[Beowulf] Table of best compiler flags for different CPUs?
landman at scalableinformatics.com
Wed May 11 09:37:44 PDT 2005
Try -fastsse with the Portland group compilers. You might also try
-Mipa=fast . For laughs, pull down PathScale as well. When it is
faster, it is usually by a bit. When it isn't, they are pretty close.
Steve Cousins wrote:
> On Tue, 10 May 2005, Kozin, I (Igor) wrote:
>>It is generally a good idea to look up what other
>>published benchmarks used to build their codes.
>>Two obvious good sources are SPEC and Stream.
>>http://www.polyhedron.com/ is quite useful too.
>>But I must admit that the obscure options used
>>there were seldom useful to me apart from building
>>the respective codes on respective hw.
> Thanks. Someone else suggested spec.org to me offline. This does have
> valuable information but I was hoping that someone had already created a
> table like this and it would save time if they might share it.
>>>>I am interested in what combinations of compilers and
>>>>switches will give a better comparison, as well as run our
>>serial or parallel? Please, post a link if published.
> I haven't really posted but here is what I have found:
> CPU: Speed CPUs Time to run: Compiler and Flags
> ============= ======= ==== ============ ====================
> PIII 866 Mhz 2 30:10 pgf77 -fast -Mconcur
> Athlon 2800+ 2.1 Ghz 1 13:26 pgf77 -fast -Mconcur
> SGI MIPS 400 Mhz 4 12:08 f77 -apo -O2
> Athlon 3000+ 2.1 Ghz 1 11:27 pgf77 -fast -Mconcur
> Athlon 2800+ 2.1 Ghz 1 10:19 pgf77 -fast -Mconcur
> Desktop G5 2.0 Ghz 1 9:15 xlf_r *
> Xeon 3.0 Ghz 1 9:08 pgf77 -fast -Mconcur
> Desktop G5 2.0 Ghz 2 7:20 xlf_r *
> Opteron 1.6 Ghz 1 7:15 pgf77 -fast -Mconcur
> Xserve G5 2.0 Ghz 2 7:11 xlf_r *
> SGI MIPS 400 Mhz 8 6:55 f77 -apo -O2
> Opteron 2.2 Ghz 1 6:10 pgf77 -fast -Mconcur
> Opteron 1.6 Ghz 2 5:37 pgf77 -fast -Mconcur
> Opteron 2.6 Ghz 1 5:32 pgf77 -fast -Mconcur
> Xeon 3.0 Ghz 1 5:29 pgf77 -fast -Mconcur
> Opteron 2.6 Ghz 1 3:12 pgf77 -fast -Mconcur
> * flags for xlf_r: -O5 -qstrict -qarch=g5 -qtune=g5 -qsmp -qunroll=yes
> The program that was run is a stripped down ocean model (Princeton Ocean
> Model) that reads in initial conditions at the beginning and then does a
> one day run without any further I/O.
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax : +1 734 786 8452
cell : +1 734 612 4615
More information about the Beowulf