Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Doubling of thinking time, greater benefits for dedicated computers?

Author: Dann Corbit

Date: 19:44:28 11/20/03

Go up one level in this thread


On November 20, 2003 at 22:41:09, Dann Corbit wrote:

>On November 20, 2003 at 22:27:14, Dann Corbit wrote:
>
>>On November 20, 2003 at 22:21:46, Dann Corbit wrote:
>>
>>>On November 20, 2003 at 19:50:12, Uri Blass wrote:
>>>
>>>>On November 20, 2003 at 18:57:24, J. C. Boco wrote:
>>>>
>>>>>Doubling the processor speed or the time for thinking adds about 80 points to
>>>>>the SSDF-computer-ratings.  But PC's are already running so fast that doubling
>>>>>their speed might mean searching at 15ply instead of 14ply.
>>>>
>>>>How do you get it?
>>>>
>>>>Based on the ssdf list even being 3 times faster does not give 80 elo.
>>>>A1200 is often more than 3 times faster than K6-450
>>>>
>>>>Uri
>>>
>>>This query:
>>>SELECT a.engine, (a.Elo - b.Elo)/(1200.0/450.0) AS EloDiff
>>>FROM Ssdf AS a, Ssdf AS b
>>>WHERE a.Engine = b.Engine
>>>
>>>And
>>>   a.Hardware like "*1200 MHz*"
>>>
>>>And
>>>   b.Hardware like "*450 Mhz*"
>>>
>>>Order by  (a.Elo - b.Elo)/(1200.0/450.0) Desc
>>>
>>>Returns this result set:
>>>
>>>engine	EloDiff
>>>Gandalf 4.32h	50.625
>>>Crafty 18.12/CB	50.25
>>>Gandalf 5.0	42
>>>Fritz 7.0	42
>>>Deep Fritz 7.0	41.25
>>>Rebel Century 4.0	40.125
>>>Hiarcs 8.0	36.75
>>>Chess Tiger 14.0 CB	32.25
>>>Gambit Tiger 2.0	27.375
>>>Deep Fritz	25.125
>>>Junior 7.0	22.875
>>>Chess Tiger 15.0	20.625
>>>Shredder 5.32	20.25
>>>
>>>And this query:
>>>
>>>SELECT sum((a.Elo-b.Elo)/(1200/450))/count((a.Elo-b.Elo)/(1200/450)) AS
>>>EloDiffAverage
>>>FROM Ssdf AS a, Ssdf AS b
>>>WHERE a.Engine=b.Engine And a.Hardware Like "*1200 MHz*" And b.Hardware Like
>>>"*450 Mhz*";
>>>
>>>Gives this result:
>>>34.7307692308
>>>
>>>~35 ELO per doubling seems like what we might expect.  Give or take a large
>>>error bar.
>>
>>This query:
>>SELECT sum((a.Elo-b.Elo)/(450/200))/count((a.Elo-b.Elo)/(450/200)) AS
>>EloDiffAverage
>>FROM Ssdf AS a, Ssdf AS b
>>WHERE a.Engine=b.Engine And a.Hardware Like "*450 MHz*" And b.Hardware Like
>>"*200 MMX*";
>>
>>Gives this result:
>>32.0888888889
>>
>>This query:
>>
>>SELECT sum((a.Elo-b.Elo)/(200/90))/count((a.Elo-b.Elo)/(200/90)) AS
>>EloDiffAverage
>>FROM Ssdf AS a, Ssdf AS b
>>WHERE a.Engine=b.Engine And a.Hardware Like "*200 MMX*" And b.Hardware Like "*90
>>Mhz*";
>>
>>Gives this result:
>>36.18
>>
>>Looks like it holds pretty steady.
>
>Now this query:
>SELECT sum((a.Elo-b.Elo)/(90/66))/count((a.Elo-b.Elo)/(90/66)) AS EloDiffAverage
>FROM Ssdf AS a, Ssdf AS b
>WHERE a.Engine=b.Engine And a.Hardware Like "*90 MHz*" And b.Hardware Like "*66
>Mhz*";
>
>Gives this result:
>51.3333333333
>
>(perhaps transition from 486 to Pentium was more dramatic)

How about this one:
SELECT sum((a.Elo-b.Elo)/(66/33))/count((a.Elo-b.Elo)/(66/33)) AS EloDiffAverage
FROM Ssdf AS a, Ssdf AS b
WHERE a.Engine=b.Engine And a.Hardware Like "*66 MHz*" And b.Hardware Like "*33
Mhz*";

Gives this result:
20.5

So it may be that rather than speed dropping off as CPU power exponentially
rises, the architecture of the chip changing is even more important than the raw
MHz speed.

I suspect that a pure MIPS ratio would give a much more steady number.  But I
have not tried that.



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.