[Wrf-users] WRF efficiency on different architectures

[Don Morton]

Dear WRF Users:

I would like to point you to a site that a handful of us have been working
on, albeit unfunded, for about a year

http://weather.arsc.edu/WRFBenchmarking/

At this page you'll find single-domain benchmark cases meant primarily for
testing raw WRF computations from domains of hundreds of thousands to over
a billion grid points.  These cases have, in fact, been used by NSF and US
Department of Defense as part of required benchmarking for acquisition
proposals.

A second set of cases is more "real world," intended to help users test the
nested domains of odd configurations that they're more likely to use.

We have lots of ambitions for this site, but again, the whole effort is
volunteer right now.  Your comments are appreciated and, if you choose to
test some of these cases we would very much appreciate your allowing us to
post these findings.

Best Regards,

Don Morton and Delia Arnold

2012/3/22 Guillermo Fernández <guillermo.fernandez at meteogalicia.es>

> I'm aware of the compilers as well as node interconnection impact in the
> efficiency but at this point my interest is about
> architecture/efficiency relation.
>
> The link provided by the wrfhelp people:
>
> http://www.mmm.ucar.edu/wrf/WG2/bench/
>
> is pretty much what I'm looking for. Those Benchmarks even with 2 years
> old are very informative. However more recent data would be great.
>
> Thank you all.
>
> Guillermo
>
>
> On 03/22/2012 01:47 AM, Saji Hameed wrote:
> > Setting aside the efficiency factor introduced by compilers, you would
> > have to consider the impact of interconnecting
> > devices.  In my case, I use relatively slow processors (1.7 GHz)
> > interconnected by fast network (InfiniBand).  Each node in
> > my system is a 4 socket AMD supermicro server with each socket having
> > a12core AMD Magnycours at 1.7GHz.  Each node
> > is interconnected through InfiniBand.  I have four such nodes and they
> > seem to giving quite good scaleup.  I do not have
> > the numbers right now, but I could post it later if there is interest.
> >
> > saji
> >
> > On Tue, Mar 20, 2012 at 9:59 PM, Guillermo Fernández
> > <guillermo.fernandez at meteogalicia.es> wrote:
> >> Hello all,
> >>
> >> I'm looking for information about WRF efficiency on different
> >> architectures (AMD, Intel, p-series,...). Any information is welcome.
> >>
> >> Regards
> >>
> >> Guillermo
> >>
> >> --
> >> MeteoGalicia
> >>     Guillermo Fernández García
> >>     Dpto. Predición Numérica e Investigación Rúa Roma, 6
> >>     15.707 Santiago de Compostela
> >>     Tel: +34 981 957 462
> >>     Fax: +34 981 957 466
> >>     guillermo.fernandez at meteogalicia.es
> >>     www.meteogalicia.es
> >>
> >> _______________________________________________
> >> Wrf-users mailing list
> >> Wrf-users at ucar.edu
> >> http://mailman.ucar.edu/mailman/listinfo/wrf-users
> >
> >
> >
>
>
> --
> MeteoGalicia
>     Guillermo Fernández García
>     Dpto. Predición Numérica e Investigación
>      Rúa Roma 6, 15.707 Santiago de Compostela
>      Tel: +34 981 957 462
>     Fax: +34 981 957 466
>     guillermo.fernandez at meteogalicia.es
>     www.meteogalicia.es
>
> _______________________________________________
> Wrf-users mailing list
> Wrf-users at ucar.edu
> http://mailman.ucar.edu/mailman/listinfo/wrf-users
>



-- 
Voice:  +1 907 450 8679
Arctic Region Supercomputing Center
http://weather.arsc.edu/
http://people.arsc.edu/~morton/ <http://www.arsc.edu/%7Emorton/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/wrf-users/attachments/20120322/6a3cdef0/attachment.html