For those interested, I came across a website that has some interesting
benchmarks posted for P4 and Xeon processors in single and cluster
configurations. The text is in Japanese, but the tables are more-or-less
self explanatory (AltaVista's translator helps somewhat:
http://world.altavista.com/). The primary table of interest is located
1/3 of the way down the following page:
http://www.hpc.co.jp/IA-Parallel/HP2SC-PP4-bench.htm
The clusters use gigabit ethernet and MPICH. Values are reported using
both PGI and Intel Fortran compilers (though versions are not apparent).
I've tried to summarize some of the values below. The columns are, in
order: compiler, cpu type, cpu speed, number of cpus, OS, interconnect
type, message-passing protocol, Prowat6 result (seconds). The actual
table also includes numbers for R10k and Alpha 21264 based
systems/clusters.
(PGI) P4 2.0GHz 8 SuSE 1000Base-TX Mpich-ch_p4 25.52
(PGI) P4 2.0GHz 4 SuSE 1000Base-TX Mpich-ch_p4 33.21
(PGI) P4 2.0GHz 2 SuSE 1000Base-TX Mpich-ch_p4 49.6
(ifc) P4 2.0GHz 1 SuSE - - 74.0
(PGI) P4 2.0GHz 1 SuSE - - 85.9
(ifc) Xeon 1.7GHz 1 SuSE - - 88.3
The website actually has a number of interesting comparisons which can be
accessed from the home page:
http://www.hpc.co.jp
I hope this proves helpful.
Robert
_________________________
R. M. Fesinmeyer
rmf_at_u.washington.edu
Department of Chemistry
University of Washington
Received on Thu Oct 11 2001 - 13:10:14 PDT
