A novel parallel QR algorithm for hybrid distributed memory HPC systems

Robert Granat
Department of Computing Science
Umeå University


Computing the eigenvalues of a square dense matrix A is at the very heart of numerical linear algebra, with applications coming from a broad range of science and engineering. In this talk we present a novel variant of the parallel QR algorithm for solving dense nonsymmetric eigenvalue problems on hybrid distributed high performance computing (HPC) systems, which outperforms the existing ScaLAPACK code by one to two orders of magnitude.

This talk presents joint work with Bo Kågström, Dept Computing Science and HPC2N, Umeå University, and Daniel Kressner, Seminar for applied mathematics, ETH Zürich, Switzerland.