Author: Joseph Ciarrochi
Date: 17:42:25 01/17/06
Go up one level in this thread
thanks for your excellent response Dan. Please excuse my simple questions. I get the gist of what you are saying but... When you talk about branching factor, do you mean how broadly the program searches. e.g., does it simultanously search two lines if it has a branching factor of 2. What does it do with 1.5 branching factor. Also, concerning the 64 bit question, i was thinking more of relative boost to strength. so is it possible for a new search to increase the *difference* between the 32 two bit and the 64 bit, even if they both have the same new search. Based on your answer to my first question, it sounds like the answer is "yes" to this question as well. I think? niavely yours, Joseph On January 17, 2006 at 20:10:00, Dann Corbit wrote: >On January 17, 2006 at 19:34:27, Joseph Ciarrochi wrote: > >>The CEGT list suggests that just maybe Rybka has made a big performance increase >>(but we really don't know yet of course). >> >>Still, i was wondering. >> >>In terms of practical results, is it possible for changes in a search algorithim >>to only be observable at longer time controls. Or should you observe differences >>in both blitz and long time controls. I guess what i'm asking is, "what sort of >>change in a search algorithim would only become apparent at longer time >>controls." > >Sure. > >Consider sorting as a model. For small data sets, the O(n^2) sorts like >insertion sort or shell sort are fastest. So, if I have 20 things to sort, >insertion sort will beat introspective sort. > >But if I have millions of things, then introspective sort is going to clobber >insertion sort because it is O(n*log(n)) and shell sort is O(n^2) > >In a similar vein, suppose that I invent a search technique that has a branching >factor of 1.5. Now, the problem with my search technique is that it takes 20 >minutes to gather data. So, if we are playing 40/2hrs, you are going to kill me >with your 2.0 branching factor program. But if we are playing correspondence >chess, I will outsearch you by 40 plies and you won't have a chance (depending >on whether I am looking at the right nodes and throwing out the bad ones, of >course). > >>Also, is it possible that some search features will work better in 64 bit >>compared to 32 bit machines and engines? > >Why wait until the future? Rybka 64 bit outperforms Rybka 32 bit by a large >margin, and (I am fairly sure) that it is exactly the same code. > >Any 64 bit operations are going to benefit from 64 bit CPUs.
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.