Author: Sune Fischer
Date: 02:05:05 04/17/02
Go up one level in this thread
On April 17, 2002 at 03:17:50, David Dory wrote: >>I don't mean to insult you, but you have never written an engine, have you? >You don't insult me. Not relevant. My chess program is slowly taking shape, >though. Since I learned to program in DOS, I find the Windows event-driven >environment difficult to program in. "event-driven"? Hmm, sounds complicated, you are not using MFC or something fancy like that? I think most just write it in portable C/C++, just let Winboard or Arena deal with all that stuff. :) As a start I would suggest to browse through a few codes for the simpler programs, TSCP springs to mind, but later take a look at the rest of them gnuchess, beowulf, phalanx, crafty, faile, pierre, gerbil (who did I forget?). I like to study how they all solve the same problems in different ways, you can get lots of ideas. Funny actually that they are not close in strength, perhaps someone could explain to me why for instance Beowulf is 300 elo weaker than Crafty..? >>I don't believe they are at the hardware limit just yet, it may come at one >>point. AFAIK Chris said Tiger had improved greatly since he started working with >>Ed (was it?). I can't find the interview just now... >> > >Yes, he said Ed had helped improve CT. So here's another chess programmer, >saying his program is stronger for the reporters. > >And I believe him. > >But it ain't much stronger. The tigers are up there now fighting with the Fritz, it wasn't always like that I think. >>I'm pretty sure software is a pretty big factor here, DAVID! >> >When you test, you use STANDARDS THAT DON'T CHANGE. You know as well as I do >that elo is a moving target, it depends upon the opponents you play, (and of >course, their hardware, and other factors, all of which are changing). The figures came form the same chart, so I think whatever they did, it was the same scale. >In my other post, I put up an UNCHANGING standard, year by year, noting the >tremendous improvement in hardware. > >Your "How many years", and elo figures above (without even the source being >noted), lacking even a time frame, pale in comparison to Jonathan Schaeffer's >FACTS. More speed mean more nodes per second, but there is the issue of the quality of the evaluation, in principle if you had a perfect eval you'd only needed to search the root to 1 ply. Perhaps Shredder would have lived longer with a better king safety eval, I don't think 10x more nodes would have saved him, the attack was that deep. Then there is the matter of how effective you can prune the tree. Improve both of those, and it is equal to a huge amount of nps. The best indicator of this, is that most programs today run within a factor of 10 of each other in speed, some about 1 Mnps and some only 100 knps, but there is very little correlation to their relative strengths. As I mentioned above, why is Crafty so much stronger than Beowulf, in fact why isn't all programs just as strong or stronger than Crafty, Crafty is opensource - no secrets. I think it is because there are a lot of subtleties in the search that needs to be understood, and doing this requires time and effort. >I thought you would really respect his documentation, since he is a chess >programming pioneer, with PHOENIX, a PH.D. and researcher in Computer Science, >and a very good chess player OTB. I 'respect' it, but I'm just hesitant to conclude anything from node numbers alone, after all we didn't have any raw data for the improvement on SOFTWARE so what's to conclude, really? >Their is no absolute data for chess software improvement over the past 20 years >that I know of. I mean real data, against standards, not a relative thing like >elo. > >The reason is simple - there has been precious little progress made in chess >software. I'd love to re-compile the old 1970's SARGON chess program for 32 bit >CPU's, and watch it in a tournament, or Lang's old 6502 REBEL program, or >Mephisto, or how about the Spracklin's Fidelity program, that'd be great. > >They wouldn't win, because they wouldn't use the new HARDWARE component >effectively, but they'd be fun. I think what Uri suggested would be interesting, but I don't have the old hardware or SOFTWARE anymore :( >>Oh, so you meant NOTHING but NONSENSE with those numbers, why post them then, >>DAVID? > >I meant just what I said. I compared some old computer hardware, with some new >computer hardware. I was NOT comparing chess program strength in that paragraph, >and I stated that clearly at the time, and repeatedly since then. Hmm, I think the issue was about the improvement in _chess strength_, sure hardware has come a long way, no argument there. >I thought the numbers were interesting. Of course, they are verifiable, and >quantifiable facts, and that doesn't seem to be something you tolerate very >well. But honestly, that factor of 15000 for the harddrives were quite funny ;) >>(oh no, I've been infected with the troll caps-lock disease, must go see doctor >>soon...) > >Perhaps the good doc can prescribe something for your allergy to Schaeffer's >standardized test results, which document the tremendous improvement in HARDWARE >to his chess program's performance. Whether you accept it or act cartoony, is >your choice. I will act cartoony and completely deny any improvement in hardware over the last 20 years, it is all a hype, "believe me", "take my word for it" ;) -S.
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.