some ab initio benchmarks

Andrew Wang andrewxwang at yahoo.com.tw
Sun Oct 19 10:39:36 EDT 2003


I have 2 pts:

1. The compilers used across different platforms were
not the same, why not use the Intel compiler for the
P4 as well?

2. What is the working set of the benchmark? If the
benchmark fit in the 6MB on-chip L3 of the Itanium2,
it is very likely to perform very well.

Another benchmark that shows the G5 wins the large
memory case, loses small/medium cases, while the
Itanium2 loses most of its advantages when the working
set does not fit the L3:

http://www.xlr8yourmac.com/G5/G5_fluid_dynamics_bench/G5_fluid_dynamics_bench.html

Andrew.

 --- Cezary Czaplewski <czarek at sun1.chem.univ.gda.pl>
的訊息:> 
> Hi,
> 
> quite recently I did some benchmarks of P4 2.8 GHz
> against dual Opteron
> 1400MHz, dual Itanium2 1400MHz and dual k7mp
> 2133MHz(MP 2600+) and some
> older machines. For comparison I am including
> benchmarks of dual P3 512
> 1200MHz I got from Wayne Fisher, The University of
> Texas at Dallas. On
> Opteron I have also tried PC GAMESS program which I
> received from Alex
> Granovsky.
> 
> 
> 1. Single point HF energy calculation for
> Ace-Gly-NMe in 6-31G*
>     (155 basis functions)
> 
> g03: mem=100MW TEST 6-31G* nosym  scf=(tight,incore)
> gamess: MEMORY=20000000 DIRSCF=.TRUE.			  [sec]
> 
>  itek       g03    Itanium2 1400MHz          efc 7.1
>      26.5
>  prototype  g03    p4 512   2800MHz           pgi4  
>      41.1
>  dahlia     g03    Opteron  1400MHz           pgi4  
>      49.5
>  m211       g03    k7mp     2133MHz(MP 2600+) pgi4  
>      83.3
>  Wayne      g03    p3 512   1200MHz           pgi4  
>      85
>  m211       gamess k7mp     2133MHz(MP 2600+) ifc7.1
>      92.5
>  prototype  gamess p4 512   2800MHz           ifc7.1
>     106.5 
>  dahlia     PCgamess   Opteron   1400MHz	        
> 112.9
>  dahlia     gamess Opteron  1400MHz           ifc7.1
>     128.5
>  itek       gamess Itanium2 1400MHz          efc 7.1
>     150.8
> 
> 2. Single point MP2 energy calculation for
> Ace-Gly-NMe in 6-31G*
>      (155 basis functions)
> 
> g03: mem=100mw rwf=a,250MW,b,250MW,c,250MW TEST
> rmp2/6-31G* nosym 
> scf=(tight,incore)
> MaxDisk=750MW
> gamess: MEMORY=50000000 DIRSCF=.TRUE.
> 
>  itek       g03    Itanium2 1400MHz           efc
> 7.1     51.7
>  prototype  g03    p4 512   2800MHz	       pgi4     
> 111.0
>  dahlia     g03    Opteron  1400MHz            pgi4 
>     150.7
>  m211       gamess k7mp     2133MHz(MP 2600+) 
> ifc7.1    154.2
>  prototype  gamess p4 512   2800MHz           
> ifc7.1    157.0
>  dahlia     PCgamess   Opteron   1400MHz		 163.8
>  dahlia     gamess Opteron  1400MHz           
> ifc7.1    191.0
>  itek       gamess Itanium2 1400MHz           efc
> 7.1    194.8
>  m211       g03    k7mp     2133MHz(MP 2600+)  pgi4 
>     251.6
>  Wayne      g03    p3 512   1200MHz            pgi4 
>     303
> 
> 3. Manfreds Gaussian Benchmark
>
http://www.chemie.uni-dortmund.de/groups/ocb/projekte/mg98b.html
> 
> 243 basis functions 399 primitive gaussians
> RHF/3-21G* Freq
>                                                     
>   [sec]
> 
> itek       g03     Itanium2  1400MHz    efc 7.1     
>   2843
