Numerical Library Eigensolver Performance on PRACE Tier-0 Systems

Parallel eigensolver operations are at the computational core of many large-scale scientific and engineering application codes. This project analyses parallel performance of established and newly developed parallel dense symmetric eigensolver numerical library routines on PRACE Tier-0 systems using real datasets from large-scale application codes. This whitepaper builds upon the research report ‘Dense Linear Algebra Performance Analysis on the Fujitsu BX900 OPL Machine†’ (from the same author) which can be found at: http://www.openpetascale.org/documents /2011_Dense_Linear_Algebra_Performance_Analysis_on_the_Fujitsu_BX900_OPL_Machine.pdf. This version of the report is updated with results from ScaLAPACK and with the recently released ELPA library routines on the PRACE Tier-0 machines CURIE and JUGENE