Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: HW based Crafty

Author: Keith Evans

Date: 15:45:40 03/30/02

Go up one level in this thread


On March 30, 2002 at 15:08:57, Slater Wold wrote:

>On March 30, 2002 at 14:37:35, Robert Hyatt wrote:
>
>
>I profiled TSCP and board.o was the most expensive to run.  This really is no
>surprise because TSCP has a *very* basic eval & search.
>
>On Crafty, either Search or Eval are going to be the most expensive, HOWEVER,
>would be a real pain to implement into hardware.
>
>I have not profiled Crafty yet, only because my computer is being upgraded.  I
>will profile it as *soon* as I can.

If you don't already have a copy of Hsu's thesis and IEEE micro articles, then
you should get copies of those. You won't find the Belle articles very
enlightening as they don't get into enough detail - just get copies of Hsu's
papers.

Then when you're designing your move generator think about how an on-chip
microsequencer might want to interface to the move generator. And think about
how both would interface with some on-board evaluation hardware. Your PCI
interface could be a target only interface that's used to communicate to the
microsequencer.

You'll probably find that using a ChipTest (move gen only) type approach just
slows Crafty down, but don't let that discourage you because you'll need a move
generator if you decide to make a more complete chess coprocessor later. IMHO
the next step for you would be to put a Deep Thought style design onto an FPGA
and maybe add transposition tables to that if your development board supports
external memory. Hsu never did glue on any transposition tables. Does anybody
know why not? Is there more to it than simply running out of time?

I think that if you start reading about all of the the features in final Deep
Blue chip then you might get a little overwhelmed. And it's a little mystifying
because some of the features are not well documented. I don't think that it
would be impossible to implement all of the features into one of the larger
Virtex parts, just keep in mind that Hsu spent a decade thinking about this
stuff so it will take a while. (You didn't propose doing this, but you might
eventually start thinking about it.)

Regards,
Keith



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.