Author: Kevin Strickland
Date: 16:10:53 01/02/02
Go up one level in this thread
>The more rounds you have, the less chances of a fluke win. >A 50% speed penalty *should* be huge. >Each doubling of the CPU count should add 50 ELO or so. > >A machine with 4x MHz ought to win 64% of the points against itself on slower >hardware (all other things being equal). > >That's a pretty significant advantage. > >At some point, and SMP machine will be built that uses 16 or more processors >effectively. That would give 76% of the points from a 200 ELO difference. > >If you could get an effective machine with 64 CPU's, that would give 300 ELO for >85% of the points. There were supposed to be some Alpha machines that were SMP >with 64 CPU's. I am pretty sure that 32 CPU versions were built, but I don't >know much about the quality of the implementations, since they used crossbar >switches. > >Like the guy on home improvement says: >"MORE POWER!" > >I do realize that there is a lot of variability. My assumption is that the top >engines are approximately a push (within 30 ELO or so) so that a doubling of CPU >power will make the doubled engine stronger. Of course, even at that there are >a lot of variables. > >In other words, it is not a lead-pipe synch to win because you are stronger, but >you ought to win. > >The commercial engines are already a bit stronger than the freely available >engines, so they are the favorites in my view. Now, double the available CPU >power, and they should have a large and clear advantage. > >Whether a large advantage translates into a win is another story. > >Consider pyotr (with an ELO of about 1400). The odds it would score 10:0 >against a field of ten opponents with ELO 2400 in round robin is about: >.003^10 = 0.000000000000000000000000059049 >Which amounts to one in 16,935,087,808,430,286,711,036,597 trials. > >Not terribly likely but if you ran enough trials it would happen. And (by >random fluxuation) it could possibly be the very first trial. It's just not >terribly likely to happen. > >Now, with engines that are fairly close, there is a lot of randomness. Even at >that, the very best engines are rather head-and-shoulders above the middle of >the road engines and so we usually see something fairly close to what we expect. I understand what you are saying, but at the same time there are much more valuable variables other than cpu speed here. I have seen a series of games between two Crafty clones. One was 2x faster than the opponent. The slower machine actually won the series of over 200 blitz games at 5 0 at a 71% winning percentage just due to the opening I believe. The slower Crafty easily had the better book. It was more diverse, and well learned. I think that could have a more impacting factor than the cpu speed used. If a commercial engine was to play the tournament with the book that is publically released, I think it would hurt it's chances by 50% as everyone have played and trained vs that book and can probably reduce the advantage that Chessbase books give the program. There is no question that the Chessbase books are very strong, if not the strongest out there. It has been observed many times in the matches of the SSDF. The Chessbase programs that use the books that are offered with the products are very good and are a class above everyone else. Crafty used to have and still does have an advantage playing the Silician. I think out of all the programs I have and watch play on the various servers I think it plays that opening that the variants of it the best. I think that is due to the opening book. I created a Crafty opening book once that was darn near impossible to beat it out of book. I hand editted the start.pgn off of Bob's server, and it took months of analyzing positions to find the best next move to put into it. Once I was satisfied that it was a decent book I tried it out on FICS and attained a 200+ rating over the 2300 I currently had on slow hardware. The best Crafty book I have ever seen was by the person that ran the account RoboElvis(C) on FICS. It was probably the strongest Crafty to ever play on FICS. It crushed all comers for the longest time until the operator switched to a non-alpha processor. I think the open books that are going to be used are going to impact play more than the cpu's being used.
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.