Author: Robert Hyatt
Date: 07:30:16 02/02/04
Go up one level in this thread
On February 01, 2004 at 23:19:16, Matthew Hull wrote: >Bravo! > >As a spectator, the power of Crafty on a big Opteron was palpable, like a shaved >1969 Dodge Charger with a 426 hemi. Even when the position was quite >uncomfortable, you could sense that if there was a defense to be found, Crafty >would find it and brick-wall the attack. This is where one can see that >attention to chess fundamentals in a program really pays off. That is one of >the things a spectator can definitely appreciate about your project. It was a >real pleasure to watch. > >I look forward to the day when Crafty runs on a 32-way. That will be some >impressive mechanical advantage. There are some "impressions" I get when playing other programs. IE with my dual xeon, when playing "deep shredder" I always have the feeling that Shredder knows everything my program knows, but searches deeper. I win some games, but it always feels like I am pushing a bus uphill to make any progress. Against fritz through version 7 (I have not noticed later versions specifically and can't say much there) things didn't feel the same. Sometimes fritz (deep fritz) would out-tactic me, sometimes it would out position me, but most of the time the two programs don't agree on evaluation and in the case of pawn structure, old versions of Fritz through version 7 were wrong more than Crafty was wrong, particularly with pawn majorities and distant passers and the like. The new hiarcs seems to feel a lot like shredder, although in this event I never felt like I was getting pushed around by anybody at all. Crafty did what it wanted, although sometimes it saw its advantage drop back to a draw as the game wore on. I will probably play an 8-way opteron on ICC within the next couple of months, but only at night or on a weekend, and not for very long at a time. Might even see something bigger from time to time as well. :) I played with a 16-way alpha a few years ago, but only for a couple of hours. It was searching about as fast as the quad-opteron (about 7M nodes per second if memory serves going back that far). But the main point for this event is that anytime you play a game, and your eval steadily increases, no matter what your opponent does, it means the underlying search is working well and that coupled with the hardware advantage, it is doing better than the opponent can control. Of course, the book is always an issue and in a couple of games we never really had any winning chances and had to work for a draw... But if you look at the logfiles for the junior game, where it needed to go deep it went _beyond_ deep. I've never seen 16 plies with queens, rooks, minor pieces, open lines, etc, except in forcing cases like QxQ and then KxQ is the only legal move... :) The search never "crumbled" as sometimes happens on my xeon when my opponent has better hardware and a better search too (shredder/hiarcs for example). I suspect Crafty felt like that bus I mentioned before, except it had the brakes on, and a couple of bulldozers behind it pushing downhill. It was _not_ going uphill willingly. :)
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.