The latest release of OpenBLAS comes with substantial improvements for single precision matrix multiplication on IBM Z. Thanks to a hand-tuned compute kernel, the implementation makes optimal use of the IBM Z instruction set to provide a big performance benefit. As an algebra library implementing the BLAS interface (Basic Linear Algebra Subprograms), OpenBLAS is a central building block for AI and other compute-intensive workloads. In a Linux system, OpenBLAS serves as compute back-end for many functions in NumPy, SciPy, R, and Octave.
For further details, see the OpenBLAS 0.3.12 release notes here.
No comments:
Post a Comment