This article is from the GNU Chess and XBoard FAQ, by Tim Mann firstname.lastname@example.org with numerous contributions by others.
It would be irresponsible to answer this question with a number, without
first explaining a few things about ratings.
The ratings that are commonly given for computer chess players are less
meaningful than they may seem. Most computer chess players (including
GNU Chess) do not play in tournaments against humans, or do so only
rarely, so they do not have official ratings from FIDE, USCF, or other
Some people have methods for rating chess programs approximately by
giving them a set of problems to work on and seeing how they do, or by
having them play tournaments against each other. Any rating number
produced by such means should be taken with a grain of salt; it may be
only a rough approximation to the rating the program would achieve in
over-the-board tournament competition against humans. The chess skills
required for solving problems or playing against other computers are not
necessarily the same as those required for play against humans. Also, of
course, tournaments among computers can rate the computers only relative
to one another, not relative to humans. Some of the computers need to be
rated by other means to give the ratings a basis to start from.
Compared with human players, computer players are strong tactically but
weak strategically, and are much better at blitz chess than at slow
chess. These differences make it more difficult to assign a meaningful
Several computers do play regularly on the Internet chess servers and
have achieved ratings there. These ratings have the advantage of being
based on many games. On the other hand, ICS ratings are only roughly
comparable to USCF or FIDE ratings. Many players have ICS ratings that
are hundreds of points higher or lower than their USCF or FIDE ratings.
Finally, unlike dedicated chess machines, or PC chess programs that run
on only a few different models of Intel processors, GNU Chess runs on
many different kinds of CPU at many different speeds. Thus its strength
depends on how fast a machine you run it on and how much optimization
your C compiler does. Some people have formulas for estimating how a
computer player's rating varies on faster or slower machines---see the
rec.games.chess FAQ for more information---but these need to be taken
with a grain of salt too.
All that said, here are some numbers.
- On the Internet Chess Club, a copy of GNU Chess running on an SGI Onyx
R4400 under the handle MaxII once achieved a blitz rating of over 2500
and a standard rating of over 2300. Current ICC and FICS ratings for
computers using GNU Chess tend to be a good deal lower.
- Wolfgang Gabriel ran the Bednorz-Toennissen Test BT2630 with GNU Chess
4.0 pl74 on a 60 MHz Pentium with 16 MB of RAM. The test gave an
estimated rating of 2213. He also ran Fritz-2 on the same hardware and
got an estimated rating of 2311.