Author: Robert Hyatt
Date: 18:48:57 01/25/04
Go up one level in this thread
On January 25, 2004 at 18:09:03, Mike S. wrote: >A poster at the CSS Forum has asked, if there are results available of newer >Crafty versions versus the most recent version of Cray Blitz: nope. First, there is no "recent version" of Cray Blitz. It's last real tournament was the ACM event in 1994 and there was no development on it beyond that point as that marked the beginning of the Crafty project that replaced CB. I played a 10 game match a couple of years ago using my quad xeon vs a Cray T90 that I had access to, and the Cray pretty well had its way with Crafty. I think the final result was 7-3 for CB. I did not save the games as I didn't really consider that to be too interesting. > >http://f23.parsimony.net/forum50826/messages/90559.htm >(german) > >(I have added the assumption that there is probably no Cray Blitz running >anywhere, anymore. But of course I'm not sure.) No idea, but I doubt it. It used to be in the Cray user's library, but that was 10 years ago. The problem of course is that the machines sold for around $60 million dollars, so they were not very easy to get access to for playing chess, the problem that led to the Crafty project in fact. > >Please Prof. Hyatt if you read this and if you have an answer, you could either >post a reply directly after the message linked above (requires no password) or >answer here, and I'll forward it. > >(I guess the poster is interested in similarities between Cray Blitz and >Crafty.... the history of Cray Blitz is well known among german computerchess >fans because there were reports about it in the eighties.) A couple of notes. 1. Cray Blitz used null-move, but R=1 and non-recursive. IE there could be no more than one null-move search in any distinct path from root to tip, where today Crafty uses R=2 or 3 depending on the depth remaining, and multiple null-move searches can be done in the same path, although two null-moves can not be done in a row. 2. Cray Blitz had a reasonable singular extension implementation, Crafty does not. 3. Crafty has some endgame knowledge that CB did not, but then Cray Blitz did some king safety evaluation things that Crafty does not due to computational costs that were not as high on a vector architecture. 4. CB was clearly a 2500-level player as we played multiple exhibition games against GM players and always did well. Crafty is probably capable of doing at least that well against the same players, perhaps. I'm not sure what else to say. CB on a T90 with 32 cpus could hit around 7M nodes per second. Crafty on the quad opteron I am using for CCT-6 (running on ICC right now in fact) is searching around 7-8M nodes per second, with an occasional burst to 10M+. Either would be a handful. :) > >Thanks, >Mike Scheidl
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.