[CPMD-list] how to speed up on amd64 machines

[Axel Kohlmeyer]

On Mon, 16 Apr 2007, Bin Pan wrote:

BP> Hi Prof. Hutter,
BP> 
BP> I said that it is very slow because I used 6 CPU's on a Dual Core AMD 
BP> Opteron(tm) Processor 870 connected by myrinet and compared the
BP> time required to finish one MD cycle with that done by 4 CPU's on a 
BP> Intel(R) Xeon(TM) CPU 3.00GHz connected by ethernet for the same system.
BP> The first needs ~90 seconds, however the second only needs ~60.
BP> I expected the first one has faster CPU and better network connection.
BP> 
BP> Are these enough info for you to give me some advice in how to 
BP> improve the speed?

first of all. the TCPU timing is not necessarily accurate.
please check the end of the job whether there is a difference
between cpu time and elapsed time.

now, there are two more issues to check:
go to the section bracketed with PARAPARAPARA lines
and check that you are really using the 6/4 cpus
as you expect.
finally, you didn't specify which BLAS/LAPACK library
you chose. that can make a large difference.

for your reference, here is the set of flags that i am
using on our dual opteron and dual-core dual opteron 
cluster with myrinet and gigabit using OpenMPI (with intel fortran).

FFLAGS = -pc64  -tpp6 -O2 -unroll
LFLAGS = -L/cmm/pkg/fftw2/lib/ -L$(HOME)/lib -lacml -ldfftw -lacml_mv 
CFLAGS = -O2 -Wall -m64
CPP = /lib/cpp -P -C -traditional
CPPFLAGS = -D__Linux -D__PGI -DFFT_FFTW -DPOINTER8 -DLINUX_IFC \
        -DPARALLEL -DMYRINET
NOOPT_FLAG = -O0 -pc64 -tpp6
CC = gcc
FC = /cmm/pkg/openmpi-1.2/bin/mpif77 -c
LD = /cmm/pkg/openmpi-1.2/bin/mpif77 -i-static 


cheers,
   axel.

BP> Thanks a lot!
BP> 
BP> Best regards,
BP> Bin.
BP> 
BP> At 02:57 4/16/2007, Juerg Hutter wrote:
BP> >Hi
BP> >
BP> >you have to give us more precise information.
BP> >What exactly do you mean by 'it is very slow'.
BP> >Do you mean slow in general, e.g. compared to
BP> >another plane wave code or slow because
BP> >of bad speedups in parallel compared to the
BP> >serial runs?
BP> >You should also give us some information on your
BP> >computer. What is the exact version of your CPUs
BP> >and what is the type of your network?
BP> >At the end of each run CPMD prints timing information.
BP> >Please attach at least one such timing information
BP> >block for a run that you consider too slow.
BP> >
BP> >regards
BP> >
BP> >Juerg Hutter
BP> >
BP> >----------------------------------------------------------
BP> >Juerg Hutter                   Phone : ++41 44 635 4491
BP> >Physical Chemistry Institute   FAX   : ++41 44 635 6838
BP> >University of Zurich           E-mail: hutter at pci.unizh.ch
BP> >Winterthurerstrasse 190
BP> >CH-8057 Zurich, Switzerland
BP> >----------------------------------------------------------
BP> >
BP> >
BP> >On Sun, 15 Apr 2007, Bin Pan wrote:
BP> >
BP> >>Hi CPMD users,
BP> >>
BP> >>I am wondering how to speed up CPMD runs on amd64 cluster using MPI
BP> >>parallelization.
BP> >>Previously I compiled CPMD using pgi compiler with acml. However, I
BP> >>found it is very slow.
BP> >>The flags I used are:
BP> >>
BP> >>CPPFLAGS = -P -C -traditional -D__Linux -D__PGI -DFFT_DEFAULT
BP> >>-DPOINTER8 -D__pgf90 -DPARALLEL -DMYRINET
BP> >>CC = gcc -O2 -Wall -m64
BP> >>FC = pgf90 -c -fastsse -tp k8-64
BP> >>LD = pgf90 -fastsse -tp k8-64
BP> >>
BP> >>Can you please let me know how to improve the performance?
BP> >>Thanks a lot!
BP> >>
BP> >>Best regards,
BP> >>Bin.
BP> >>
BP> >>_______________________________________________
BP> >>CPMD-list mailing list
BP> >>CPMD-list at cpmd.org
BP> >>http://cpmd.org/mailman/listinfo/cpmd-list
BP> 
BP> 
BP> 
BP> ---------------------------------------------------------------------
BP> Bin Pan
BP> Ph.D. Candidate
BP> Department of Chemical Engineering
BP> Massachusetts Institute of Technology
BP> Tel: 617-253-6675
BP> E-mail: binpan at mit.edu
BP> Room E19-528, 77 Mass. Ave. Cambridge, MA
BP> ---------------------------------------------------------------------- 
BP> 
BP> _______________________________________________
BP> CPMD-list mailing list
BP> CPMD-list at cpmd.org
BP> http://cpmd.org/mailman/listinfo/cpmd-list
BP> 

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