Efficiency of general Krylov methods on GPUs - An experimental study

Hartwig Anzt, Jack Dongarra, Moritz Kreutzer, Gerhard Wellein, Martin Köhler

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

13 Scopus citations

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.

Original languageEnglish
Title of host publicationProceedings - 2016 IEEE 30th International Parallel and Distributed Processing Symposium, IPDPS 2016
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages683-691
Number of pages9
ISBN (Electronic)9781509021406
DOIs
StatePublished - Jul 18 2016
Externally publishedYes
Event30th IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2016 - Chicago, United States
Duration: May 23 2016May 27 2016

Publication series

NameProceedings - 2016 IEEE 30th International Parallel and Distributed Processing Symposium, IPDPS 2016

Conference

Conference30th IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2016
Country/TerritoryUnited States
CityChicago
Period05/23/1605/27/16

Keywords

  • Algorithmic bombardment
  • BiCGSTAB
  • CGS
  • GPU
  • IDR(s)
  • Krylov solver
  • QMR

Fingerprint

Dive into the research topics of 'Efficiency of general Krylov methods on GPUs - An experimental study'. Together they form a unique fingerprint.

Cite this