Author: Vincent Diepeveen
Date: 06:24:06 06/24/04
Go up one level in this thread
On June 23, 2004 at 22:50:38, Mike Byrne wrote: >Do we have now have machines on desktops equal to Deep Blue in strength as Don >Daily suggested may happen in 10 years or less back in 1999? > >http://chessprogramming.org/cccsearch/ccc.php?art_id=44793 > >Let's hear your opinions! Deep Blue searched 10-12 ply using forward pruning last 4 ply and using singular extensions first 9 ply. Because it used a very inefficient fullwidth search and even more inefficient parallel search. Further its evaluation function was like gnuchess 4.0 with a few bugfixes here and there. In short deep blue is a very bad example to compare. It's like DIEP on a supercomputer. I won 2 ply by the supercomputer compared to my dual k7. Now if you consider the dual k7 i have is a very efficient processor which is clocked at 2.127Ghz and that each supercomputer cpu is clocked at just 500Mhz and has less L1 cache than even an itanium2 and the L2 cache is even off chip, then you *start* to realize that such huge hardware gives you only an advantage (or disadvantage) at the time you use it. So you should see it that deep blue was a 1997 program which used hardware to get +2 ply over a software program at a dual pentiumpro 200Mhz in those days. That's how you should see deep blue at those days. That's how diep was in november 2003 too. I just took the +2 ply (and all the burocratic problems) that's all. There is never a good compare to the future because the software progresses so much. Where i am sure that in 1997 world champs it would have been butchered too (it was a microevent though), it's clear that by 1999 Deep Blue would have been butchered clear for even outsiders. 10-12 ply searches were by then also for programs using a lot of extensions very normal. They were *not* in 1997. The only ones getting 10-12 ply back then were programs either doing shit in qsearch or forward pruning (that includes crafty, bob has perhaps forgotten he was forward pruning in those days, and was doing a 1 pawn futility in qsearch and had zero extensions, not even solving a mate in 6 like WAC141 which diep 1995 already solved instantly at a 486; so bob claiming he searched 10-12 ply back then doesn't count). Let's face it. 4000 moves openingsbook it had; handmade, the rest was randomly generated book moves from games, which as proven by 1999 was a suicidal approach. So against the commercial opponents in 1999 it would have been butchered in book for sure. In 1995 at least it had a chance, but blundered itself in opening (O-O?? against fritz for example). There are more horrible games from it there. They are really beginners level. In those days i could win any rook endgame from such programs, against crafty even with 2 pawns down. In 2004, forget it if you have a bit less space against 'em ;)
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.