[RC5] Performance on P-IV

Jonathan Smith jsmith at cygnacom.com
Thu May 10 13:45:13 EDT 2001

If the distributed.net client doesn't recognize the P4 yet, it certainly
mean that their isn't a P4 optimized core available.  The PII core and the
Athlon core have been reoptimized 5 or 6 times over 2 or 3 years to get the
current performance.  It is unsurprising that the P4, which uses a radically
different internal architecture and certainly has different bottlenecks that
either the PII or the Athlon does poorly when running a core designed to
squeeze the last little bit of performance out very different chip.  
The cores are tuned to take advantage of things like how long it take
specific processors to fetch data, taking into account cache delays, which
instructions run faster, and other very specific items.  This is why 3
processors which run the same instruction set and are very similar, the
PII/III, the Pentium Pro, and the Athlon all have different cores.  The
internals are different enough to warrant specific optimization even though
almost no other software bothers due to the hassle and extra code not being
normally worth the performance difference.
I'm not saying you should rush out and buy a P4, but it is silly to say it
sucks before anyone has even tried to optimize the software for it.

-----Original Message-----
From: Ferry van Steen [mailto:td at salesint.com]
Sent: Thursday, May 10, 2001 5:31 AM
To: rc5 at lists.distributed.net
Subject: [RC5] Performance on P-IV

Hey I've got some sad sad sad statistics. I hope there is an explanation for
this because else I have to advise everybody NOT to buy Pentium-IV's because
this is sad.....
Ok.. the distributed net does NOT recognize P-IV's yet so it does a micro
bench and selects Core #2 RG Class 6 which is the same core running on
P-II's. Knowing this you would suspect a performance increase on a P-IV 1400
compared to P-II 400 of 1400/400 = 3.5 which would mean 350% performance
increase in MHz so the pure calculative power should increase 350% also...
My P-II 400 on Core #2 RG Class 6 Win2K with some services runs 1,128,585,13
according to dnetc -benchmark rc5
The P-IV 1400 (1.4GHz) running -benchmark rc5 Core #2 RG  Class 6 only makes
a LOUSY 1,824,190,86 on a freshly installed WinME without any services.
This is only a lousy 161.64% (182419086/112858513*100) increase over a 350%
increase of MHz's
To give another example
(This runs on linux btw, and the tests I've done the rc5 clients run faster
on linux in any case (using the exact same hardware RC5 goes faster under
linux than under win)
My AMD Athlon 1000 (1 GHz) scores a friggin' absolutely awesome 3,538,971.00
Hahaha a 1GHz Athlon is almost TWICE as fast as a stinking P-IV 1400 which
has 400MHz more under ít's belt. This is scary shit... Hehe Guess I
shouldn't buy intel... Atleast not until everything is optimized because
this is really sad.
Hope somebody can explain this....

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.distributed.net/pipermail/rc5/attachments/20010510/9cc3299c/attachment-0002.htm

More information about the rc5 mailing list