Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Has anyone kept a record of reported "Techniques" used in DeepBlue?

Author: Robert Hyatt

Date: 05:58:15 11/14/01

Go up one level in this thread


On November 13, 2001 at 19:23:00, Dann Corbit wrote:

>On November 13, 2001 at 18:03:18, Robert Hyatt wrote:
>
>>On November 13, 2001 at 15:37:52, Slater Wold wrote:
>>
>>>On November 13, 2001 at 13:02:34, Joshua Lee wrote:
>>>
>>>>I have seen many argue this point weather or not DB had much in the way of
>>>>intellegence which makes no sense do you really think that IBM would risk just
>>>>having a node cruncher?  Don't you think that Someone on the DB team would
>>>>explain how there are many positions where brute force wouldn't work even if
>>>>every cpu in the world were connected together?
>>>>
>>>>What technigues Can be used together and on micros? Has anyone tried to make
>>>>their program more like this enigma?
>>>>
>>>>Here's what i found:
>>>>
>>>>sophisticated quiescence search
>>>>  - endgame heuristics
>>>>  - a few small endgame databases
>>>>  - position repetition detection
>>>>  - calculates mobility
>>>>  - evaluates space
>>>>  - close to 50 tables to evaluate a chess move
>>>>      (implied that this includes:
>>>>          piece square tables
>>>>          pawn bitmaps
>>>>          open file
>>>>          coefficient updates after each move (incremental evaluation)
>>>>
>>>>Deep Blue can recognize (in hardware) approximately 6,000
>>>>chess-specific features
>>>
>>>DB's features were 75% hardware and 25% software.
>>>
>>>With a micro, they have to be 100% software.
>>>
>>>Hence the catch on trying to make something "like" it.
>>>
>>>DB was the last chess "super computer".  Chess software is made to make money,
>>>or as a hobby.  Chess hardware is made to spend money, and lots of it.
>>>
>>>Although DT I only cost about $5,000.  Less than *most* servers now.  A LOT less
>>>than Hyatt's quad 700.  (A 700mhz Xeon CPU is about $1,300 right now.  That's
>>>$5,200 just in CPU's.)
>>
>>
>>My total machine is worth about $10,000 now.  We just bought another one
>>identical to my machine to use for our departmental file server.  Only
>>difference it that the new one has 6 36 gig 10K scsi drives and a raid-5
>>(hardware) controller.  The chassis is now under $2K.  $5K for the processors,
>>a couple of hundred bucks for 512mb of RAM, and then the disks of your choice
>>will give you a quad for under 10K.  I don't know if Intel has released any
>>quad 900 or quad 1000 certified processors.  They will work in this chassis
>>when/if they do.
>
>You can easily price out an 8 CPU Xeon box at $100K.  At some point, I suspect a
>Cray is cheaper!  When there are 5 digits in the price {not counting the
>decimals} it's not a PC anymore.  Beyond that, we're talking mainframe.
>;-)
>
>Here is one that starts at 25K {FOR THE BARE BONES MODEL!!}
>http://rcommerce.us.dell.com/rcomm/config.asp?order_code=PE8450
>I worked up a price of $102,443.00 by fiddling around with various options.
>With Linux, it drops clear down to: $98,608.00
>;-)


With a cray, you will always see seven zeros in the price tag.  IE the T932
I ran on a few years ago sold for $60,000,000. roughly.  :)

The 8-way machines are a kludge for a couple of reasons and I don't particularly
like them.

1.  They are really supported by the chipset as two clusters of 4 processors.

2.  Memory still stays at 4-way interleaving, which means that memory contention
is much worse on those machines.  This is probably why they were originally sold
with only 2M L2 caches.  I don't know if that is still a limit or not, but I
have run on one and it definitely has a noticable memory bottleneck.



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.