Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: An Experiment that disproves Hyatt's 1000X NPS Theory

Author: Uri Blass

Date: 05:31:04 09/18/05

Go up one level in this thread


On September 18, 2005 at 07:49:30, ALI MIRAFZALI wrote:

>On September 18, 2005 at 07:39:50, Uri Blass wrote:
>
>>On September 18, 2005 at 07:24:43, ALI MIRAFZALI wrote:
>>
>>>On September 17, 2005 at 21:51:04, Robert Hyatt wrote:
>>>
>>>>On September 17, 2005 at 17:10:00, ALI MIRAFZALI wrote:
>>>>
>>>>>On September 17, 2005 at 13:42:07, Albert Silver wrote:
>>>>>
>>>>>>On September 17, 2005 at 10:04:32, ALI MIRAFZALI wrote:
>>>>>>
>>>>>>>Hyatt has claimed many times that a Nodes Per Second Factor of one thousand
>>>>>>>times would not be overcome by the program with the less Nodes per second.In
>>>>>>>this Experiment it was shown conclusively that this is false .Although I played
>>>>>>>4 games ,I do not think the result would have been different if I had played a
>>>>>>>hundred more.Time Control 40 MOVES IN 2 HOURS followed by sudden death in 1
>>>>>>>hour.Hardware: GNU CHESS 4.11 a program from 1996 ran a celeron 1.8 Gig machine
>>>>>>>;Chess Tiger on Palm ran on the Palm Tungten E.NODES PER SECOND:ON THE
>>>>>>>AVERAGE:CHESSTIGER ON PALM 500 per second ,GNU CHESS 4.11 500000 per second on
>>>>>>>the celeron 1.8 Gig.1000X DIFFERENCE.Hyatt and some other people have always
>>>>>>>argued about the supremecy of DeepBlue based on its speed.I think these days
>>>>>>>these arguments are false;and Speed does not mean as much as it used to.Deep
>>>>>>>blue would be crushed by todays program's.A lot of STRENGTH is EVALUATION
>>>>>>>FUNCTION.Take a look at these games:
>>>>>>>Match ended in 2-2 draw.
>>>>>>
>>>>>>The idea of testing this is certainly interesting but the conditions seem rather
>>>>>>dubious IMHO. For one thing, 4 games really is COMPLETELY meaningless, andwith
>>>>>>all due respect to claim you don't think the result could have been different
>>>>>>shows how much you don't understand this.
>>>>>>
>>>>>>BTW, does Tiger really only get 500 nodes per second on your Palm? That seems
>>>>>>ridiculously low. I don't have Tiger, nor a Palm for that matter, but on my Dell
>>>>>>Pocket PC at 624MHz, I get about 50,000 nps on average for Fruit 2.1.
>>>>>>
>>>>>>Note that if one is to believe the results of Hiarcs site
>>>>>>(http://www.hiarcs.com/phresults.htm), Tiger on the Palm has inordinately bad
>>>>>>results (they claim it plays over 400 points worse than Hiarcs on identical
>>>>>>hardware, which is HUGE), so perhaps it isn't the ideal choice.
>>>>>>
>>>>>>                                        Albert
>>>>>Yes indeed Tiger does get 500 Nps on the palm Tungsten E.Note that it is Hyatt's
>>>>>claim that I am disproving.According to him a NPS of 1000X factor would be
>>>>>impossible to overcome even in 4 games
>>>>
>>>>
>>>>No, what you are disproving is a false statement you are making.  Since I never
>>>>said what you claim, it would be just a bit difficult to disprove it, since it
>>>>was never said.
>>>>
>>>>Grow up or try again...
>>>Well ,this is what you have said in the past:Programs of similar search
>>>techniques or something to this effect.I do remember you and me having this
>>>conversation in another thread .At that time I was going to try to find a 1000X
>>>NPS difference Experiment.And why do you call GNU chess4 a lousy program?For
>>>1996 standards it was good.But this is the whole point:Deep Blue was also 1997
>>>standards.
>>
>>2 points:
>>
>>1)Gnu was not good for 1996 standards.
>>Genius3(a program of 1994) win it convincingly and programs that are not close
>>to Genius3's level are not good for 1996 standards.
>>
>>2)Programs that were good for 1996 standards may be lousy programs later.
>>
>>Uri
>You may be correct Uri :but the point is:Was Deepblue good for 1997 standards?
>your point no 2 proves what I am saying Deepblue would get crushed by todays
>programs

I have the same opinion that deep blue whould get crushed by today programs but
it cannot be proved by np experiments.

I agree that if we wait enough time the improvement in software may be
equivalent to 1000:1 hardware advantage.

I do not think that the improvement from 1997 to 2005 is enough for it but I
also do not think that it is easy to compare speed of Deep Blue and speed of
other programs.

Note that deep blue had bugs (I remember that it did a stupid blunder because of
bug in the first game)

Parallel search is not a simple task and we do not know if there were other bugs
and what was the influence of them so I think that it is better not to discuss
about things that we cannot check.

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.