A survey of sparse matrix-vector multiplication performance on large matrices

TL;DR

This paper surveys the performance of sparse matrix-vector multiplication on large matrices across various software libraries, hardware architectures, and matrix formats, providing a comprehensive comparison of their efficiency.

Contribution

It offers a detailed third-party performance comparison of multiple SpMV implementations on large matrices, hardware, and formats, which was previously lacking.

Findings

Performance varies significantly across libraries and hardware.

Certain formats outperform others depending on the architecture.

The survey highlights optimal configurations for large-scale SpMV computations.

Abstract

We contribute a third-party survey of sparse matrix-vector (SpMV) product performance on industrial-strength, large matrices using: (1) The SpMV implementations in Intel MKL, the Trilinos project (Tpetra subpackage), the CUSPARSE library, and the CUSP library, each running on modern architectures. (2) NVIDIA GPUs and Intel multi-core CPUs (supported by each software package). (3) The CSR, BSR, COO, HYB, and ELL matrix formats (supported by each software package).