Parallel Chess Programs?
Patrick Surry
implemented on parallel machines and/or techniques for same? We are
interested in developing a very simple such program as a learning
experience.
Thanks for your help,
Patrick
---------------------------------------------------------------------------
Engineers think that theory approximates reality | Mathematicians never
Physicists think that reality approximates theory | make the connection
---------------------------------------------------------------------------
Patrick Surry, JCMB 2254, Edinburgh, Scotland e||) Edinburgh Parallel
EMail: p...@epcc.ed.ac.uk Tel: (031) 650-5960 c||c Computing Centre
---------------------------------------------------------------------------
Robert Hyatt
The ones I'm familiar with:
1. Cray Blitz. Runs on shared memory parallel machines like the ones from
Cray Research. Several publications describe it's parallel search.
2. Sun Phoenix. Runs on a distributed network of sun workstations and does
not require shared memory, but does need fast network interconnections like
ethernet or better. Again, several publications exist. Jonathan Schaeffer
is the contact here, although they have stopped work on the chess program
to pursue development on Chinook...
3. Deep Thought/Deep Blue. This is based on a symbiosis of a fast host
computer with multiple special-purpose chess processors to assist it in
carrying out the tree search. Might not be as applicable to your needs.
4. CalTech developed a chess program for the hypercube machine. Don't recall
the name, but it was designed for a specific architecture and might not be
applicable to your needs.
5. Bert Wendroff at Los Alamos worked on a parallel version of his LaChex
program, but later discontinued the parallel search due to the difficulties
of finding time on a parallel supercomputer to play chess...
6. Monty Newborn tried a couple of parallel search strategies, and again there
are publications to describe his work (ostrich).
As a final note, "beware" of trying to play around with parallel chess search
strategies. They are both addictive and difficult to debug.
--
!Robert Hyatt Computer and Information Sciences !
!hy...@cis.uab.edu University of Alabama at Birmingham !
Paul Rubin
Later it got flushed in favor of a single processor chess engine
that played stronger chess because of other reasons.
Ralf Stephan
> In article <C1Er...@dcs.ed.ac.uk> p...@epcc.ed.ac.uk (Patrick Surry) writes:
>
> >Can anyone give me any information on chess programs that have been
> >implemented on parallel machines and/or techniques for same? We are
> >interested in developing a very simple such program as a learning
> >experience.
>
And
7. "Zugzwang" from the University Paderborn (Germany) which runs on 1024
T-825 transputers and scored 4/5 in the last computer WC.
They plan to upgrade to 65535 processors and want to use the T-9000
whenever it mayeth be available... (Team: R.Feldmann, P.Mysliwietz)
ralf
--
************************************
One program's worth a thousand words.
Peter Rainer
>hy...@cis.uab.edu (Robert Hyatt) writes:
>> In article <C1Er...@dcs.ed.ac.uk> p...@epcc.ed.ac.uk (Patrick Surry) writes:
>>
>> >Can anyone give me any information on chess programs that have been
>> >implemented on parallel machines and/or techniques for same? We are
>> >interested in developing a very simple such program as a learning
>> >experience.
>>
>> The ones I'm familiar with:
>> [...]
>And
>7. "Zugzwang" from the University Paderborn (Germany) which runs on 1024
> T-825 transputers and scored 4/5 in the last computer WC.
> They plan to upgrade to 65535 processors and want to use the T-9000
> whenever it mayeth be available... (Team: R.Feldmann, P.Mysliwietz)
>ralf
Although we would love to see Zugzwang running on 64K prozessors,
we do not plan to upgrade. If you consider that one transputer
costs about 3000 $ with mem and all the stuff you need, you now
the reason.
But we do indeed wait for the T9000.
Peter Mysliwietz