Impressive stream results (nvidia nforce/crush)

Bill Broadley bill at math.ucdavis.edu
Tue Feb 26 14:12:43 EST 2002


> Sounds good. What's the exact model?

NV7m (microatx)

> The new durons are based on the palomino core . . . albeit that's only
> the 1200's, IIRC.

I just swapped CPU's, kinda interesting.

Linux-2.4.17, redhat-7.2 based (modifications for diskless), NV7m,
2 PC2100 Dimms.

Duron 950: (100 Mhz FSB, 100 Mhz memory, 66 Mhz agp)
Function      Rate (MB/s)   RMS time     Min time     Max time
Copy:         742.4336       0.0889       0.0862       0.0899
Scale:        443.9945       0.1495       0.1441       0.1507
Add:          958.3234       0.1030       0.1002       0.1037
Triad:        570.7626       0.1728       0.1682       0.1740

Athlon XP1600, 1.4 Ghz: (133 Mhz FSB, 133 Mhz memory, 66 Mhz agp)
-------------------------------------------------------------
Function      Rate (MB/s)   RMS time     Min time     Max time
Copy:         827.1189       0.0777       0.0774       0.0786
Scale:        726.8103       0.0887       0.0881       0.0901
Add:          994.5098       0.0977       0.0965       0.1059
Triad:        858.9759       0.1121       0.1118       0.1140

Apparently the XP's prefetch helps in some cases where nvidia's doesn't,
i.e. scale and triad.  Doesn't manage much higher peak bandwidth which
is fairly impressive.  

In any case these are the best numbers I've seen from an Athlon based
system.

-- 
Bill Broadley
Mathematics/Institute of Theoretical Dynamics
UC Davis
_______________________________________________
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