TY - GEN
T1 - Efficiency of general Krylov methods on GPUs - An experimental study
AU - Anzt, Hartwig
AU - Dongarra, Jack
AU - Kreutzer, Moritz
AU - Wellein, Gerhard
AU - Köhler, Martin
N1 - Publisher Copyright:
© 2016 IEEE.
PY - 2016/7/18
Y1 - 2016/7/18
N2 - 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.
AB - 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.
KW - Algorithmic bombardment
KW - BiCGSTAB
KW - CGS
KW - GPU
KW - IDR(s)
KW - Krylov solver
KW - QMR
UR - http://www.scopus.com/inward/record.url?scp=84991666764&partnerID=8YFLogxK
U2 - 10.1109/IPDPSW.2016.45
DO - 10.1109/IPDPSW.2016.45
M3 - Conference contribution
AN - SCOPUS:84991666764
T3 - Proceedings - 2016 IEEE 30th International Parallel and Distributed Processing Symposium, IPDPS 2016
SP - 683
EP - 691
BT - Proceedings - 2016 IEEE 30th International Parallel and Distributed Processing Symposium, IPDPS 2016
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 30th IEEE International Parallel and Distributed Processing Symposium Workshops, IPDPSW 2016
Y2 - 23 May 2016 through 27 May 2016
ER -