Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: IBM would risk 37 billion dollar

Author: Robert Hyatt

Date: 12:31:40 03/30/01

Go up one level in this thread


On March 30, 2001 at 13:42:36, Eugene Nalimov wrote:

>There *are* 32-CPU Xeon computers. Look at
>http://www.unisys.com/hw/servers/enterprise/7000/default.asp
>
>Eugene


OK... I wasn't thinking of _that_ kind of architecture.  IE it isn't the
same as my 4-way xeons.  There has always been a batch of "tightly-coupled
cluster" machines lying around.



>
>On March 30, 2001 at 11:40:50, Robert Hyatt wrote:
>
>>On March 30, 2001 at 09:51:47, Vincent Diepeveen wrote:
>>
>>>On March 30, 2001 at 09:17:59, Robert Hyatt wrote:
>>>
>>>>On March 30, 2001 at 08:44:47, Vincent Diepeveen wrote:
>>>>
>>>>>On March 28, 2001 at 23:14:30, Robert Hyatt wrote:
>>>>>
>>>>>>On March 28, 2001 at 16:38:31, Rajen Gupta wrote:
>>>>>>
>>>>>>>chinook ckeckers, (the checkers equivalent of deep blue)is ready to face all
>>>>>>>comers on its web site(admittedly a single processor version, but the same
>>>>>>>software)why cant deep blue or even deep blue junior do the same? is it because
>>>>>>>it is scared of being exposed for what it really is?
>>>>>>>
>>>>>>>rajen
>>>>>>
>>>>>>
>>>>>>Why would they be afraid to expose it as an incredible chess-playing
>>>>>>machine?  Can't we figure that out from the Kasparov match?
>>>>>
>>>>>As you know stocks and shares have dropped past months/weeks quite
>>>>>a bit. Still IBM is worth 167.3%
>>>>>
>>>>>After match IBM stock went up most likely because of Deep Blue 22% in 1997
>>>>>right after winning match.
>>>>>
>>>>>So letting deep blue lose on the web now from all commercial progs
>>>>>and a bunch of chessplayers who will figure out its weaknesses
>>>>>would be risking 22% x 167.3 = 36.806 billion dollar
>>>>
>>>>
>>>>
>>>>First, Deep Blue would not lose on the web to all commercial programs.  Some
>>>>might tend to report things optimistically.  IE "My program beat Deep Blue"
>>>>when it should be "My program beat deep blue one game out of 23."
>>>
>>>Ok first of all from chesstechnical viewpoint i would be amazed if
>>>deep blue at slow level would ever get further as a draw against me.
>>
>>Didn't you say something like that for the Dutch tournament?  I would
>>_never_ say that a program could at best draw me.  I know that even very
>>old programs can bite me from time to time.
>>
>>
>>
>>>
>>>Note i assume it is possible to play slow level against it as i
>>>see it on average wasted like 30 seconds a move, which probably
>>>means it either had a hell slow operator or it needs like zugzwang
>>>quite some time to communicate to the processors to start the search.
>>
>>There is _no_ communication delays to mention.  At the beginning of a search,
>>they do sometimes have to download evaluation tables to the chess engines, but
>>this is only done when the values need to be changed.  Hsu didn't like to play
>>5 second per move games because a couple of seconds were lost doing this.  But
>>only a couple of seconds.  Any other time lost is simply operator time.  The
>>machine was strong enough, and the match was so visible to everyone, I suspect
>>they chose to be _very_ careful in move input/output rather than trying to save
>>every last second and produce too many "take-back" situations that would
>>subject them to criticism.
>>
>>
>>
>>
>>>
>>>So i definitely doubt it.
>>>
>>>But let's approach it from how IBM will approach it. It will be
>>>a business decision, and those decisions are usually taken by managers
>>>who probably never heart of the 'en passant' rule (i'm not implying
>>>that most readers of CCC therefore know the rule).
>>>
>>>Managers think in terms of numbers. The number is quite convincing
>>>to NEVER ever again let Deep Blue live when it is interesting to let it
>>>play. It's obvious that in 2001 programs are hell better now as in 1997.
>>
