Author: Dann Corbit
Date: 17:05:55 11/13/02
Go up one level in this thread
On November 13, 2002 at 19:52:57, Russell Reagan wrote: [snip] >Are chess programs different? By many magnitudes in some cases? Yes, but are >they revolutionarily different? I don't think so. I think all of the top >programs are in the same generation. I don't think any top program has made any >breakthrough of the significance that every other chess program suddenly becomes >a previous generation chess program (I think this is what Bob D. meant by saying >all chess programs use basically the same algorithms). What do you think? Some programs are revolutionarily different. It takes an exponential increase in ability to gain a small increase in playing strength. For instance, one additional ply is not going to buy you much going from (for instance) 10 to 11 plies in a search. And yet some programs are 1500 ELO stronger than others. That's not a minor difference. The search implementation itself can be a source of knowledge (Christophe Theron has had several interesting discussions on this idea). Phalanx is a good example of a program where an extra-clever search nets a great deal of power. Ruffian is also an engine where the search part of the algorithm was given extra attention. It is also possible to make the evaluation much smarter. Pruning and extension ideas to look at more/less things down certain paths can have a huge effect. The difference between two engines that have an ELO difference of just 200 points is remarkable. That means that you could put the less powerful engine on a 1GHz machine and the faster one on a 200 MHz machine and they would have about the same performance. Now, imagine a 500 ELO difference. How about 1000? To achieve this kind of performance (Ruffian, ChessMaster, Chess Tiger, etc.) *requires* innovation. Just writing good, clear code is not going to put you in that elite class.
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.