Einstein vs SETI performance by processor type
17 Apr 2006 22:02:18 UTC
Topic 191096
(moderation:
Hi all
The aim of this comparision is to maximize scientific calculations output for peoples who participates in both projects, Einstein and SETI.
For measure of relative project effectivenes on given processor I chose mean SETI WU time to mean Einstein WU time ratio. If all processors would be equal in performance for both projects this ratio would remains constant (although both WU times will be shorter for more powerful processors of course).
I got next ratios:
[pre]
Athlon Thunderbird 256 KBytes cache 2.259
AthlonXP Thoroughbred 256 KBytes cache 1.769
AthlonXP Barton 512 KBytes cache 1.341
Athlon 64 Winchester 512 KBytes cache 0.710
Athlon 64 Venice 512 KBytes cache 0.702
Pentium 4 Prescott 1024 KBytes cache 0.375
[/pre]
So old Athlon w/o any SSE support much more applicable for Einstein than for SETI.
All results were aquired on best optimized applications available for both projects on particular processor type. So probably lack of 3DNow! support for SETI app. and small cache make Athlon unappropriate for SETI relative to other processor types.
But it seems P4 Prescott performs very bad for Einstein project. Not only by relative ratio but by WU completion times too (for SETI Athlon 64 ~3k sec, P4 ~3k sec too; for Einstein Athlon 64 ~4k sec but P4 ~ 8,5k sec - much worse!).
Any explanation of P4 strange behaviour is welcomed.
If someone has such statistic for other processor types please post it in this thread.
Einstein vs SETI performance by processor type
)
There've been official postings on why AMD significantly outperforms Intel in einstien, and IIRC it was that the Einstien app is branchy and the p4's long pipeline becomes a major liability.
I cannot give any figures,
)
I cannot give any figures, because the computer no longer has figures for Seti.
But I moved a Celeron 1.1GHz (Tualatin) L1 cache 2 * 32Kbyte, L2 cache 256Kb, to 100% crunching on Einstein because it performs much better here. On Seti it was out performed by my P3 933MHz but the reverse is true here.
Andy
DanNeely, Winterknight Thanx
)
DanNeely, Winterknight
Thanx for info
Could you list cache size for your P3 too?
It seems for like arhitectures big cache size wins much more for SETI than for Einstein as it was mentioned already on these forums.
RE: DanNeely,
)
Heh, I was just on a page listing those specs for PIIIs and Celerons (I noticed my celeron doing better than someone else's PIII of same frequency so I was curious myself)
specs
It's 256k L2 according to that
RE: DanNeely,
)
The L2 cache for the P3 (coppermine) is also 256Kb, wonder if it is FSB speed, P3 is 133MHz (supposedly mine reports 132 in Everest and PCWizard) whilst Celeron is FSB 100MHz.
Andy