Preferred Networks Releases CuPy v11
TOKYO – July 28, 2022 – Preferred Networks, Inc. (PFN) today released CuPy™ v11, the new major update to the open-source library for general-purpose matrix calculation.
CuPy v11 provides the following new features:
- Distributed computation of sparse matrix
cupyx.distributed, a module for collective and peer-to-peer communications, now supports sparse matrices. This allows users to perform computation against sparse matrices in multi-node, multi-GPU environments.
- Arm SBSA & JetPack 5 support
CUDA on Arm SBSA is now officially supported. This allows users to deploy GPU workloads at lower cost by using the combination of Arm-based servers and NVIDIA GPUs. In addition, the support for JetPack 5, the latest SDK for NVIDIA Jetson embedded platform, has also been added in this release.
- Enhanced NumPy/SciPy-compatible APIs
96 compatible APIs are added, improving the interoperability with NumPy and SciPy.
Check 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.