Speed of chess programs?
Jan Eric Larsson
Hi!
I'm well aware that speed is far from everyhting, and that it is
difficult to compare different computers, but still I wonder
whether anyone has gathered any statistics about the following:
1. What are the relative "chess" speed differences between a
SPARC 10, a SPARC 20, a Pentium 60, a Pentium 100, and a
486/60?
2. Is there some kind of list of current fast ones, like Stobor,
Don Dailey's latest, etc.? (For example in nps).
Thanks beforehand!
Jan Eric Larsson Phone: +1 415 725 3859
Knowledge Systems Laboratory Fax: +1 415 725 5850
Department of Computer Science E-mail: Lar...@KSL.Stanford.Edu
Stanford University
701 Welch Road, Building C "We watched the thermocouples dance to the
Palo Alto, CA 94304, USA spirited tunes of a high frequency band."
Robert Hyatt
I can give you two results for Crafty, if anyone else has downloaded
Crafty, you, too can be a contributor to this. Would be useful info
for all of us:
two tests that are easy to do:
1. start crafty, and type "perf". This will run a pretty quick performance
test that first generates all legal moves 10,000 times and then computes the
number of moves generated per second. On the Sparc-20, this figure is
around 900K per second, and is around 600K on a pentium-100. It then
does the same thing again, but after generating each move, it makes it on
the board, to test the generate/make time. On the Sparc-20, this is around
200K, on the pentium 110K. If you have a pentium, run these tests and let
me know the results and machine type. Note that these are *low* estimates
of these numbers because they are from the initial board position, which
only has 4 knight and 16 pawn moves. running this test in middlegame
positions push these numbers up significantly. You can type "perf" at
any time without "busting" the program, just be sure it isn't "pondering"
(type ponder off) or you might rip something.
2. run crafty, and watch the "nps" figure. on the ss20, this starts at
around 10K and reaches 25K in some endings. On the pentium-100, it is
about 1/2 this speed. Note that I'm now optimizing, so I hope that these
numbers will increase substantially as I clean up from all the testing I've
done the past few months.
Current tests seems to say that the sparc-20 is about 2x a pentium-100 when
running Crafty. The Dec ALpha/150mhz is right between these two, the big
alpha is significantly faster than the sparc-20, but I haven't run perf
test on one in a while, if you have one, let me know. For Comparison, on
a full-blown Cray T90, Crafty searches about 5M nodes per second. The
perf figures aren't meaningful because that test is *not* parallelized
and therefore doesn't see the full benefit of 32 processors.
--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170