Re: AMBER: performance of amber7/xlf/macosX 10.3 with JAC on xserve g5.2.0GHz

From: Tru Huynh <tru.pasteur.fr>
Date: Tue, 25 May 2004 16:31:34 +0200

On Mon, May 24, 2004 at 12:12:39PM -0700, Mengjuei Hsieh wrote:
> Hi,
>
> You can check
> http://apple.sysbio.info/~mjhsieh/archives/000236.html
> http://apple.sysbio.info/~mjhsieh/archives/000234.html
> for some tweaks I did before.

Thanks,

my final figures for amber7:

*** GNU: -> (3600*24/1175) = 73.5 ns/day
real 19m35.889s
user 19m35.170s
sys 0m0.460s
| Setup wallclock 1 seconds
| Nonsetup wallclock 1175 seconds
MACHINE:
setenv MACHINE "linux/FreeBSD/Windows PC/Mac OSX"
setenv MACH Linux
setenv MACHINEFLAGS
setenv CPP "cpp -traditional "
setenv CC "gcc -fomit-frame-pointer -O3"
setenv SYSDIR Machines/standard
setenv G77 "g77 "
setenv LOAD "g77 -O6"
setenv LOADLIB "-lm"
setenv G77_COMPAT "-fno-globals -ff90 -funix-intrinsics-hide"
setenv G77_OPT "-funroll-all-loops -fno-globals -ff90 -funix-intrinsics-hide -malign-natural -fno-globals -Wno-globals"
setenv L0 "$G77 -c -g -O0 $G77_COMPAT"
setenv L1 "$G77 -c -O1 $G77_OPT $G77_COMPAT"
setenv L2 "$G77 -c -O2 $G77_OPT $G77_COMPAT"
setenv L3 "$G77 -c -O3 $G77_OPT $G77_COMPAT"
setenv RANLIB ranlib

*** XLF: -> (3600*24/781) = 110 ns/day

real 13m0.913s
user 13m0.320s
sys 0m0.550s
| Setup wallclock 0 seconds
| Nonsetup wallclock 781 seconds
MACHINE:
setenv MACHINE "IBM rs6000"
setenv MACH rs6000
setenv MACHINEFLAGS "-DMEM_ALLOC -DCLINK_PLAIN "
setenv CPP /opt/ibmcmp/xlf/8.1/exe/cpp
setenv SYSDIR Machines/rs6000
setenv LOAD "xlf -O5"
setenv LOADLIB " -L/opt/ibmcmp/xlf/8.1/lib -lxlfpad "
setenv CC xlcA
setenv FC_COMPAT "-qxlf77=leadzero -qarch=g5 -qtune=g5 -qunroll=yes"
setenv L0 "xlf -O3 $FC_COMPAT -c"
setenv L1 "xlf -O3 $FC_COMPAT -c"
setenv L2 "xlf -O4 $FC_COMPAT -c"
setenv L3 "xlf -O5 $FC_COMPAT -c"
setenv RANLIB ranlib


Final conclusions for our test with amber7/c29b1 JAC on opteron and g5:
The G5 has only closed the gap for MD simulations. It is not faster/slower.

Xserve G5 .2GHz + IBM compiler == Opteron .2GHz (NetBSD+gcc.3.4.0)
for both c29b1 and amber7 on the JAC benchmark.

Best regards,

Tru

-- 
Dr Tru Huynh          | http://www.pasteur.fr/recherche/unites/Binfs/
mailto:tru.pasteur.fr | tel/fax +33 1 45 68 87 37/19
Institut Pasteur, 25-28 rue du Docteur Roux, 75724 Paris CEDEX 15 France  
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Tue May 25 2004 - 15:53:01 PDT
Custom Search