Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Tiger against Deep Blue Junior: what really happened.

Author: Chris Carson

Date: 17:05:11 07/27/00

Go up one level in this thread


On July 27, 2000 at 19:14:32, Jeremiah Penery wrote:

>On July 27, 2000 at 14:07:00, Chris Carson wrote:
>
>>4.  So 97 DB has 480 chips and 200M NPS.  So what.  It needed it, todays
>>    programs get almost the same results with 2.5M NPS due to better
>>    evals and search techniques (DJ on 8x-700 for example).
>
>I don't think there is any data showing that, against humans, NPS is all _that_
>important.  Once you get to some critical level that facilitates a minimum
>acceptable search depth, anything above that may not help so much.  I think
>unless you do something quite different in the search logic (as apparently DB
>did), those extra nodes aren't doing much for you (relatively).
>

I agree with you.

>>5.  Todays programs are still just 1 or two clock doublings from
>>    beating the 97 DB performance (TPR).  It will also require an
>>    equal amount of sw improvements, but I have faith in our programmers.
>>    FIDE may save 97 DB here if rated players can not play against comp.
>>    next year (at least the games may not be rated by FIDE).  It would
>>    be funny if FIDE saved DB.  :)
>
>Again TPR != STRENGTH!  How many times must I say it?  Also, comparing the DB
>TPR to any other TPR is meaningless for several reasons which I won't reiterate.
>

Well, I can accept that you have a different opinion.

>>6.  DT beat old programs on 6502/386/486 and won the 1989 WCCC.  Good
>>    for it.  Has no relevance since programs today have better results
>>    against the same programs (see SSDF list).  Lost the 1995 WCCC
>>    to Fritz3 running on a P-90.  I was not impressed.
>
>Comparing DT '95 or before to DB2 '97 is pretty pointless - it's approximately
>the same as trying to compare Rebel 3 on an Apple IIe with one of the plug-in
>cards to Rebel Century on a PIII-GHz machine.  IIRC, the DB team admitted that
>even DB '96 had a somewhat weaker evaluation than the top Micros of the time,
>but it was probably still better than the Micros even of today because of its
>search - Have you seen DTs results on the Nolot test?  I'd be willing to bet
>that any micro of today can't come even close to matching them even when
>searching twice as many nodes.
>

This is the core of the debate.  We agree about the eval and I have no
way to prove superior search for either (hey I like DB, I just like the
micro's also).  We have different opinions and I can accept that.

>>7.  You have not proven any 90% superiority over 8x-700 by 97 DB.  There
>>    is a statistical basis for a 25% 97 DB superiority.  That will diminish
>>    quickly over the next 18 months.  Ed was right with 25% in my opinion.
>
>This is all pointless speculation.  Until we see some games, nobody is ever
>going to agree.
>

I can agree to disagree.  I can respect an oposing opinion.

>>8.  97 DB made tactical errors, no shame here, just facts.  Todays micros
>>    make errors, but they are still evolving and in 18 months that number
>>    will be fewer (still there, just harder to find).  Amir has the
>>    positions if you want them.
>
>I would definitely like to see these "tactical errors".  And Rd1 doesn't count
>because it was caused by a bug.
>

Well, a bug counts for me, but I can respect it if you do not count it.

>>9.  Todays micros have played hundreds of games against FIDE rated players
>>    and the results are published for everyone to study.
>
>Hundreds of games at tournament time control?  Can you point me towards these
>games?
>

See: Tony's page: http://home.interact.se/~w100107/welcome.htm
SSDF calibration games and over 200 associated with my list since
1997.  These meet FIDE criteria see www.fide.com.

>>10. Using games since 97 at 40/2 the programs of today have a TPR of
>>    2544 (as I predicted in Feb 200 when I started this and you had a
>>    fit and said no way).  2 programs recently scored above 2600 and
>>    one above 2700.  The programs of today are GM level on P-200s
>>    and above (FIDE says 2500 is GM level).  Guess you lost that one.
>
>I would really like to see a program on a P-200 beat a GM at 40/2 in a match.
>And again TPR != STRENGTH!

See Walters post, he did an independant stat analysis on the data
I collected, it included P-200 HW and the mid point was 2544.  I do think
a P-200 is less likley than faster HW, but there have been some
impressive results and it would not be impossible from a stat point
of view.

I can respect that you may not accept this, I can agree to disagree.

>
>>    I for one
>>    an bored with your same old 480 chips and 1B NPS top proves
>>    something.  Not to me, but I am flattered that you will die tring
>>    to change my mind or mock me in your posts.  I guess you consider
>>    me your academic nemisis.  You certainly follow me around a lot
>>    and disagree every chance you get.
>
>And you haven't been doing the same? :)

I am not inocent, I make mistakes, I admit them, I try to work
it out with the person and can agree to dis-agree.

This debate has lasted a long time.  I hope I have not offended you
and that you can respect my side of the debate even if you do not
accept it.  Hey I feel just as strongly.

Best Regards,
Chris Carson



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.