scourge
October 5th, 2007, 12:00 PM
It's an Xboard chess engine called Sloppy: http://koti.mbnet.fi/~ilaripih/sloppy/
Yeah, I shamelessly borrowed the style sheets from the Bluefish homepage.
Anyway, Sloppy is licensed under GPL 3 and is better than the default xboard engine (Gnuchess) in many ways:
- stronger
- doesn't lose on time in "x moves in y minutes" games
- doesn't crash when a game reaches 300 moves
- doesn't go crazy if the game ends up being a draw by the 50-move rule
- Displays the full principal variation. Useful for analyzing games.
What I'd like you to do is run a couple of tests (would you kindly ;) ). First you'll need to compile Sloppy, which is as easy as downloading the source code, entering the right folder and typing "make". Then please run Sloppy and enter the following commands:
- "bench" - this will run Sloppy's benchmark
- "divide 7" - runs the parallel perft test to depth 7
- "quit" - quits the program
Then you should dump the contents of your console here. It would also be helpful to know the OS and the CPU you have. On my Feisty 64-bit Core 2 Duo E6600 pc it looks like this:
Sloppy 0.1.0 by Ilari Pihlajisto
Build date: Oct 5 2007
Debugging level: 1
Optimized for 64-bit
Initializing...
Found 2 CPUs
Using "book in memory" book mode
No opening book was found
Book learning ON
Hash table size: 36 MB
...Done
Type "help" to display a list of commands
White: bench
Running benchmark at search depth 8...
100% [========================]
Benchmark finished in 26.43 seconds.
Main nodes searched: 10298009
Quiescence nodes searched: 17290626
Total nodes per second: 1043640
Average branching factor: 2.72
Hash table hit rate: 12.35%
White: divide 7
0: a3 106743106
1: b3 133233975
0: c3 144074944
1: d3 227598692
1: f3 102021008
0: e3 306138410
1: g3 135987651
0: h3 106678423
1: a4 137077337
0: b4 134087476
1: c4 157756443
0: d4 269605599
0: f4 119614841
1: e4 309478263
0: g4 130293018
1: h4 138495290
0: Na3 120142144
1: Nc3 148527161
0: Nf3 147678554
1: Nh3 120669525
Perft(7): 3195901860 nodes.
Time: 12.89 seconds.
Processing speed: 247878838 nodes per second.
White: quit
The times and nodes per second values will be different, and the order of the moves in "divide" may also be. But the node counts, branching factor and hash table hit rate should be the same on every machine. The "divide" test uses all your cpus/cores, and the number before a move is the thread number that was used. So on a quad-core machine there should be threads 0 - 3, and they should get used somewhat evenly.
So the "divide" test should be a lot faster on multi-core systems, but "bench" shouldn't. Both of the tests should however get a great speed boost on 64-bit platforms.
Thanks for your help. If you want to play against Sloppy you should install Xboard: sudo apt-get install xboard.
Then something like: xboard -size Medium -fcp ./sloppy
Yeah, I shamelessly borrowed the style sheets from the Bluefish homepage.
Anyway, Sloppy is licensed under GPL 3 and is better than the default xboard engine (Gnuchess) in many ways:
- stronger
- doesn't lose on time in "x moves in y minutes" games
- doesn't crash when a game reaches 300 moves
- doesn't go crazy if the game ends up being a draw by the 50-move rule
- Displays the full principal variation. Useful for analyzing games.
What I'd like you to do is run a couple of tests (would you kindly ;) ). First you'll need to compile Sloppy, which is as easy as downloading the source code, entering the right folder and typing "make". Then please run Sloppy and enter the following commands:
- "bench" - this will run Sloppy's benchmark
- "divide 7" - runs the parallel perft test to depth 7
- "quit" - quits the program
Then you should dump the contents of your console here. It would also be helpful to know the OS and the CPU you have. On my Feisty 64-bit Core 2 Duo E6600 pc it looks like this:
Sloppy 0.1.0 by Ilari Pihlajisto
Build date: Oct 5 2007
Debugging level: 1
Optimized for 64-bit
Initializing...
Found 2 CPUs
Using "book in memory" book mode
No opening book was found
Book learning ON
Hash table size: 36 MB
...Done
Type "help" to display a list of commands
White: bench
Running benchmark at search depth 8...
100% [========================]
Benchmark finished in 26.43 seconds.
Main nodes searched: 10298009
Quiescence nodes searched: 17290626
Total nodes per second: 1043640
Average branching factor: 2.72
Hash table hit rate: 12.35%
White: divide 7
0: a3 106743106
1: b3 133233975
0: c3 144074944
1: d3 227598692
1: f3 102021008
0: e3 306138410
1: g3 135987651
0: h3 106678423
1: a4 137077337
0: b4 134087476
1: c4 157756443
0: d4 269605599
0: f4 119614841
1: e4 309478263
0: g4 130293018
1: h4 138495290
0: Na3 120142144
1: Nc3 148527161
0: Nf3 147678554
1: Nh3 120669525
Perft(7): 3195901860 nodes.
Time: 12.89 seconds.
Processing speed: 247878838 nodes per second.
White: quit
The times and nodes per second values will be different, and the order of the moves in "divide" may also be. But the node counts, branching factor and hash table hit rate should be the same on every machine. The "divide" test uses all your cpus/cores, and the number before a move is the thread number that was used. So on a quad-core machine there should be threads 0 - 3, and they should get used somewhat evenly.
So the "divide" test should be a lot faster on multi-core systems, but "bench" shouldn't. Both of the tests should however get a great speed boost on 64-bit platforms.
Thanks for your help. If you want to play against Sloppy you should install Xboard: sudo apt-get install xboard.
Then something like: xboard -size Medium -fcp ./sloppy