Author: blass uri
Date: 07:04:44 06/01/99
Go up one level in this thread
On June 01, 1999 at 08:59:46, Andrew Dados wrote: > It's interesting to see how many people here oppose SSDF testing because of >speed advantage, say 3 times, between hardware (P90 vs P200) leading to match >scores like 14:4 (more or less - threads on subject emerge regularly) and yet >refuse to estimate thousandfold speed difference impact 'correctly'.... Common >sense can be misleading... This speed advantage leads to difference of less than 100 elo and 14:4 is difference of more than 200 elo. The expected result between p200 and p90 is not 14:4 but something like 11:7 or 10.5:7.5 We cannot learn from these results to estimate thousandfold speed difference impact 'correctly' because the difference in rating may go down. Some positions after the opening may be practically solved and the result in part of the cases will stop to be function of the speed of the computer but only function of the opening choice. This happens even at low speeds(not often) when top programs play repetition lines after they get out of book. Junior5(P200) drew against Comet32(P90) in the SSDF games and doing the hardware 1000 times faster cannot help to avoid these draws. Fritz5 also did a draw by repetition and in this case it was even worse because the repetition was part of the book and Fritz5 engine did not play. You need to improve the programs to avoid these draws. Uri
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.