[Pw_forum] a problem with 64-bit OS by multiprocess calculation

[Axel Kohlmeyer]

On Wed, 2009-07-15 at 13:18 +0300, Alexej Mazheika wrote:
> Hi
> 
> I apply Espresso-package under Linux with mpich2 applying it for
> parallel processes calculations. The Espresso I compiled from
> source-files with mpich2-compilers. I noticed a strange phenomenon.
> The CPU of my PC is Intel Core Quad (64-bit). When I make my
> calculations under 64-bit linux-system (openSUSE 11.0) it lasts longer
> than under 32-bit one (openSUSE-11.0). For example, the average time
> of SCF procedure of my task is 8 min under 64-bit OS vs 7 min under
> 32-bit OS. Can this problem be binded with installation or work of
> Espresso? If yes, then how to solve it? I asked the developers of
> mpich2 about it, and they couldn't help me. 

i don't think that this is a strange phenomenon at all.
in general, a 64-bit compilation of a code should be expected
to be slower than a 32-bit compilation. all memory pointers
are twice the size and thus require more memory bandwidth and 
incur a reduced cache efficiency.

only, it usually doesn't show so much on x86_64 since the 32-bit 
x86 cpu design is to horribly register starved and AMD doubled 
the number of registers to make up for it. however, if your job
runs only for a few minutes then you may still see the additional
overhead from the larger pointers. for a larger problem, that 
should be reduced. if not, there may be different compilers or
compiler settings.

on non-x86 CPUs a 64-bit mode compilation on codes like Q-E 
is typically about 10-20% slower than a 32-bit compile.

cheers,
   axel.
> 
> regards,
> 
> Alex Mazheika, Research institut for Physical Chemical Problems of
> BSU, Leningradskayastr. 14, 220050 Minsk, Belarus
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum

-- 
=======================================================================
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.