Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Question for Robert Hyatt about Deep Blue moves

Author: Slater Wold

Date: 10:34:23 10/11/02

Go up one level in this thread


Bob, you have a lot more patience than I.

I was going to respond with almost the same exact post, but knew it would do no
good, so I refrained.

It's obvious.  Some people read, listen, think.  Some people just read and talk.



On October 11, 2002 at 12:36:52, Robert Hyatt wrote:

>On October 11, 2002 at 08:02:47, Vincent Diepeveen wrote:
>
>>On October 11, 2002 at 00:59:36, Slater Wold wrote:
>>
>>Slate, everyone recognizes that deep thought was an
>>absolute beginner. Do you agree?
>
>
>Yes.  But a beginner so strong that it produced a 2650+ performance rating
>over 25 consecutive games against GM-only opposition, at 40 moves in 2
>hours.
>
>One _hell_ of a "beginner" I would say...
>
>Has your program even beaten a GM at 40/2?  Crafty hasn't played one at that
>time control, but Cray Blitz did and won.  As did HiTech.  And Deep Thought.
>
>
>>
>>But please remind that deepthoughtII was getting 10
>>million nodes a second.
>
>
>Wrong number.  Ask Hsu.  at one point they had 14 processors and got close to
>that, but not for more than a couple of games...  Hsu generally claims DT was
>2-3M nodes per second...
>
>
>
>
>>
>>Now it only gets 20 times faster and from absolute
>>piece square tables it has a kind of gnuchess
>>evaluation, also with precalculated values for the
>>parameters, instead of an independant leaf evaluation.
>>
>>In short it's not so impressive. If they play
>>beginners level with 10 MLN a second, and they
>>do not use nullmove, then what level do they play
>>with 126MLN a second, knowing that they had severe
>>parallel losses. Practical speed was about 5% of the
>>total speed.
>>
>>Of course same is valid for DTII.
>
>
>Again, ignorance is bliss.  Deep Thought II had _nothing_ to do with deep blue,
>in terms of parallel search.  Don't you _ever_ read anything?  Deep Thought II
>used a Sun workstation with multiple chess chips in it.  Deep Blue had a two-
>level parallel search using 30 IBM SP processors, each with 16 chess processors.
>
>No relationship whatsoever, with respect to parallel search.
>
>
>
>>
>>I need to remind you too that if i make a program that's
>>not evaluating much. Say only 40 parameters, which is
>>about DBII's eval (as published in artificial intelligence)
>>that it is not so hard to get 2 million nodes a second
>>on a single K7.
>>
>>In fact older fritz versions which are in the same league
>>like that 40 parameters, if they would get optimized to
>>K7 would get a hell of a lot more nodes a second.
>>
>>At a dual K7 it clocked against me at 2.2MLN nodes a second.
>>
>>And we all know how bigtime it was slowed down the past few years
>>by adding knowledge and doing more sophisticated search.
>>
>>DB never did a sophisticated search, so getting more nodes a second
>>is a hell of a lot easier then too!
>>
>>>On October 11, 2002 at 00:25:27, K. Burcham wrote:
>>>
>>>>
>>>>
>>>>Robert,  I know you are more qualified than most to have an opinion about this
>>>>comparison of Deep Blue to Deep Fritz. I know you have many reasons to have
>>>>formed this opinion. As you know I find the electrical/mechanical machine we
>>>>call Deep Blue very fascinating for its time.
>>>>
>>>>I have heard your comparisons about hardware, software, search depth, memory,
>>>>search methods, etc. that explains some of your reasoning about comparing Deep
>>>>Blue to todays programs.
>>>>
>>>>My question is, separate from all of this technical discussion, do you have
>>>>several moves that you have studied with todays programs that you know these
>>>>programs cannot find?
>>>>What have you found to be the most difficult of Deep Blues moves?
>>>>Would you please post your findings here for others to study?
>>>>
>>>>thanks
>>>>kburcham
>>>
>>>I have read just about everything there is to read about DB.  Lectures, thesis,
>>>ICCA journals, everything.  While I'm not Bob, I will take a stab at answering
>>>your questions.
>>>
>>>
>>>DB and later DBII were massive machines.  DBII had 400+ CPUs designed for the
>>>sole purpose of examining a chess board.  While I believe that the software was
>>>not as advanced as todays, I do have to remind you they did some pretty fancy
>>>extensions (among other things) that cannot be afforded on a PC at this time.
>>>
>>>Did DB or DBII ever make a move that cannot be reproduced by a computer now?
>>>Not that I have found.  If you give a certain engine a certain amount of time,
>>>most will always find DB moves.  Some faster, and some slower.  Some a lot
>>>faster, and some a lot slower.
>>>
>>>Why?
>>>
>>>Anyone can look at the games played against DBII and tell Kasparov was *not* on
>>>top of his game.  DBII was never given the oppurtunity to shine, thus, it never
>>>did.  There are I believe 2-3 positions in DB/DBII's history that still give
>>>some programs a hard time, but nothing that switching PCs or programs won't
>>>solve.
>>>
>>>However.
>>>
>>>You can look back at Deep Thought and Deep Thought II and pick some of it's
>>>games to compare.  Granted, you will not have a one on one comparison, but you
>>>will have a close one.  DBII was fairly stronger than DTII, however, they are in
>>>somewhat the same realm.
>>>
>>>Nolot.
>>>
>>>Back in the 90's when Nolot released his position suite, he said that it would
>>>be 10+ years before computers where getting these moves (and lines) right.  DTII
>>>completly squashed Nolot.  At that time, no program or PC could even *dream* of
>>>getting those positions solved.  At present day, *some* programs with good HW
>>>can get *some* of the Nolot positions correct.  AFAIK none do as well overall as
>>>DTII.  And please, try to keep in mind DTII was 2 generations before DBII.  2M
>>>nps vs 200M nps.  (Ok, 150M nps.  Whatever number you want to pick.)
>>>
>>>TPR.
>>>
>>>DTII was crushing GMs at a time where if you had brought a PC to play a GM,
>>>people would laugh at you.  After the Kramnik-DF7 match, it's looking like
>>>you'll get this same attitude nowadays.
>>>
>>>
>>>I will leave you with this; why if we have advanced SO much in computer chess,
>>>is the "best" computer chess software on the best computer hardware getting torn
>>>apart by a GM?  Because he got the software in advanced?  Give me a break.
>>>That's part of normal chess.  You study your opponenets before you play them.
>>>You don't think Frans and his crew were going over ever Kramnik game played?  I
>>>am sure it helped Kramnik, but nothing like most people think.
>>>
>>>
>>>Look at the DT games.  Look at the DF games.  It is clear to me that there is a
>>>pretty huge difference in quality of play.
>>>
>>>
>>>
>>>
>>>
>>>** All of you anti-DB guys can go ahead and reply all you want.  I've heard your
>>>side 1,000 times.  I don't agree.  So go ahead, waste the keystrokes.  It's your
>>>time, not mine. **



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.