10 years of Computer Chess

[Houdini]

It is also interesting to know if shredder may get more draws at longer time control.

Houdini can only win 99% of the games because Shredder commits a lot of mistakes. If you increase the TC, less mistakes will be made.

It would be interesting to have an idea of the equivalent TC on 2003 hardware (Pentium 4?) of Kai's 1'+1" match.

Robert

[Laskos]

The beast of the early 2003, undisputed SSDF leader, undisputed Playchess Engine Room favorite Shredder 7.04 stands up against the current leader:



A game

Code: Select all

[Event "10 years"]
[Site "PC"]
[Date "2012.11.08"]
[Round "75"]
[White "Houdini 3 Pro x64"]
[Black "Shredder 7.04"]
[Result "1-0"]
[ECO "B90"]
[Annotator "0.40;0.47"]
[PlyCount "57"]
[TimeControl "1+1"]

{W=13.2 plies; 1,
077kN/s; Perfect2012t.ctg  B=9.2 plies; 535kN/s; Perfect2012t.ctg} 1. e4 {B 0}
c5 {B 0} 2. Nf3 {B 0} d6 {B 0} 3. d4 {B 0} cxd4 {B 0} 4. Nxd4 {B 0} Nf6 {B 0}
5. Nc3 {B 0} a6 {B 0} 6. Be3 {B 0} e5 {B 0 Both last book move} 7. Nf3 {0.40/
15 2} Be7 {0.47/10 1 (Nc6)} 8. Bc4 {0.23/15 6 (Be2)} O-O {0.37/10 2 (Be6)} 9.
Qd3 {0.34/15 3 (0-0)} Nc6 {0.40/9 1 (Be6)} 10. a3 {0.33/14 1 (0-0)} Bg4 {0.29/
8 1} 11. h3 {0.52/14 2} Bxf3 {0.33/9 1 (Be6)} 12. gxf3 {0.47/13 0} b5 {0.44/10
1} 13. Ba2 {0.46/15 2} Qd7 {0.33/9 1 (Nd7)} 14. Rg1 {0.68/14 1 (0-0-0)} Qxh3 {
0.40/8 1} 15. O-O-O {0.61/16 1} b4 {0.67/8 2 (g6)} 16. Nd5 {1.84/13 2} Nxd5 {
1.16/9 2 (bxa3)} 17. exd5 {3.21/11 0} b3 {1.63/9 1 (Nd4)} 18. Bxb3 {4.46/11 0 
(Rh1)} Nd4 {4.67/10 5} 19. Bxd4 {4.46/10 0} Qc8 {4.81/10 2 (g6)} 20. Bc3 {5.38/
14 2 (Rh1)} Bf6 {4.84/8 1} 21. f4 {5.66/16 3 (Rg4)} Re8 {5.54/8 1} 22. Ba4 {6.
32/14 1 (Rde1)} Re7 {5.74/8 1 (Rd8)} 23. fxe5 {7.35/15 1 (Rde1)} Bxe5 {6.11/10
1} 24. Bxe5 {7.36/14 0} Rxe5 {5.72/10 1} 25. f4 {7.19/13 0} Rh5 {5.84/10 1} 26.
Bc6 {7.19/12 0 (Rde1)} a5 {8.40/10 5 (Rh3)} 27. Bxa8 {10.62/11 0} Qxa8 {8.93/9
1 (g6)} 28. Rxg7+ {9.29/10 0} Kxg7 {8.93/10 1} 29. Qc3+ {8.95/9 0} 1-0

[D]q7/5pkp/3p4/p2P3r/5P2/P1Q5/1PP5/2KR4 b - - 0 29[/D]

Observing the games, Houdini outplays Shredder in the opening, if not finished, in the middlegame, and if not finished, then surely in the endgame. Tremendous achievement in these less than 10 years.

It is clear that houdini3 outsearches Shredder7.04 and get bigger depths.

I wonder what is the hardware that you used and how many cpu did the engines use?

It is also interesting to know if shredder may get more draws at longer time control.

1 core each engine (ponder off) on a 2 core AMD x2. Houdini 3 sure outsearches Shredder 7.04, but it seems to me that it evaluates better too (often the engines are diverging in their evaluations). I will not test at longer TC, as it will probably take 100 or more long games to be conclusive.

[hammerklavier]

Interesting .... but 1 + 1 TC is ridiculous. Try the same with 40/120.

[mclane]

indeed

http://talkchess.com/forum/viewtopic.ph ... ini+komodo

[thierrycatalan]

I am surprised by such a score...
I had kept memory than schredder 7 was an excellent software, able to beat many human GMI
I knew that houdini was stronger, but I would have never thought to 99.5 points / 100

Finally between chesssoftware, list elo ratings have a certain reality ...
and 10 years show progress very well.

[Laskos]

Interesting .... but 1 + 1 TC is ridiculous. Try the same with 40/120.

What's so ridiculous? That's slower than Playchess Engine Room bullet. Several good rating lists have 5 + 3 or 40/4 controls, which are only ~4 times slower, besides that, much of engine development is about testing at 10-20 times faster controls like 3'' + 0.05''

[jplchess]

Give the TC a compromise: G/10 minutes for 50 games.

There will be more hardware changes than software changes in the 21st century. In terms of software, "we are running out of change". The easy ideas have already been figured out.

If you did 20 years of Computer Chess, try Mephisto Genius version 2 or version 3.

The humble chessplayer,
Jonathan Lee

[Adam Hair]

It is also interesting to know if shredder may get more draws at longer time control.

Houdini can only win 99% of the games because Shredder commits a lot of mistakes. If you increase the TC, less mistakes will be made.

It would be interesting to have an idea of the equivalent TC on 2003 hardware (Pentium 4?) of Kai's 1'+1" match.

Robert

Probably no more than 2'+2". The fastest desktop processor available when Shredder 7.04 came out appears to be the Pentium 4 3.2 Northwood. Kai said he used an AMD X2. According to this Fritzmark comparison, the X2 single core does not appear to be much faster than the Pentium 4.

[Laskos]

Give the TC a compromise: G/10 minutes for 50 games.

There will be more hardware changes than software changes in the 21st century. In terms of software, "we are running out of change". The easy ideas have already been figured out.

Well, last 10 years were 21st century and your statement is not quite right. An i5 quad core is no more than 12 times faster than a reasonably good Pentium or AMD of 2003, that's some 3.5 doublings and no more than 300 Elo points improvement. Software from Shredder 7.04 to Houdini 3 improved by 600 points. Therefore, until now, in the 21st century, software beats hardware. I don't know if that can stay for long, but I wouldn't be surprised if the next 10 years will bring another 400-500 points purely from software.

If you did 20 years of Computer Chess, try Mephisto Genius version 2 or version 3.

The humble chessplayer,
Jonathan Lee

[Cubeman]

Impressive results, but you mentioned that in many games Houdini got advantage out of the opening. Was Houdini using a different more modern book and Shredder was left with it's own book from 2003 ? Or were they both using the same book?