[Wien] Benchmark-Compilation options

[Martin Hilgeman]

Dear Clazo,

The build options should be fine and not too compiler dependent. For
this benchmark, the diagonalization of the Hamiltonian is the most time
consuming, so it is very dependent on the efficiency of the linear
algebra being used.

What I have used on the R_LIB line is:

  -lgoto_it2 -L/opt/intel/mkl/8.0.2/lib/64 -lmkl_lapack64 -lmkl -lvml

I would rather not use SCSL, is slower here than MKL. My timing on an
Altix 3700 with 1.6 GHz 9MB L3 cache Itanium 2 processor:

       TIME HAMILT (CPU)  =    10.5, HNS =    15.7, DIAG =    93.3

   ===> TOTAL CPU       TIME:    121.6 (INIT =      2.0 + K-POINTS =
119.6)
   > SUM OF WALL CLOCK TIMES:    122.0 (INIT =      2.2 + K-POINTS =
119.9)
      Maximum WALL clock time:    122.227270841599
      Maximum CPU time:           121.663085937500

Your Altix has a smaller L3 cache, but faster FSB. I get on a system
with the same processor (1.6 GHz, 6 MB L3) but slower FSB:


       TIME HAMILT (CPU)  =    11.2, HNS =    15.8, DIAG =    94.0

   ===> TOTAL CPU       TIME:    123.2 (INIT =      2.1 + K-POINTS =
121.1)
   > SUM OF WALL CLOCK TIMES:    123.6 (INIT =      2.2 + K-POINTS =
121.4)
      Maximum WALL clock time:    123.825130939484
      Maximum CPU time:           123.211914062500


This is all with WIEN2k_06.1.

Lastly one note on the use of the GOTO library. It uses by default a
number of threads that equal to the number of processors of your
machine. To get reproducible timings, I recommend you to set the number
of threads to 1:

  % setenv GOTO_NUM_THREADS 1


regards,

-Martin

On Fri, 2006-06-09 at 08:30 +0200, Peter Blaha wrote:
> The benchmark numbers are directly from the SGI benchmark-center 
> (Martin Hilgeman <hilgeman at sgi.com>). I do not know more details
> about compiler options, maybe you ask him directly.
> 
> For your reference: The "partial " timing is
>        TIME HAMILT (CPU)  =     9.9, HNS =    15.6, DIAG =    94.3
> 
> Regards
> 
> > I compiled the wien code  VERSION_05.6: 20.10.2005  in a SGI Altix 4700 system with Itanium2 Madison 9M Prozessoren (1.66GHz/6MB L3 Cache) , ifort version 9.0 + mkl8.0 + libgoto_ITANIUM2-r1.00.a,  and I carried out the standard test_case provided in the wien2k web site in order to see how well the tunning of the compiler and libraries was. I obtained a running time of 160 Sec. The running time for a similar setup reported in the wien2k Benchmarks site is:
> > 
> > Itanium2(1.6GHz,SGI Altix 3700)  122 sec    ifort9.0 +mkl8.0, libgoto_itanium2_64p-r1.00
> > 
> > Thus, faster than the time I got. The compiling options  (most of them are the default options recommended by the siteconfig script) are the following:
> > 
> >      O   Compiler options:        -O3 -tpp2 -ftz -ip -fno-alias -cm -w -FR -DALTIX -DINTEL_VML
> >      L   Linker Flags:            -L../SRC_lib -Vaxlib
> >      P   Preprocessor flags       '-DParallel'
> >      R   R_LIB (LAPACK+BLAS):     -L/fibus/fs1/16/vt4cl/LIB/GotoBLAS/libgoto_ITANIUM2-r1.00.a -lscs -L/opt/intel/mkl/8.0.2/lib/64/ -lvml
> > 
> > I downloaded and compiled the last version of GOTO library from the web.
> > 
> > Has somebody having a similar system been able to reproduce the running time posted in the wien2k Benchmark page (122 sec)? If so, could you post the options used to compile the code?
> > 
> > The running time I get (160 sec) is already good, but I think, if the code is able to run faster in the computational setup available, I should try to tune the compiler and libraries in order to get the best possible performance.
> > 
> > Thank you in advance, 
> > 
> > 
> > Cesar Lazo
> > Technische Universität Hamburg-Harburg
> > 
> > 
> > 
> > 
> > 
> > 
> > _____________________________________________________________________
> > Der WEB.DE SmartSurfer hilft bis zu 70% Ihrer Onlinekosten zu sparen!
> > http://smartsurfer.web.de/?mc=100071&distributionid=000000000071
> > 
> > _______________________________________________
> > Wien mailing list
> > Wien at zeus.theochem.tuwien.ac.at
> > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> > 
> 
> 
>                                       P.Blaha
> --------------------------------------------------------------------------
> Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
> Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
> Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
> --------------------------------------------------------------------------
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
-- 
Martin Hilgeman                    Chemical Applications Engineering
                                   Phone:  +31(0)30-6696885
SGI                                E-mail: hilgeman at sgi.com
The Netherlands                    URL:    http://www.sgi.com/go/chembio