home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: rec.games.chess
- Path: sparky!uunet!zaphod.mps.ohio-state.edu!magnus.acs.ohio-state.edu!usenet.ins.cwru.edu!gatech!willis1.cis.uab.edu!hyatt
- From: hyatt@cis.uab.edu (Robert Hyatt)
- Subject: Re: Wanted: details of (computer) evaluation functions
- Message-ID: <1992Dec30.034032.526@cis.uab.edu>
- Keywords: evaluation, position, positional, judgment
- Organization: University of Alabama at Birmingham
- References: <1992Dec29.035140.12403@cis.uab.edu> <1992Dec29.133014.13406@fccc.edu>
- Date: Wed, 30 Dec 1992 03:40:32 GMT
- Lines: 56
-
- In article <1992Dec29.133014.13406@fccc.edu> vanderli@orion.fccc.edu (Michel van der List) writes:
- >In article <1992Dec29.035140.12403@cis.uab.edu>,
- > hyatt@cis.uab.edu (Robert Hyatt) writes:
- >> I would recommend the following:
- >>
- >> 1. Chess Skills in Man and Machine, Springer-Verlag. A chapter on chess 4.0
- >> tells a lot.
- >>
- >> 2. Computers, Chess and Cognition, edited by Marsland and Schaeffer,
- >> Springer-Verlag. It has chapters on Cray Blitz, Hitech, Deep Thought, and
- >> others. The Cray Blitz chapter completely spells out it's evaluation
- >> function, search extensions, etc. I can give you an electronic summary
- >> of the chapter if you would like. You will need a unix-like troff
- >> formatter.
- >
- >First off let me say that I always enjoy Mr Hyatt's posts on computer
- >chess. His insights are always very interesting. Secondly I have two
- >questions:
- > - Could you post the above mentioned summary, I am sure there would be
- > quite a bit of interest in it.
- > - A few weeks ago you mentioned Sparc Blitz (I think thats what you
- > called it). Could you comment on it's strength?
- >
-
- First, and I hope either Jonathan or Tony reads this, is there any problem
- with nroff'ing part of the chapter I wrote and then posting that here? I
- know that it could possibly lose a few sales, but maybe it might also cause
- some readers to purchase the book to get the data on HiTech and Deep Thought
- after seeing what I wrote on Cray Blitz. Comments Tony or Jonathan?
-
- Second, "sparc blitz" is nothing more than the pure fortran version of Cray
- Blitz (it has about 15,000 lines of Cray assembly language that really speeds
- things up by a factor of five (5) over the pure fortran) running on a sparc.
- I have a sparc-2 on my desk and S-B searches about 1,000 nodes per second.
- Yeah, I know that's not very fast. Especially when we do 200,000 per sec on
- a YMP 8 and over 500,000 on a C90. The real problem with "sparc blitz" is
- that it't search extensions have been set up for the 9-10 plies we get on a
- YMP8. Running on a sparc, the current code simply extends too many lines
- too deeply and as a result, doesn't search as deep on average as I would like.
- Not having played any games, I don't know what to expect, but I suspect 6-7
- plies at 3 mins per move.
-
- It does work, although a few Cray-specific features (dealing with memory
- allocation, etc.) don't function. I occasionally use it to test problems
- I see in here, but don't expect to use it in a tournament. I have been
- working on converting to C for a year or so, very sporadically, and would
- expect that version to be somewhat better on a sparc after I optimize it
- away from the cray-specific things we do now. (some 64bit arithmetic
- dealing with hashing is a real pain and is slow as it is written in
- fortran.)
-
- Bob
-
- --
- !Robert Hyatt Computer and Information Sciences !
- !hyatt@cis.uab.edu University of Alabama at Birmingham !
-