Author: fca
Date: 06:43:39 08/11/98
Go up one level in this thread
On August 11, 1998 at 08:14:32, Robert Hyatt wrote: >On August 11, 1998 at 08:06:36, Tom Kerrigan wrote: > >>Aside from software differences, the Pentium MMX/200 has a 66MHz L2 cache >>(possibly smaller than 512k) whereas the Pentium II/300 has a 512k 150MHz L2 >>cache. If a program really bangs on the L2 cache, it will go much faster on the >>Pentium II. >> >>Cheers, >>Tom >> >>On August 10, 1998 at 22:23:46, fca wrote: >> >>>I am puzzled, Uri. >>>http://www.rebel.nl/bench.htm shows the P2/300 about 50% - 60% faster than the >>>P200MMX for Rebel. >>>Intel benchmark pages ( SPECint (base)95 ) give +80%. >>>While obviously different programs benefit by different amounts by CPU upgrades, >>>your figures show +150% which is a huge difference! >>>Any ideas? > > >Besides Tom's comment about Cache, there are other issues. The original pentium >had a serious design flaw in the pipieline that would let one pipe stall far >more than necessary. This was corrected in the MMX versions. Also the original >pentium had a horrible bug in branch prediction, and would perform poorly on >programs with lots of branches (A chess program typically falls in this category >because of all the if...then...else... cases in the evaluation. The MMX also >fixed this bug. > >So such a big jump on a poorly-behaved program Ah - so Junior is poorly behaved, but Rebel is not? Hmmm.... :-) > is not a surprise. The typical >benchmark program is much better behaved than a chess program, with far fewer >branches, and more regular memory references. Of course. > Chess programs are pretty bad >actors and stress hardware... Please recall the big difference was between +100% per Uri (+150% adjusted down by say 20% because no F5 shell: +150% == 2.5x, 2.5x less 20% = 2x , 2x == + 100%) and the +50% per the Rebel benchmarks. This is not primarily a cpu benchtest vs chess program argument therefore. It is a Rebel vs Junior one! Why such a big difference (if cache is not most of the answer)? this is one reason that it seems that Crafty is >going to be in the new SpecInt98 benchmark... it will be the first 64 bit >integer program and the first non-numerical algorithm that will really stress- >test compiler optimizers... So I can load SpecInt98 and play the Grob? ;-) Kind regards fca
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.