Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: AMAZING results. Congratulations for Christophe!

Author: Uri Blass

Date: 15:46:48 04/04/01

Go up one level in this thread


On April 04, 2001 at 17:24:50, Christophe Theron wrote:

>On April 04, 2001 at 15:18:41, Chessfun wrote:
>
>>On April 04, 2001 at 12:52:57, Christophe Theron wrote:
>>
>>>On April 04, 2001 at 10:43:24, stuart taylor wrote:
>>>
>>>>On April 04, 2001 at 09:13:03, Jonas Cohonas wrote:
>>>>
>>>>>On April 04, 2001 at 09:00:51, Vincent Diepeveen wrote:
>>>>>
>>>>>>On April 04, 2001 at 08:54:59, Aloisio Ponti Lopes wrote:
>>>>>>
>>>>>>>Bullet 1 minute/game matches on ERT:
>>>>>>>
>>>>>>>Gambit Tiger 2 x Yace 0.99    76 x 24
>>>>>>>Gambit Tiger 2 x Crafty 18.1  66,5 x 33,5
>>>>>>>Gambit Tiger 2 x LG2000v2.9   90 x 10 !!!!!
>>>>>>>
>>>>>>>A. Ponti
>>>>>>
>>>>>>how about setting both engines to 'instant move'
>>>>>>and then play a few thousands of games?
>>>>>
>>>>>Oh my god
>>>>
>>>>If Kramnik or Kasparov were playing at instant time, or even 1 minute per game,
>>>>I think their level of play would be quite weak.
>>>>  I don't see the real need of concentrating on making programs to play 2-300
>>>>elo points higher on lightning speed than at longer controls, much more so than
>>>>the way humans do it.
>>>>  I don't even see what the great fun is in testing it. It's a little bit
>>>>interesting, but not all that much. I don't think it can be used to demonstrate
>>>>the quality of the program. Maybe the contrary is true, as the longer timings
>>>>are the interesting thing.
>>>>S.Taylor
>>>
>>>
>>>
>>>When you get a new program I think it's quite natural to begin with fast time
>>>controls, so you get an overview on a large number of games in a reasonnable
>>>amount of time.
>>>
>>>Do you think Aloiso is going to stop now? I guess he is going to try longer time
>>>controls.
>>>
>>>Fast time controls are also important because they are commonly used on the
>>>chess servers.
>>>
>>>Another thing to consider is that game in one minute on a PIII-800 computer must
>>>be something like game in 4 hours on the hardware of 10 years ago.
>>>
>>>If 10 years ago you or other chess enthusiasts were interested in games in 4
>>>hours, then I don't understand why you would regard game in 1 minute on today's
>>>computers with such contempt. These games show probably an even better chess
>>>level (because software has been vastly improved).
>>>
>>>And finally, you could complain if the program in question (Tiger) was only
>>>specialized for fast time controls, but as you will see it is not.
>>>
>>>
>>>
>>>    Christophe
>>
>>
>>I'm not sure that you can compare exactly as all programs behave differently
>>especially such fast lightning games. I would think a lot of it depends on
>>pruning especially at such fast speeds.
>>
>>Currently I have been running a 15 engine 1 min 1 sec inc Lightning
>>tournament all play all 80 games each, 8400 games total, this is autoplayed
>>and has been running now for about 10 days and currently Tiger leads Gambit
>>by about 5 with about a 20 point gap till the next engine, Nimzo 7.32 is in
>>last and the Shredder's have yet to play. All programs using their original
>>opening books except crafty which is using general.ctg. All 5 men tablebases
>>to programs that use them.
>>
>>While I agree both Tiger's are also very strong at longer controls currently
>>with the results I have, most matches are being won by say 50-30. At longer
>>controls while my results were very good they weren't this impressive.
>
>
>It can be an effect of the so-called "dimishing returns" in computer chess.
>
>The effect has not been demonstrated yet, so it is still speculation, but many
>people expect more computing power (or time to compute) to give less and less
>strength increase.
>
>If it is true, the consequence is that at longer time controls the strength
>differences between programs tend to be less and less perceptible.

No
This consequence is not result of diminishing returns.

Suppose the following table for program A and B(the numbers are the rating of
programs A and B on different speeds)

speed 1  A 2200 B 2250
speed 2  A 2300 B 2360
speed 4  A 2390 B 2460
speed 8  A 2470 B 2550
speed 16 A 2540 B 2630
speed 32 A 2600 B 2700

This table suggests diminishing return for both A and B when the difference
becomes bigger on faster hardware or at slower 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.