>>So?  IBM's SP is also a lot faster now.  And Hsu re-designed the chip yet
>>again although he didn't build one.  So DB _would_ be a lot better today as
>>well.
>>
>>
>>
>>>
>>>Like any commercial program of today will beat the hell out of this deep
>>>blue thing. If it's not book, then it's because of pawn structure, if it's
>>>not because of that, then it'll be exchanging some crucial pieces in the
>>>middlegame etcetera.
>>
>>I don't know of any commercial program today that can "beat the hell" out of
>>my program on good hardware.  And I wouldn't begin to claim that I would have
>>any real chance vs DB at all.  I might draw a game here and there, and I might
>>win a rare game if I am lucky.  But Crafty at 20M nodes per second is not going
>>to be a serious threat to them.  Although it would be an overwhelming threat to
>>anything else.  For a comparison, bring your program on a single cpu machine and
>>try crafty on my quad to see how you do.  Then figure that Crafty could run 20x
>>faster on a big Compaq machine, and think about how you would do against that.
>>And then DB is somewhere way beyond that rainbow...
>>
>>
>>
>>
>>
>>>
>>>However this is all of no concern of business men. They see a 37 billion
>>>dollar risk.
>>>
>>>Even if that risk would be 0.05 as you say (though i think it's more
>>>like a 100% sure risk) then count the win versus lose problem.
>>>
>>>I'm bad in math, so are the business men, so they will do next
>>>math:
>>>
>>>  overall risk: 22% of stocks/shares = 37 billion
>>>
>>>  working risk: 0.05 x 37 billion    = 1.85 billion risk
>>>  working win : ??   x ??            = 0
>>>
>>>What can they WIN with it? What is their working win?
>>>Their working risk is arguably 0.05.
>>>
>>>Most likely it's more like 60%.
>>>Things go bad in this world economically now because of big crises in Japan
>>>(or whatever).
>>>
>>>So SUPPOSE intel goes play them. With DIEP at a 32 processor Xeon.
>>>Very cheap for them to do. They have plenty of 32 processor Xeons idling
>>>day and night and by dealing a blow to IBM they might want to
>>>risk giving me a bit of system time, which i of course will blindfolded
>>>accept.
>>
>>There aren't any 32 processor xeon machines.  There are 8-way xeon
>>machines however.  But even using 4 of those would leave you 100x slower
>>than DB.
>>
>>
>>
>>>
>>>If i start winning bigtime suddenly they will write the sponsor name
>>>with big capitals suddenly and ask a few TV teams to get some free
>>>drinks.
>>>
>>>They by accident film how IBM technology is getting kicked ass by Intel.
>>>
>>>No here the more likely scenario to happen one day.
>>>
>>>A manager proposes this idea to play again with deep blue for
>>>'commercial purposes'. A manager who thinks even simpler as i do
>>>(as i would play just for the sake of chess) is doing the above math,
>>>and he fires the person(s) who proposed the idea!
>>>
>>>>Second, I doubt anyone would argue that with the current positive impression
>>>>the general public has about the Deep Blue project, IBM would be foolish to do
>>>>_anything_ to tarnish that image.  When the "shine" wears off, DB might re-
>>>>surface, I don't know.  But at present, when you are on top, there is no need to
>>>>return to the ring too quickly.
>>>
>>>My point is that they will never get back. If they do expect a big bunch
>>>of managers to get fired.
>>
>>
>>They will get back _if_ it seems to be financially rewarding.  When the name
>>"deep blue" has faded from memory, reviving it would be worthwhile, most likely.
>>
>>
>>
>>>
>>>>>
>>>>>Last Trade
>>>>>Mar 29 ยท 95.04 Change
>>>>>0.00 (0.00%) Prev Cls
>>>>>95.04 Volume
>>>>>0 Div Date
>>>>>Mar 10
>>>>>Day's Range
>>>>>0.00 - 0.00 Bid
>>>>>N/A Ask
>>>>>N/A Open
>>>>>0.00 Avg Vol
>>>>>9,203,545 Ex-Div
>>>>>Feb 7
>>>>>52-week Range
>>>>>80.0625 - 134.9375 Earn/Shr
>>>>>4.44 P/E
>>>>>21.41 Mkt Cap
>>>>>167.3B Div/Shr
>>>>>0.52 Yield
>>>>>0.55



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.