Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Is Hardware Over -valued? Why is Diep doing so Poorly?

Author: Robert Hyatt

Date: 13:15:04 07/11/02

Go up one level in this thread


On July 11, 2002 at 13:58:40, Keith Evans wrote:

>On July 11, 2002 at 13:01:14, Robert Hyatt wrote:
>
>>On July 11, 2002 at 12:19:40, Keith Evans wrote:
>>
>>>On July 11, 2002 at 11:57:05, Robert Hyatt wrote:
>>>
>>>>On July 11, 2002 at 09:03:20, Rolf Tueschen wrote:
>>>>
>>>
>>>>>The Hsu team of IBM did avoid it by introducing GM Benjamin into the testings.
>>>>>IMO he had to check nonsense and contradictions of the play in relation to the
>>>>>books.
>>>>
>>>>Apples and oranges.  The DB guys _never_ tried to tune their program while
>>>>running at 1/1000th the speed of the real machine.  That was the problem we
>>>>fell into.  They had access to their machine all the time and could test in
>>>>any way they wanted.  We had little access to the Cray except right before
>>>>an annual computer chess tournament, so all our development was _forced_
>>>>onto a VAX, which was far slower.
>>>>
>>>
>>>How did you do your Cray development on a Vax? Did Cray supply an emulator?
>>>
>>>Regards,
>>>Keith
>>
>>
>>No.  We were a FORTRAN-based program, and we just ran the cray-optimized
>>version right on the vax after compiling with the vax compiler.  It was
>>horribly inefficient since the vax didn't know beans about vectors but it
>>ran perfectly, just slowly...
>>
>>We always had FORTRAN versions of everything, even if we rewrote things in
>>Cray Assembly for speed.  That way we could change the fortran first (a far
>>easier task) to check out a new idea.  If it worked, we then had to modify
>>the assembly code (a much harder task).
>
>So the only time that you had to test your Cray assembly code was right before a
>tournament? I guess that you carefully checked each others work then.
>
>Regards,
>Keith


We could do very short tests thruout the year, on occasion.  IE we had enough
time to confirm that the pure FORTRAN and the FORTRAN + assembly versions would
produce identical node counts for test positions.  But we could not get enough
time to play actual games for testing and tuning, very often.  That was a real
problem...



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.