MAGMA
A library for BLAS and LAPACK on GPUs
Area: Mathematical libraries
CASS member: FASTMath
Description
MAGMA (Matrix Algebra on GPUs and Multicore Architectures) is a node-level mathematical library for dense linear algebra algorithms on GPU-accelerated HPC systems. MAGMA provides BLAS, batch BLAS, LAPACK, and batch LAPACK functionality on NVIDIA, AMD, and Intel GPUs, often complementing the industrial numerical libraries. MAGMA pioneered hybrid CPU-GPU algorithms for dense matrix factorizations, linear solvers, eigenvalue, and singular value decompositions. MAGMA has been partially adopted by NVIDIA and AMD into their numerical linear algebra libraries. It is an integral component of many DOE applications, and is part of xSDK and E4S.
Target audience
Computational scientists from various domains, numerical analysts, HPC scientists, certain AI applications, developers of other math libraries such as finite elements analysis, sparse direct solvers, computational statistics, computer vision, and others.