Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The law of diminishing returns

Author: K. Burcham

Date: 14:53:47 07/12/02

Go up one level in this thread





ED, I have also thought very much about this same thing.
The reason for my interest is because of what it means when
2 opponents play each other in a long standard game, when one
of the opponents has small hardware. This small hardware, if the game
is long enough, will still reach respectable depths. But the question
in this situation for me would be how much is being stored in hash to make
the next move if you have large mhz and large hash. in other words,
to see what i am talking about here, i have always wanted to run two
computers with same program on strong hardware. when playing this game
with two computers, lets say both have 2000 mhz amd with 1 gig ram and hash set
at 432 megs. but on computer A, select "clear hash after every move".
on computer B select "store moves". in this example it would be interesting to
see how many times in an actual game against a GM or another program using
these two computers, that computer A would not choose the same move as computer
B, if most all moves in a game were to take 3 to 5 minutes per move, such as a
120/10 game.

I have noticed when i play on one of the game servers using 2x1533mhz,
with smp program, if i play against another program with 120/10 time
control, most moves are still made in 2 to 5 minutes after book moves
are played. yes some moves might take 5 to 8 minutes, but not many. when playing
against another program, it seems the opponent's move is expected most of the
time. when playing against a human GM, lots of moves are not expected by the
program, and the eval depth drops after making the unexpected GM move. if the GM
plays a line that the program has not been looking at in its past moves the
depth stays very low with all programs, if the program has looked at most of the
moves involved with the GM move then the program depth will stay high in eval.
so it seems the hardware of today cannot play at the depths in your chart where
you show 0 change in moves selected.

i do not doubt your accuracy at all, but i think in most games against other
programs with strong hardware we will not see the same results as your chart,
because your program will not have looked at the expected move by the opponent
in a real game, especially against a human. and at depths of between 11 and 16,
your chart shows lots of changes your program might make at these depths. also
as each one of these changes are made, it would seem that the possibility of
more changes would increase substantially with each following change made by the
program in an actual game. in other words, one change will lead to another
change, to another change. where as if it is just a test position, then i can
see where your chart could show your statistics.

it has been proven, many many times this year, that any of the top programs on
the strongest hardware, 2500 mhz intel, or 1800 mhz amd, both with very large
hash, in a very long time control game----either program can lose or win  at any
time, and i dont mean the games where one program wins because of book.

so it will be interesting to see how the draws increase in games over the next
five years as hardware and mhz improves. and i fully expect the human to program
win loss ratio to change with these hardware improvements.

the chart is very interesting, it reminds me of some negative comments that
Chris has made about upgrading mhz.

kburcham





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.