With the explosive rise in hardware parallelism, running applications efficiently on modern high-performance computing systems requires a total redesign of the underlying numerical methods. This new paradigm must include implementations that focus on node-level parallelism, reduced global communication volume, and relaxed synchronization requirements. The Fixed-Point Methods for Numerics at Exascale (FiNE) research group aims to tackle these challenges head on by developing a new class of algorithms for extreme-scale computing.
Software Projects:
The numerical linear algebra package Ginkgo
The numerical linear algebra package MAGMA
The Exasclae Computing Project PEEKS
Major Collaborations:
Marc Baboulin, University of Paris-Sud
Jack Dongarra, University of Tennessee
Jens Saak, Max Planck Institute for Dynamics of Complex Technical Systems
Enrique S. Quintana-Ortí, University Jaime I
Daniel B. Szyld, Temple University
Thomas Huckle, Technical University of Munich
Gerhard Wellein, Friedrich-Alexander University Erlangen-Nürnberg
External website containing additional information.