Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crosstable

Author: Robert Hyatt

Date: 14:31:35 08/16/05

Go up one level in this thread


On August 16, 2005 at 16:44:26, Uri Blass wrote:

>On August 16, 2005 at 16:33:36, Robert Hyatt wrote:
>
>>On August 16, 2005 at 16:14:37, Uri Blass wrote:
>>
>>>On August 16, 2005 at 16:05:39, Ingo Althofer wrote:
>>>
>>>>On August 16, 2005 at 15:57:30, Gerd Isenberg wrote:
>>>>>                      1    2    3    4    5    6    7    8
>>>>>1. Shredder    6      x    1    1    0    1    1    1    1
>>>>>2. Zappa       5      0    x    1    0    1    1    1    1
>>>>>3. Junior      4.5    0    0    x    1    1    1/2  1    1
>>>>>4. Fruit       4.5    1    1    0    x    1    1    0    1/2
>>>>>5. Crafty      2.5    0    0    0    0    x    1    1/2  1
>>>>>6. Diep        2.5    0    0    1/2  0    0    x    1    1
>>>>>7. Jonny       1.5    0    0    0    1    1/2  0    x    0
>>>>>8. Deep Sjeng  1.5    0    0    0    1/2  0    0    1    x
>>>>
>>>>Hello Gerd,
>>>>
>>>>thanks for the full information.
>>>>
>>>>Congratulations to Stefan for winning the Blitz title.
>>>>Good luck for you for the remaining round of the WCCC!
>>>>
>>>>Ingo.
>>>
>>>congratulation for stefan and congratulation for fruit for scoring more than
>>>Crafty and showing that number of processors is not the only important thing.
>>>
>>>It seems that fruit one processor is not weaker than Crafty 8 processors based
>>>on the results.
>>>
>>>I wonder if there were time losses because of bad operators.
>>>
>>>Uri
>>
>>
>>Your logic is sometimes horrible.  Blitz is not normal chess.  If you think it
>>is, you are _sadly_ mistaken.  I watched IM Mike Valvo give 1-5 time odds to GM
>>players and beat 'em at blitz.  But not OTB in standard time controls...
>
>I understood from another post that you say that Crafty is not tuned for blitz
>because of some reasons.
>
>I understand the reason of the parallel search but I still need to see example
>for significant different results between blitz and long time control for
>programs without bugs.
>
>I know that usually when program improve they improve in all time controls.
>I do not know of evaluation changes or search changes that make programs weaker
>at blitz but stronger at long time control.

Quite simple.  You can extend more in blitz, to avoid tactical mistakes.  But if
you double the search depth, those extensions can add too much overhead.

But that is irrelevant here.  It is about the SMP search tuning, and the minimum
search depth from the tips where I allow splits to be done.  What if that
minimum depth is greater than the blitz search depth for quick searches?

You can believe what you want.  I personally don't have time to drag up examples
of where tuning for blitz is bad for long time controls and vice-versa.  But it
is an absolute fact that it happens.  Perhaps not to all programs, I don't know
about that.  But for parallel search, at least for one with a reasonable level
of performance, tuning makes all the difference.  I ran for _days_ on the
opteron, testing different settings over dozens of positions, to find the
optimal average...



>
>In thoery it can happen but I need to see a proof for it and I believe that
>fabien mainly test in blitz time control(he can correct me if I am wrong)
>because usually productive changes in blitz of adding knowledge to the
>evaluation are also productive at long time control.
>
>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.