVIDIA CUDA is not installed. This makes it easier to write codes that support both NumPy and CuPy.
Please see the announcement in the CuPy blog for more details.
PFN will continue to swiftly incorporate the latest research outcomes into CuPy while collaborating with supporting companies and open source communities.