Author: Wayne Lowrance
Date: 13:02:47 02/25/06
Go up one level in this thread
On February 25, 2006 at 09:32:15, William Penn wrote: >On February 25, 2006 at 01:37:12, Wayne Lowrance wrote: > >>CC is my interest in Rybka as published little way back. I run my programs >>against each other in infinate analysis. This takes a lot of time to get a idea >>of performance. What I am asking can anyone tell me of testing where Rybka and >>Fritz8 or 9 compare in, say 2 hours per move as for example. >> >>Thanks much >>Wayne > >I haven't heard of any ratings at such long time controls, but it might be >possible with a large number of people cooperating. Maybe it also could be done >wwith a good set of test positions. In the meantime we each try different >engines and then proceed based on instinct. > >I don't have Fritz 9 (no DVD player), but have compared the Rybka betas, Fruit >2.2.1, and Shredder 9. I believe Shredder has been left behind. Rybka or Fruit >are better. > >Fruit behaves better than Rybka and more predictably at long analysis times, >also produces lengthier analysis outputs. Rybka concatenates the moves to 1 pair >(2 half-moves) after several hours in infinite mode in most positions. Also it >is impossible to predict how long it will take between legs with Rybka. I call >them legs - the successive analysis lines output in the engine window. Fruit's >multiplier for successive legs tends to be about 2, and behaves fairly >predictably. Rybka's multiplier is more like 3, but variable, not easy to >predict; it might be 5 or more. For example you might have let the analysis go >for 2 hours, and wonder how long you'll have to wait til the next one decides to >burp itself out into the real world. With Rybka you might have to wait a long >time, 10-15 hours. With Fruit it's a pretty good bet that the next one will show >up in about 4 hours total run time. These are important considerations for CC >players. I'm sure most CC players understand this kind of thinking, others may >not. > >My conclusion at this point is that Rybka is useful at intermediate infinite >analyis times, say 30 minutes to 2 hours. At longer run times it's too >unpredictable and often drives me crazy waiting. So I tend to prefer Fruit if >I'm going to let the analysis run over 2 hours, but I go in cycles. Currently >I'm in an upcycle that favors Rybka, tomorrow it may be Fruit again. Obviously I >don't really know anything, everything is just hunches... :) >WP Very interesting, Thank you Wayne
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.