> prototype  g03     p4 512    2800MHz pgi 4	      
> 8084
> dahlia     g03     Opteron   1400MHz pgi 4          
>   9332
> m211       g03     k7mp 2133MHz(MP 2600+) pgi 4     
>  10289
> Wayne      g03     p3 512    1200MHz	pgi 4        
> 12920
> galera     g03     p3xenon    700MHz    pgi 3       
>  19317
> m001       g03     p3         650MHz    pgi 4       
>  22824
> 
> 4. test397.com from gaussian03
> 
> 882 basis functions,  1440 primitive gaussians
> rb3lyp/3-21g force test scf=novaracc
>                                                     
>  [sec]
> 
> itek       g03     Itanium2  1400MHz    efc 7.1     
>  6733
> prototype  g03     p4 512    2800MHz	pgi 4       
> 12980
> dahlia     g03     Opteron   1400MHz    pgi 4       
> 17879
> m211       g03     k7mp 2133MHz(MP 2600+) pgi 4     
> 20521
> Wayne      g03     p3 512    1200MHz    pgi 4       
> 24521
> galera     g03     p3xenon    700MHz    pgi 3       
> 39353
> 
> 5. Gaussian calculations of NMR chemical shifts for
> GlyGlyAlaAla
> 
> 207 basis functions,   339 primitive gaussians
> %MEM=800MB
> B3LYP/GEN  NMR                                      
> [sec]
> 
> itek       g03     Itanium2  1400MHz    efc 7.1     
> 275
> prototype  g03     p4 512    2800MHz	  pgi 4     
> 614
> dahlia     g03     Opteron   1400MHz      pgi 4     
> 849
> m211       g03     k7mp 2133MHz(MP 2600+) pgi 4     
> 948
> Wayne      g03     p3 512    1200MHz      pgi 4    
> 1134
> 
> some details:
> 
> g03 is GAUSSIAN 03 rev. B04 with gaussian blas
> compiled with 32-bit pgi4.0
> 
> gamess is  VERSION 6 SEP 2001 (R4) compiled with
> 32-bit ifc 7.1, for P4 I 
> have used additional options -tpp7 -axKW
> 
> Opteron (dahlia) had 64bit GinGin64 Linux and I had
> to use static 32-bit 
> binaries.  It should have SuSE Linux Enterprise soon
> and I will repeat 
> tests using PGI 5.0 64-bit compiler when it will be
> ready.
> 
> Itanium2 (itek) uses gamess  VERSION = 14 JAN 2003
> (R3) compiled with
> 64-bit efc and GAUSSIAN 03 rev. B04 with mkl60
> compiled with 64-bit efc 
> 7.1
> 
> P3xenon (galera) uses gamess VERSION = 6 SEP 2001
> (R4) compiled with ifc
> 6.0 and GAUSSIAN 03 rev B.01 with gaussian blas
> compiled with pgi 3.3
> 
> 
>                                 czarek
> 
>
----------------------------------------------------------------------
>                      Dr. Cezary Czaplewski
> Department of Chemistry                  Box 431
> Baker Lab of Chemistry
> University of Gdansk                     Cornell
> University
> Sobieskiego 18, 80-952 Gdansk, Poland    Ithaca, NY
> 14853
> phone: +48 58 3450-430                   phone:
> (607) 255-0556
> fax: +48 58 341-0357                     fax: (607)
> 255-4137
> e-mail: czarek at chem.univ.gda.pl          e-mail:
> cc178 at cornell.edu
>
----------------------------------------------------------------------
> 
> 
> 
> 
> 
> 
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or
> unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf 

-----------------------------------------------------------------
每天都 Yahoo!奇摩
海的顏色、風的氣息、愛你的溫度,盡在信紙底圖
http://tw.promo.yahoo.com/mail_premium/stationery.html
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list