Re pondering

Discussion of computer chess matches and engine tournaments.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
bezzy3004

Re pondering

Post by bezzy3004 » Fri Jan 23, 2009 6:22 pm

I am the original poster of this topic and am sorry i have started a feud! I only wanted to know how to obtain the best quality chess possible. ponder on or off simple.

User avatar
kranium
Posts: 1824
Joined: Thu May 29, 2008 8:43 am

Re: Re pondering

Post by kranium » Fri Jan 23, 2009 6:31 pm

i think all will agree that an engine is stronger if it can ponder between moves...

User avatar
Matthias Gemuh
Posts: 3238
Joined: Thu Mar 09, 2006 8:10 am
Contact:

Re: Re pondering

Post by Matthias Gemuh » Fri Jan 23, 2009 7:00 pm

kranium wrote:i think all will agree that an engine is stronger if it can ponder between moves...

This is only true if some extra CPU is used that would otherwised not be used.

If the same amount of CPU is used in both cases, an engine is WEAKER if pondering is ON. Remember that each ponder miss is a waste of CPU.
Hash tables may help to reduce the waste but cannot eliminate it.

Matthias.
My engine was quite strong till I added knowledge to it.
http://www.chess.hylogic.de

User avatar
Graham Banks
Posts: 33236
Joined: Sun Feb 26, 2006 9:52 am
Location: Auckland, NZ

Re: Re pondering

Post by Graham Banks » Fri Jan 23, 2009 7:13 pm

bezzy3004 wrote:I am the original poster of this topic and am sorry i have started a feud! I only wanted to know how to obtain the best quality chess possible. ponder on or off simple.
As you can see, there are arguments both ways. :wink:
Whereas I do all my testing with ponder off in order to meet CCRL criteria, I think you'll have more fun watching a ponder on tournament in progress.
In the past, I recall seeing some testing that was done to measure how much stronger engines played with ponder on as opposed to ponder off. The difference was negligible.

Cheers, Graham.
My email addresses:
gbanksnz at gmail.com
gbanksnz at yahoo.co.nz

User avatar
hgm
Posts: 23773
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: Re pondering

Post by hgm » Fri Jan 23, 2009 7:17 pm

I still have little to add to what I originally said:

If you want to play a number of games on a dual, play two games simultaneous with ponder off, with twice longer time control. The engines will then play stronger than when you play one game at the time at the nominal TC with ponder on. In both cases you use 100% of the CPU, but in the former case 20-30% of the CPU time will be wasted because engines are pondering about the wrong move.

The other arguments that have been mentioned, like "butthen engines take tuns in thinking, in stead of thinking all the time" you should only pay attention to if you care abotthem and are willing to sacrifice quality for it or wait longer for your results.

User avatar
hgm
Posts: 23773
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: Re pondering

Post by hgm » Fri Jan 23, 2009 7:22 pm

Graham Banks wrote:In the past, I recall seeing some testing that was done to measure how much stronger engines played with ponder on as opposed to ponder off. The difference was negligible.
This is interesting, and certainly true for micro-Max (as it does not know how to ponder :lol: ). But I would have expected a 20-Elo improvement, or so, for engines with a good ponder implementation.

User avatar
Graham Banks
Posts: 33236
Joined: Sun Feb 26, 2006 9:52 am
Location: Auckland, NZ

Re: Re pondering

Post by Graham Banks » Fri Jan 23, 2009 7:29 pm

hgm wrote:
Graham Banks wrote:In the past, I recall seeing some testing that was done to measure how much stronger engines played with ponder on as opposed to ponder off. The difference was negligible.
This is interesting, and certainly true for micro-Max (as it does not know how to ponder :lol: ). But I would have expected a 20-Elo improvement, or so, for engines with a good ponder implementation.
20 ELO is negligible in the bigger scheme of things. Just my opinion.
My email addresses:
gbanksnz at gmail.com
gbanksnz at yahoo.co.nz

mcostalba
Posts: 2684
Joined: Sat Jun 14, 2008 7:17 pm

Re: Re pondering

Post by mcostalba » Fri Jan 23, 2009 8:10 pm

Graham Banks wrote: 20 ELO is negligible in the bigger scheme of things. Just my opinion.

I can assure you that when you are fighting hard with your code just to squeeze that extra bit of speed or tweaking a parameter to death just to find the optimal tuning you are very very far from the bigger scheme of things in that moment :D :D

User avatar
Matthias Gemuh
Posts: 3238
Joined: Thu Mar 09, 2006 8:10 am
Contact:

Re: Re pondering

Post by Matthias Gemuh » Fri Jan 23, 2009 8:57 pm

hgm wrote:... I would have expected a 20-Elo improvement, or so, for engines with a good ponder implementation.

Such a huge gain is only possible if some opponents ponder so badly that they lose that much Elo approx.

Graham is right that games with ponder ON are more fun to watch, though.

Matthias.
My engine was quite strong till I added knowledge to it.
http://www.chess.hylogic.de

User avatar
Graham Banks
Posts: 33236
Joined: Sun Feb 26, 2006 9:52 am
Location: Auckland, NZ

Re: Re pondering

Post by Graham Banks » Fri Jan 23, 2009 9:00 pm

mcostalba wrote:
Graham Banks wrote: 20 ELO is negligible in the bigger scheme of things. Just my opinion.

I can assure you that when you are fighting hard with your code just to squeeze that extra bit of speed or tweaking a parameter to death just to find the optimal tuning you are very very far from the bigger scheme of things in that moment :D :D
True. I certainly didn't mean to downplay the efforts of the engine authors in this regard. 8-)
My email addresses:
gbanksnz at gmail.com
gbanksnz at yahoo.co.nz

Post Reply