Crushed! Thinker vs. Colossus, 26-4

[Mike S.]

I found the following ratings (32-bit single cpu):

Code: Select all

           Col.2008b  Thi.54a/i   Diff. ~Exp.
---------------------------------------------
CCRL 40/4    2755       2938(*)   +183   74%
CEGT 40/20   2679       2898(**)  +219   78%
CCRL 40/40   2742       2938      +196   75%

*) from version 5.3b inert
**) 64 bit version; Thinker doesn't seem to gain much from it

Colossus is a good engine! Next to the ratings of version 2008b we find good old engines, like various Chessmaster 10 settings, Tiger 15 or -2004, or SlowChessBlitz WV2.1. Nevertheless...

6m+4s, Intel D945 3.4 GHz (~P4)
128 MB hash each, pondering off
Windows Vista, Fritz 10 interface
generic opening book from GM games
set to max. 15 moves, min. 5 games
3/4 + some 5 piece tbs. (for Colossus)
files on flash mem., 128 MB tbs. cache

Thanks to Olivier Deville for his collection of recent GM games,
http://www.open-aurec.com/chesswar/download.html

(I started a match between these two, because I couldn't includ them in a test on my other computer, where I still have Win98 and they cannot run there.)

Code: Select all

6m+4s, D945/3.4, 2009
                        
1   Thinker 5.4a inert  +23/-1/=6 86.67%   26.0/30 --> perf. +318 Elo
2   Colossus 2008b      +1/-23/=6 13.33%    4.0/30

Brutal. In terms of Elo relative to Colossus' ratings, this is a performance above Rybka 2.x level. Maybe Thinker is an extra good Anticolossus? As far as I could see, everything was technically ok (CPU usage, hash memory allocation, time consumption). Thinker's white score was 14.0/15, +13=2-0. Fabulous!

I had planned to play a 40 games match, but decided to stop the slaughtering after 30 games. Colossus won only once, in the second game:

[d]r1r3k1/pp1q1pp1/3pb2p/P2p4/7R/1PBQ2P1/3PPR1P/6K1 w - - 0 23

Analysis by Colossus 2008b:

23.Bxg7 Kxg7
-+ (-1.78) Depth: 1/1 00:00:00
-+ (-1.78) Depth: 1/1 00:00:00
23.e3
+/= (0.56) Depth: 1/1 00:00:00
(...)
23.Rf1 Rc5 24.Bd4 Rc6 25.Kg2 a6 26.b4 f5 27.Kh1 Rac8 28.Rh5
= (-0.10) Depth: 9/26 00:00:00 587kN
= (-0.10) Depth: 11/31 00:00:03 2878kN
23.Bxg7 f5 24.Bxh6 Rc1+ 25.Kg2 Kf7 26.Qd4 Kg8 27.Rf1 Rc5 28.b4 Rc2
+/= (0.26) Depth: 11/31 00:00:22 17647kN
+- (2.13) Depth: 11/31 00:01:01 47067kN

(in the game, it was played with +2.25/11 after 39 seconds)

Two quick Thinker wins (resign was set to "late" in the GUI):

[Event "6m+4s, D945/3.4"]
[Site "COMPUTER-11"]
[Date "2009.01.10"]
[Round "23"]
[White "Thinker 5.4a inert"]
[Black "Colossus 2008b"]
[Result "1-0"]
[ECO "D03"]
[PlyCount "49"]

1. d4 Nf6 2. Nf3 g6 3. Bg5 Bg7 4. Nbd2 d5 5. e3 O-O 6. c3 h6 7. Bh4 Bf5 8. Qb3
b6 9. Ne5 Ne4 10. Nxe4 Bxe4 11. O-O-O Bf5 12. Be2 Bxe5 13. dxe5 c6 14. g4 Bc8
15. Qb4 Re8 16. Qf4 Qc7 17. Bd3 Ba6 18. Bxg6 fxg6 19. Qxh6 Rf8 20. Qxg6+ Kh8
21. e6 Qe5 22. Bxe7 Rxf2 23. g5 Qg7 24. Qe8+ Qg8 25. Bf6+ 1-0

[Event "6m+4s, D945/3.4"]
[Site "COMPUTER-11"]
[Date "2009.01.10"]
[Round "24"]
[White "Colossus 2008b"]
[Black "Thinker 5.4a inert"]
[Result "0-1"]
[ECO "D03"]
[PlyCount "66"]

