Author: Robert Hyatt
Date: 14:52:43 04/03/02
Go up one level in this thread
On April 03, 2002 at 16:37:25, Keith Evans wrote: >On April 03, 2002 at 15:52:43, Robert Hyatt wrote: > >>On April 03, 2002 at 13:23:57, Keith Evans wrote: >> >>I hope I didn't say much about chiptest, because it didn't last very long. >>But deep thought was definitely "belle on a chip" according to the write-up >>by Hsu in the book "Computers, chess and cognition". Farther into the article, >>Hsu says "the main difference between the processors is the enhanced hardware >>evaluation function." (where 'the processors' is referring to the chiptest >>machine vs the deep thought chip.) >> > >I don't believe that I have a copy of "Computers, chess and cognition" but I do >have a copy of Hsu's PhD thesis (F.-h. Hsu, Large-Scale Parallelization of >Alpha-Beta Search: An Algorithmic and Architectural Study, PhD thesis, Carnegie >Mellon Univ., Computer Science Dept., Feb. 1990) which contradicts this. I don't >have it in front of me right now, but it is quite clear that the only custom >silicon in Deep Thought is the single chip move generator. The only other >arguably custom chips are Xilinx LCA's which are used to implement the >evaluation logic. > >The slow evaluation really just looks at pawn structure and how rooks control >files. Based on what I read about Belle it was more extensive than this, but >perhaps more limited in it's searching algorithms. Hsu expresses some amazement >at how well his relatively simple evaluation performed in competition. > >>No idea about the above usage of FPGA. Ken used that acronym in a paper he >>wrote prior to 1984... > >My guess would be that he referred to FPLAs, but it doesn't really matter. >Whatever they used was quite simple compared to what is available today, and >they had to do a lot of board level and system level design as a result. Do you >know how much time they spent designing, building, and debugging Belle? Seems >like at least a year long full-time project for the average engineer - >especially if you include all of the different versions. > >Regards, >Keith Roughly 6 months from "the idea" to "the box". At the 1979 ACM event he was using his old hardware move generator and hardware evaluation and hardware make/unmake move stuff, and nothing more. That was in October/November. By the next summer and the WCCC event, he had the new Belle up and going and he won that event... There were exactly two of them doing the work... Ken and Joe Condon... Both were hardware types although Ken was also a software person (Joe was not).
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.