[Beowulf] scaling SPECint_base2000 for harpertown

Tom Elken tom.elken at qlogic.com
Wed Nov 21 13:09:59 EST 2007


 > -----Original Message-----
> [mailto:beowulf-bounces at beowulf.org] On Behalf Of Tom Elken
> Sent: Wednesday, November 21, 2007 9:02 AM
> To: andrew holway; Beowulf Mailing List
> Subject: RE: [Beowulf] scaling SPECint_base2000 for harpertown
> 

> > http://www.moonet.co.uk/2004
> 
> >From what I can interpret, this URL shows SPECint_base2000 = 11751
> for a 3.0 GHz Harpertown at -O2 optimization.
> 
> The highest result on the SPEC website for SPECint_base2000 I 
> could find is 3108 with full optimization (-fast +FDO, etc.). 
> 
> How did you come up with the 11751 score?

Andrew and I exchanged some private e-mails and here is where it came
from:

Someone from Intel ran SPECint with gcc/g++ using low optimisation flags
(-O2 -pthread -fPIC) on one core of the system(s).  Then they multiplied
the result by the # of cores, 8, to come up with the 11751 score on the
3 GHz CPU.

Note that this NOT what the SPEC rate benchmark would do -- SPEC rate
would run 8 copies simultaneously and time the slowest completion of
each benchmark -- and the score is on a different scale than
SPECint_rate_base2000 uses.

But if you interpret it as a measure of single core performance and not
full system performance, it's fine.

-Tom


> 
> If you were running SPECint_rate_base2000 at 8 cores, at full 
> optimization, the score would likely be in the 200-300 range.
> 
> Cheers,
> -Tom Elken
> 
> 
> > 
> > Cheers
> > 
> > Andy
> > _______________________________________________
> > 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
> 

_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

!DSPAM:4744753a110769691080364!



More information about the Beowulf mailing list