[Pw_forum] Pentium-M versus Itanium
akohlmey at cmm.chem.upenn.edu
Thu Mar 13 21:11:24 CET 2008
On Thu, 13 Mar 2008, Miguel Martínez Canales wrote:
MMC> Dear all,
a couple of comments:
MMC> real 1m44.686s
MMC> user 1m41.374s
MMC> sys 0m1.864s
MMC> real 0m8.411s
MMC> user 0m7.828s
MMC> sys 0m0.140s
those examples make bad benchmarks.
they compute too little, what you are mostly
seeing is the time to write the output... :)
MMC> Big Bad Itanium
MMC> real 1m48.417s
MMC> user 2m15.136s
MMC> sys 0m6.756s
MMC> real 0m13.028s
MMC> user 0m12.775s
MMC> sys 0m0.627s
MMC> Is it normal that user time > real time even when I am running pw.x as
MMC> mpirun -np 1 pw.x? Is it normal that my almost 4 year old laptop running a
as of version 10, MKL defaults to automatically switching to
multi-threaded at its own whim (see docs) unless you tell it
to run serially. hence the user time increase.
MMC> bloated distro is running the examples clearly faster than an Itanium
MMC> machine? I swear I haven't used fancy flags anywhere, other than the
MMC> default ones. I will also run espresso on an opteron machine, to see the
you'll have to run an input that at least stays on the CPU for
something of the order of 10mins to get some realistic numbers.
MMC> PS: I also swear that the very same Itanium machine used to run twice as
MMC> fast as the same laptop, although the free FORTRAN compilers have changed a
MMC> lot in the last couple of years.
Axel Kohlmeyer akohlmey at cmm.chem.upenn.edu http://www.cmm.upenn.edu
Center for Molecular Modeling -- University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425
If you make something idiot-proof, the universe creates a better idiot.
More information about the users