Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Deep Junior has proved that Hardware is way Over-rated

Author: Dann Corbit

Date: 11:35:03 07/19/00

Go up one level in this thread


On July 19, 2000 at 13:08:06, pete wrote:

>>You're probably correct there, but it's an _educated_ guess of 50-70 ELO per
>>doubling of CPU.  That means an ELO difference of (estimate 60)*8 = 540 *
>>.75(smp loss) = 405 ELO.  That's probably too high, but it's probably not off by
>>more than a factor of 2 in either direction (WAG).
>>
>
>I don't think this is a reasonable guess at all .
>
>It probably _is_ an educated guess for comp-comp .
>
>From all I have seen reported by others or with my own eyes in comp-human
>hardware is of much lesser value ( to what extent yet unmeasured ) .

Have you actually done some math, or is it just by impressions?  Often,
impressions can be wrong, but math rarely tells a lie.

>I am not sure that say a Junior on a PIII500 will outperform a Junior on a
>P233MMX at all by a significant margin in games against humans .

I am.  On the same time control, you are essentially given double the time
because of being almost exactly twice as fast.  That will give nearly one more
ply (let's say 2/3 of a ply on average).  Given enough moves, this will produce
an enormous benefit.  Let's say that 15% of finding a better move is right, or
maybe only 10% (the goes deep studies).  We only get 2/3 of this so we have
about .067 chance every move of improvement.  So what is the chance that we
don't make a better decision after n moves?
(1 - .067)^n
After only ten moves, the odds are 50% that we will have improved our position.
After twenty moves the odds are 75%.
After 30 moves, the odds are only 13% that we won't have made an improvement.

>This is a non-educated guess too for sure but goes together well with what I
>have seen so far .

There may be a logical basis to your idea, if people know how to play
anti-computer chess against it.  That is because it will still lose no matter
what.  But even against anti-computer players, I suspect that seeing deep
tactical shots has a better chance to wreck the anti-computer strategy.

It would make an interesting study anyway.



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.