Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: to sarah: Re: any difference in reults between 2 and 1 computer

Author: Joachim Rang

Date: 11:25:26 07/24/02

Go up one level in this thread


On July 24, 2002 at 13:41:23, Rajen Gupta wrote:

>On July 24, 2002 at 13:06:11, Timothy J. Frohlick wrote:
>
>>Rajen,
>>
>>If both programs were given the same computers with ponder-on there would be
>>essentially no change in performance relatively.
>
>this is the question that i was asking ie relative performance wih respect to
>each other. i entirely agree that the quality of moves would be certainly better
>for each programme when played on 2 computers with ponder on. i just wondered
>whether that would affect relative ratings. i have read some postings where it
>was suggested-bob hyatt i think once mentioned that crafty should alway be
>tested against other programs using 2 computers with ponder on. he even stated
>that playing crafty on a single machine against other programs with ponder off
>was meaningless as that disproportionately reduced the strength of crafty!
>
>i wonder if tht is true as i feel that the relative ratings should be
>approximately the same keeping in mind the usual statistical error.
>
>maybe sarah can tell me as she has played a lot of such matches.
>
>rajen
>
>>The moves would probably be better.  The only way to determine any difference in performance is to play
>>Program A with ponder-on against itself on an identical computer with
>>ponder-off.  I would think that a 30 to 75 ELO point improvement would be seen
>>on the program running with ponder-on since it would be thinking almost twice as
>>long.


I don't think so: The rate of right "ponder-guesses" is not often above 50 %, so
you gain only 50 % time benefit.

>>
>>The more expensive way of doing this would be to pay a 2600 level GM to play
>>games against your program with both on/off ponder.
>>
>>Tim Frohlick
>>
>>PS I think that Sarah Bird may have done this experiment.
>>
>>
>>
>>On July 24, 2002 at 12:28:21, Rajen Gupta wrote:
>>
>>>would there be a significant difference in overall results, if 2 programmes were
>>>to play against one another on 2 computers with ponder on as opposed to playing
>>>the same programmes, engine vs engine with  similar time controls n a single
>>>processor computer?
>>>
>>>and are there any programmes which do significantly better, ratings wise, when
>>>played with ponder on vs ponder off?
>>>
>>>rajen

There is an experiment with 4 Engines, but with insufficient games.
Unfortunately it is only in german, nevertheless I will post the URL:

http://www.pcschach.de/Experimente/Experiment3/experiment3.html

According to the results it seems, that CT 14 suffers most from ponder=off and
Deep Fritz gains most.



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.