Author: Enrique Irazoqui
Date: 15:20:58 02/28/98
Go up one level in this thread
On February 28, 1998 at 16:58:44, Thorsten Czub wrote: >>>64 MB hash catapults Fritz maybe into the tops. >> >>44. > >You get 44 if you have 64 MB RAM. Sorry. I mixed up RAM and hash in >mind. > >> >>>The book and the learning help even more. >> >>I think the book hurts F5 more than helps. It's not tuned. And in any >>case, having a book and a learner is not so terribly unfair, isn't it. >>They all have them. > >Yes. But they have implemented it different. And I don't think that: >"They all have them" >gives any information that helps us to understand the results. You mentioned the book and the learner, Thorsten. I only replied to you. Point is: they all have a book, they all have a learner; lets take a look at what is different, since you disagree with the testing procedure. >>>The 3 times speed give Fritz the 60 ELO points ahead ! >> >>Nonsense. F5 played matches on P200MMX vs. P200MMX and also on P200MMX >>vs. P90s, just like Rebel 9, Mchess 7, Nimzo98... Same method for all. >>It's the average what counts. > >The question is - for me - if this is really a linear and turnable thing >! >I doubt if the things behave identical if you do the same thing with >different programs. Couldn't it be that some programs profit more from >the same misbehaviours ? Speculation, speculation... But let's accept it. Some will benefit more than others from the increase in speed. We take a look at the SSDF list and then we find that Fritz programs benefit from faster processors at least as much as the slow searchers. This always puzzled me. But it's there... I know of no data that proves your point regarding Fritz 5. Enrique >If they test Hiarcs6 200 Mhz vs. Fritz5 P90 >and Fritz5 200 Mhz vs. Hiarcs6 p90, do the results really SHOW the right >thing or could it be that hiarcs cannot WIN as much games with the >faster hardware (due to the fact it is a slow program and doesn't profit >much from hash and cpu) >than fritz can win with the faster hardware because fritz comes deeper >even on similar hardware ! >I have problems to explain this in english because I don't know the name >for the word Aequivalent ( mathematical sign <=> ). >If hiarcs6 is not aequivalent to fritz5 than doing the same with 2 >programs does not proof that it produces exact results. > >>>The big ahead-advantage of fritz is not real. It is just SPECULATIVE in >>>the moment due to a few parameters. >>>If the others react and change these parameters, the advantage will >>>decrease. >>>Believe me. >> >>Could be. As I said before, this 55 points lead took me by surprise too. >>I expected Fritz 5 to be more or less even with Nimzo98, Rebel 9 and >>Hiarcs 6. But then, SSDF played twice as many games than I did. And I >>absolutely trust them. > >Right. This was my idea too. Fritz5 with all best chances and conditions >(= big hash) comes into the top ! >But not leading them with 50 ELO ! > > > >>My point is not only the strength of Fritz 5 but the criticism about >>SSDF procedures when they have been a constant over the years, not a >>variable. >> >>Enrique > >Okok. I understand your point. But still I think our main problem is >that we CANNOT relate measuring human elo with measuring computer elo. >There are mechnaisms in computerchess (because computerchess has so >strange axioms/settings) that could lead us to false conclusions >although the maths is right.
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.