Author: Richard A. Fowell (fowell@netcom.com)
Date: 21:22:06 05/11/98
Go up one level in this thread
On May 11, 1998 at 14:59:20, Bruce Moreland wrote:
>And if they never play again, DB is even less interesting.
>
<massive snip>
>>To kick an open door, suppose they are stupid, but tactical strong
>>then why if they solve for example all Nolot test set positions, then
>>why didn't they publish it?
>
>They haven't published anything significant yet.
>
>bruce
The closest I've seen is the article in IEEE's magazine
Computer (October 1997) called "Deep Blue's Hardware-Software Synergy"
( This article wasn't written by Hsu - it was based on interviews with
Hsu and Campbell, and Campbell commented on early drafts).
Much of the article speaks to features of Deep Thought (as opposed to
Deep Blue), and is occasionally hard to tell which one they are talking
about.
However, it seems that Deep Blue used:
- 32 Gbyte disk for opening and endgame databases (with 32Gbyte backup
disk)
- 32 P2SC processors assigned as:
one 135 Mhz node as system master and backup I/O processor
one 135 Mhz node as primary I/O processor and slave
30 120 Mhz computation nodes, each with 16 chess ASICs (total 512
ASICs)
ASIC is 1.7 million transistors, 0.6 micron technology, 2-4 million
moves/second
ASIC functions
- generates moves
- evaluates position (2/3 of chip area)
- simple raw search
- 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"
* Only about 1/3 of the available parameters were used in the first
Kasparov match, because the team didn't have time to tune the system.
* Deep Blue relies largly on manual tuning, based on Joel Benjamin's
assessment that Deep Blue misevaluated a position. Special [software?]
tools were then used to decompose the evaluation and see the effect
of modifying parameters.
"Initially, the team perceived a gap between Deep Thought's evaluation
function and that of top commercial chess programs. However, over the
past two years, Deep Blue's evaluation function has become superior to
other top programs. It has even defeated several programs while in
debugging mode, which uses only a single processor and ASIC."
- Computer, page 32, October 1997
-RAF
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.