🚀 go-pugleaf

In article <9836@cs.tulane.edu> jlin@convex1.tcs.tulane.edu (Jonah Lin) writes:
>
>I am just curious, how many nodes does GNUchess search on a SPARCstation
>on average?
>Also, how much stronger will GNUchess be if it is running on a machine
>2-3 times faster than the SPARC?
>

Of course, it varies depending on what SPARCstation you run on (1, 1+, IPC,
SLC, ELC, IPX, 2 etc.).  On a SPARCstation 1+, a rough average in the middle
game appears to be 3000 to 4000 nodes per second.  The actual value is strongly
dependent on the position of course.  Surely sufficient speed exists in even
a SPARCstation 1+ to play strong master if not senior master chess -- but a
more refined program is needed to attain those levels.  I base this perhaps
controversial assessment on the fact that Motorola 68000 chip running at
16 Mhz (FAR FAR less of a cpu than a SPARCstation 1+ [which is in itself an
old, relatively slow processor these days]) with a good program is easily
a master level player (Fidelity Mach III Master, Mephisto Lyon 16 bit, others).

Guessing the improvement of a program's strength based on solely speeding up
the hardware is not an exact science either, but Larry Kaufmann (IM and
computer chess guru) estimates (I'm going from memory here ... forgive any
mistakes -- they're mine) that a doubling of processor speed typically means
that about 70 points are gained.  A speed increase of 5 typically enables a
program (another generalization coming) to search a ply deeper which usually
translates to about 200 rating points ... I hope I'm remembering these figures
correctly...  Subscribe to Computer Chess Reports (approx. $10 per year) for
LOTS of info on these types of topics.  Then you won't have to trust my
memory!


--
Disclaimer:  Any opinions expressed are probably my own, but certainly NOT
             those of my employer.

Roy Brunjes                  roy.brunjes@East.Sun.COM