Author: Christophe Theron
Date: 08:59:46 11/23/99
Go up one level in this thread
On November 23, 1999 at 03:54:48, Ed Schröder wrote:
>>Posted by Christophe Theron on November 22, 1999 at 18:25:35:
>>
>>>>Was it better than Sargon II, or just equal?
>>>
>>>It was better because Sargon was outplayed by search depth in most
>>>cases. In that days Rebel was able to look 6 plies deep all very selective
>>>and much holes involved but very effective playing a program thinking
>>>just 4 plies deep.
>>
>>6 plies on a TRS-80 in the time Sargon took to compute 4?
>>
>>I have the old Sargon II for TRS-80 manual just in front of me. A green manual:
>>"Hayden computer program tapes, Sargon II: A computer chess program by Dan and
>>Kathe Spracklen". I have kept it as a souvenir (I also have the cassette,
>>but I don't dare to open the box).
>>
>>The manual says that it took 6 minutes (average) to compute to ply depth 4.
>>
>>You are saying that in 6 minutes you were able to compute to ply depth 6 on a
>>TRS-80???
>
>6 plies indeed all very selective but a friend of mine had doubled the
>processor speed from 1.77 Mhz to 3.5 Mhz. On standard 1.77 Mhz Rebel on
>40/120 was only able to compute 4 plies and some moves on 6 plies.
>
>The first Rebel was a strange animal. I had to re-invent the wheel
>completely as I had no access to documentation, I wasn't even aware
>documentation existed.
>
>No alpha/beta, no windows, no Q-search, the program could only think
>in steps of 2 plies. Thus 2,4,6,8 etc.
???
How did it work? Why the always even depth?
>>If I had known that at that time, I would have immediately given up chess
>>programming. My TRS-80 was able to compute 3 plies in 6 minutes, and I was
>>disgusted by Sargon's 4 plies in the same time. Not to talk about Sargon's
>>superior evaluation...
>
>I agree, Sargon's evaluation was the best of all.
I think it was only piece square tables (probably one for opening, one for
middlegame and one for the endgame), with a dynamic evaluation for the pawn
structure. That's all, and that's why they were so fast.
>>>>>In the past I have written an Apple IIE emulator for the PC.
>>>>
>>>>That's incredible! Did you write it just to run a 6502 Rebel with more
>>>>comfort, or did you make a standalone product of this emulator?
>>>
>>>The emulator was only meant to work for my convenience improving
>>>Rebel. No Apple 2E hard disk, keyboard or screen support, it only
>>>emulated my engine.
>>
>>
>>You mean you were able to run your 6502 binaries on the PC I suppose?
>
>I used a PC 6502 cross compiler to generate the binary then my emulator
>directly executed the 6502 code.
Nice. That's of course much faster than cross-compiling, then downloading to the
Apple (or the standalone 6502 prototypes).
Never had problems with subtle differences between the real thing and the
emulator?
>>>It was not used very much as the so-called "dedicated computer" market
>>>collapsed after the introduction of the 80486 as then it was the first
>>>time producers like Mephisto, Fidelity and Saitek couldn't compete anymore
>>>regarding raw processor speed as the 486 definitely was superior to the
>>>chips they were using.
>>>
>>>The only exception was TASC who came with the ChessMachine concept and
>>>therefore was able to compete a bit longer but then when the Pentium
>>>90 came the hardware superiority once and for all was decided in favor
>>>of the PC.
>>>
>>>Summarizing my conversions:
>>>- TRS80 (basic)
>>>- TRS80 (assembler)
>>>- APPLE 2E (assembler)
>>>- ARM RISC (assembler)
>>>- PC (C++)
>>>- PC (assembler)
>>>
>>>Every stage was about 1.5 - 2 years before I was pleased with the results.
>>>
>>>I wonder what is next. The upcoming 64-bit of Intel?
>>
>>
>>Porting to C to take advantage of all the optimizations the modern
>>compilers can offer I assume.
>>
>>
>>
>>>>> I can run and old
>>>>>6502 Rebel on the PC. It's a super fast emulator, speed loss not more than
>>>>>a factor 2-3 if I remember well. At that time I used it for a while to
>>improve
>>>>>my 6502 engines because a 386 20 Mhz was much more comfortable than the
>>>>>Apple IIE at 2 Mhz.
>>>>>
>>>>>If I can dig up the thing in 2010 I will give you a very hard time :)
>>>>
>>>>At that time you'll have a 1GHz Apple IIe.
>>>>
>>>>Check your sources. It would be sad to see you resigning because you get a
>>>>stack overflow after a 20 plies search! :)
>>>
>>>Worse, 16 plies maximum and no hash tables :(
>>
>>
>>What happened after 16 plies?
>
>Just a static evaluation of the position. In these days Q-search was
>unknown for me.
That's funny. That's exactly what my first program did. At the horizon it used a
static exchange evaluator to stop the search and give an evaluation.
The SEE was complex and time consuming. I could compute only 10 positions per
second. Pathetic.
But writting the SEE in itself was already a good challenge.
I have been using a similar technique until 1993.
The fact that a QSearch runs very quickly is counter intuitive. Common sense
says that it can have to compute complex sequences of capture/capture on several
plies with a branching factor of 2 or 3.
Without trying it actually, I would still be using the inferior SEE only
approach.
>It was not a big deal, on a TRS-80 3.5 Mhz Rebel on 40/120 was doing
>6 plies and some interesting moves even hit 8 plies. 10 plies was out
>of the question.
Yes. Without QSearch and extensions, there was no risk to go very deep... :)
Christophe
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.