[RC5] Differing processor speeds
b.ford at qut.edu.au
Fri Jul 17 10:24:51 EDT 1998
> Date: Wed, 15 Jul 1998 10:15:38 -0600
> From: Steve Bird <sbird at UATC.com>
> Subject: [RC5] Differing processor speeds
> Ok, here's my set-up.
> At home I have 2 machines....a P200MMX and a P2-266, both of them are
> Intel-based processors. When I benchmark both machines I get
> 1.9M/keys/second on the P200MMX and 2.15M/keys/second on the P2->
> 266. Both of these figures are from the DES benchmarking section of the
> My questions are:
> 1) Are these results unusual?
The results are not unusual. I used Pentium pairing rules when creating the
MMX sboxes (which by the way are at
http://www.tals.dis.qut.edu.au/staff/bruce/des-ii.htm ). The need to do alot of
op mmreg,mem instructions appears to affect the PII's performance, probably
because they generate 2 micro-ops.
There was no time to develop specialized x86 processor cores as are
present for RC5.
> 2)If so, is the P200MMX score too high or is the P2-266 score too low?
> 3) In almost every other application I have (ranging from games to the
> RC5-64 - based benchmark), the P2-266 outshines the P200MMX to the
> tune of about 2.5:1 in raw performance(actually 3:1 in RC5-64), Why isn't
> this same level of performance represented in the DES benchmark. To
> me, it almost seems as if the results would be consistent with a
> (hypothetical) Pentium 266(ie...the extra instructions of the P2 aren't even
> used). So whats up with this?
See above. Also, in RC5, the rotate left instuction is used heavily. On the
Pentium the rol reg,cl takes 4 clocks non-pairable(don't believe what the Intel
docs say), on a PII - 1 clock pairable. It makes a huge difference.
Maybe by DES-II-3 we will have PII specific bitslicing MMX cores.
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest
More information about the rc5