Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Air-Cooled @ 10GHz, today

Author: Jonathan Parle

Date: 22:37:05 02/13/02

Go up one level in this thread


On February 13, 2002 at 13:17:44, ALI MIRAFZALI wrote:

>On February 13, 2002 at 08:51:31, Martin Andersen wrote:
>
>>On February 13, 2002 at 04:18:26, Jonathan Parle wrote:
>>
>>
>>>
>>>I'm not really sure doubling CPU speeds is worth 60 points. I know this was a
>>>demonstrable rule back in the late 80's, but I feel that the gains are now much
>>>smaller.
>>
>>According to the latest comment to the SSDF rating list, the
>>average rating increase for the programs, going from K6-2-450
>>to Athlon 1200 is 72 points.
>>1200/450 = 2.67. So calculating a 60 point increase when doubling CPU
>>speed does seem fair.
>>
>>
>>http://w1.859.telia.com/~u85924109/ssdf/list.htm
>>
>>
>>Martin.
>I do not think this is right .Or else the 1996 Deep Blue would have had a rating
>of well above 3000.

My view is that a significant component of the rating increase is attributable
to the greater memory available for hash table use and the more modern CPU
architecture rather than the pure speed of the processor. Last time I checked
the rating list, most of the K6-2 machines had 128Mb RAM whereas the athlons
often had 256Mb RAM. Given the differences in architecture between a K6 and
Athlon 1200, I don't think you can really say, all other things being equal,
that an Athlon 1200 is necessarily 2.67 times faster for chess than a K6-2
processor. I think the effective increase would be more than this. A bit like
the difference between the 6301Y and 6501 processors back in the 1980's. Back
then, you needed to run a 6301Y at around 8 Mhz to equal the performance of a
6501 at 2 Mhz. Furthermore, due to the exponential increase in the search tree
(even with selective pruning) as search depth is increased, I believe there will
be ever diminishing returns from going to ever faster CPUs. As an aside, I think
programmers should take advantage of faster CPUs by increasing chess knowledge
rather than trying to increase the number of positions examined.



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.