[Beowulf] Athlon64 / Opteron test

Daniel Fernandez daniel at labtie.mmt.upc.es
Mon May 17 10:01:50 EDT 2004


Those results are a bit surprising, it seems that the performance gains
from 32bit (Athlon) to 64bit (Opteron) is nearly 0.5% . 

Considering that is using only single precision and x87 operations run
equally fast at same clockspeed on both processors ( not at all ) it's
not bandwith limited ... ¿ How about double precision performance
results ?

-- 
Daniel Fernandez <daniel at labtie.mmt.upc.es>
Heat And Mass Transfer Center - CTTC
www.cttc.upc.edu
c/ Colom nº11
UPC Campus Industrials Terrassa , Edifici TR4


On Fri, 2004-05-14 at 18:26, Karl Bellve wrote:
> 
> 
> We have a project that I  have timed on various CPUs, and compilers.
> It makes heavy use of FFTW (single precision).
> 
> Running native i386 code does very well for us on the Opteron, unlike
> the Itanium 2.
> 
> 
> Itanium 2 1.0  GHz on Red Hat Linux Advanced Workstation release 2.1AW
> 234 seconds (IA64 code and ECC, -fno-alias -IPF_fma -Ob2 -ipo
> -restrict)
> 1515 seconds (i386 code and GCC)
> 
> AMD Opteron 1.4 GHz on United Linux 1.0
> 189 seconds (gcc, x86_64 code)
> 221 seconds (gcc, i386 code)
> 
> Athlon 1.5 GHz  (AMD Athlon(tm) MP Processor 1800+) on Redhat 8.0
> 190 seconds (gcc, i386 code)
> 
> Intel Xeon 1.7 GHz on Redhat 8.0
> 267 seconds (gcc, i386 code)
> 
> Sun Intel Xeon 2.8 GHz on Redhat 7.3
> 150 seconds (gcc, i386 code)
> 
> Alpha 667 GHz
> 430 seconds (Compaq compiler for both, optimized code)
> 
> Intel 2.2Ghz Xeon on RH 7.3
> 208 seconds (gcc, i386 code)
> 
> -- 
> Cheers,
> 
> 
> 
> Karl Bellve
> 
> ______________________________________________________________________
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

_______________________________________________
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