Efficiency of general Krylov methods on GPUs - An experimental study

Beitrag bei einer Tagung
(Konferenzbeitrag)


Details zur Publikation

Autor(en): Anzt H, Dongarra J, Kreutzer M, Wellein G, Köhler M
Verlag: IEEE Computer Society
Jahr der Veröffentlichung: 2016
Seitenbereich: 683-691
ISBN: 9781509021406


Abstract


This paper compares different Krylov methods based on short recurrences with respect to their efficiency whenimplemented on GPUs. The comparison includes BiCGSTAB, CGS, QMR, and IDR using different shadow space dimensions. These methods are known for their good convergencecharacteristics. For a large set of test matrices taken from theUniversity of Florida Matrix Collection, we evaluate the methods'performance against different target metrics: convergence, number of sparse matrix-vector multiplications, and executiontime. We also analyze whether the methods are «orthogonal»in terms of problem suitability. We propose best practicesfor choosing methods in a «black box» scenario, where noinformation about the optimal solver is available.


FAU-Autoren / FAU-Herausgeber

Kreutzer, Moritz
Regionales Rechenzentrum Erlangen (RRZE)
Wellein, Gerhard Prof. Dr.
Professur für Höchstleistungsrechnen


Autor(en) der externen Einrichtung(en)
Max-Planck-Institut für Dynamik komplexer technischer Systeme / Max Planck Institute for Dynamics of Complex Technical Systems
Oak Ridge National Laboratory


Zitierweisen

APA:
Anzt, H., Dongarra, J., Kreutzer, M., Wellein, G., & Köhler, M. (2016). Efficiency of general Krylov methods on GPUs - An experimental study. (pp. 683-691). IEEE Computer Society.

MLA:
Anzt, Hartwig, et al. "Efficiency of general Krylov methods on GPUs - An experimental study." Proceedings of the 30th IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2016 IEEE Computer Society, 2016. 683-691.

BibTeX: 

Zuletzt aktualisiert 2018-20-11 um 20:50