Author: Ron Langeveld
Date: 10:58:39 06/13/01
Go up one level in this thread
On June 13, 2001 at 06:14:37, José Carlos wrote: >On June 13, 2001 at 06:02:38, Christophe Theron wrote: > >>On June 13, 2001 at 01:14:51, Jouni Uski wrote: >> >>>I run these 3 programs in my test suite, which contains 100 hard, but correct >>>ECM positions. I compared solved positions after 5s, 20s, 1m, 3m and 10 minutes >>>in my AMD 450Mhz (hash 90-128MB). Here's results: >>> >>> 5s 20s 1m 3m 10m >>>Chess Tiger 14 30 49 62 77 84 >>>Goliath Light 17 46 74 84 91 >>>Crafty 18.7 12 30 47 64 82 >>> >>>Here's same as graph: >>> >>> | x >>>90 + >>> | >>> | >>> | x t >>> | c >>>80 + >>> | >>> | t >>> | x >>> | >>>70 + >>> | >>> | >>> | c >>> | t >>>60 + >>> | >>> | >>> | >>> | >>>50 + t >>> | c >>> | x >>> | >>> | >>>40 + >>> | >>> | >>> | >>> | >>>30 + t c >>> | >>> | >>> | >>> | x = Goliath >>>20 + t = Tiger >>> | c = Crafty >>> | x >>> | >>> | c >>>10 + >>> | >>> | >>> | >>> | >>> ---------------------------------------------------------------- >>> 5s 20s 1m 3m 10m >>> >>>Interestingly Crafty gets more positions almost linear. Tiger starts best, but >>>then Goliath goes over. This is no big surprise, when it peaks over 1,4MNPS. >>> >>>Jouni >> >> >> >>A program's NPS is probably one of the worse indicator about anything related to >>playing strength or tactical abilities. >> >>Like saying that a chess program is good because the engine is over 800Kb in >>size. >> >> >> >> Christophe > > The interesting thing of the graph is the shape of the curves. Although the >x-axis scale is not constant (which makes the "Crafty gets more positions almost >linear" statement not correct) the shape of the curves show different strength >increase with time for the three programs. > Of course, you can argue that this is just a test, and doesn't prove anything >itself. And I agree with that. But it will mean something _if_ further tests >give similar results. It has been a while since I had my last mathclass, and frankly, I would have to admit that a refresh course woudn't hurt. But maybe you would like to attend that course as well. Of course the scale is not linear. There's a good reason for this: it reflects 'some' branching factor, the time needed to get an extra ply of depth. A second reason why the scale makes more sense than a linear one is the fact that after having solved 74 positions in one minute the remaining population to solve is only 26 and they are the hard ones. Solving 13 (=50%) more in 3 minutes is actually good performance. In any case, i think the non-linear scale makes the graph easier to interpret. Ron > > José C.
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.