1. d4 Nf6 2. Nf3 g6 3. Bg5 Bg7 4. Nbd2 d5 5. e3 O-O 6. c3 h6 7. Bxf6 exf6 8.
Qb3 Nc6 9. c4 dxc4 10. Bxc4 Na5 11. Qc3 Nxc4 12. Qxc4 Be6 13. Qb5 b6 14. O-O f5
15. Rac1 Re8 16. Nc4 a6 17. Qb4 Bd5 18. Rfe1 Bxf3 19. gxf3 f4 20. Qd2 Qg5+ 21.
Kh1 Rad8 22. Rg1 Qh5 23. Rg4 b5 24. Na5 Qh3 25. Qe2 Rxd4 26. Nc6 Rd6 27. Na5
Bxb2 28. Rc2 Bd4 29. Nb3 c5 30. Nxc5 Bxc5 31. Rxc5 Red8 32. Rg1 Rd2 33. Qf1
Qxf3+ 0-1

[pichy]

I found the following ratings (32-bit single cpu):

Code: Select all

           Col.2008b  Thi.54a/i   Diff. ~Exp.
---------------------------------------------
CCRL 40/4    2755       2938(*)   +183   74%
CEGT 40/20   2679       2898(**)  +219   78%
CCRL 40/40   2742       2938      +196   75%

*) from version 5.3b inert
**) 64 bit version; Thinker doesn't seem to gain much from it

Colossus is a good engine! Next to the ratings of version 2008b we find good old engines, like various Chessmaster 10 settings, Tiger 15 or -2004, or SlowChessBlitz WV2.1. Nevertheless...

6m+4s, Intel D945 3.4 GHz (~P4)
128 MB hash each, pondering off
Windows Vista, Fritz 10 interface
generic opening book from GM games
set to max. 15 moves, min. 5 games
3/4 + some 5 piece tbs. (for Colossus)
files on flash mem., 128 MB tbs. cache

Thanks to Olivier Deville for his collection of recent GM games,
http://www.open-aurec.com/chesswar/download.html

(I started a match between these two, because I couldn't includ them in a test on my other computer, where I still have Win98 and they cannot run there.)

Code: Select all

6m+4s, D945/3.4, 2009
                        
1   Thinker 5.4a inert  +23/-1/=6 86.67%   26.0/30 --> perf. +318 Elo
2   Colossus 2008b      +1/-23/=6 13.33%    4.0/30

Brutal. In terms of Elo relative to Colossus' ratings, this is a performance above Rybka 2.x level. Maybe Thinker is an extra good Anticolossus? As far as I could see, everything was technically ok (CPU usage, hash memory allocation, time consumption). Thinker's white score was 14.0/15, +13=2-0. Fabulous!

I had planned to play a 40 games match, but decided to stop the slaughtering after 30 games. Colossus won only once, in the second game:

[d]r1r3k1/pp1q1pp1/3pb2p/P2p4/7R/1PBQ2P1/3PPR1P/6K1 w - - 0 23

Analysis by Colossus 2008b:

23.Bxg7 Kxg7
-+ (-1.78) Depth: 1/1 00:00:00
-+ (-1.78) Depth: 1/1 00:00:00
23.e3
+/= (0.56) Depth: 1/1 00:00:00
(...)
23.Rf1 Rc5 24.Bd4 Rc6 25.Kg2 a6 26.b4 f5 27.Kh1 Rac8 28.Rh5
= (-0.10) Depth: 9/26 00:00:00 587kN
= (-0.10) Depth: 11/31 00:00:03 2878kN
23.Bxg7 f5 24.Bxh6 Rc1+ 25.Kg2 Kf7 26.Qd4 Kg8 27.Rf1 Rc5 28.b4 Rc2
+/= (0.26) Depth: 11/31 00:00:22 17647kN
+- (2.13) Depth: 11/31 00:01:01 47067kN

(in the game, it was played with +2.25/11 after 39 seconds)

Two quick Thinker wins (resign was set to "late" in the GUI):

[Event "6m+4s, D945/3.4"]
[Site "COMPUTER-11"]
[Date "2009.01.10"]
[Round "23"]
[White "Thinker 5.4a inert"]
[Black "Colossus 2008b"]
[Result "1-0"]
[ECO "D03"]
[PlyCount "49"]

1. d4 Nf6 2. Nf3 g6 3. Bg5 Bg7 4. Nbd2 d5 5. e3 O-O 6. c3 h6 7. Bh4 Bf5 8. Qb3
b6 9. Ne5 Ne4 10. Nxe4 Bxe4 11. O-O-O Bf5 12. Be2 Bxe5 13. dxe5 c6 14. g4 Bc8
15. Qb4 Re8 16. Qf4 Qc7 17. Bd3 Ba6 18. Bxg6 fxg6 19. Qxh6 Rf8 20. Qxg6+ Kh8
21. e6 Qe5 22. Bxe7 Rxf2 23. g5 Qg7 24. Qe8+ Qg8 25. Bf6+ 1-0

[Event "6m+4s, D945/3.4"]
[Site "COMPUTER-11"]
[Date "2009.01.10"]
[Round "24"]
[White "Colossus 2008b"]
[Black "Thinker 5.4a inert"]
[Result "0-1"]
[ECO "D03"]
[PlyCount "66"]

1. d4 Nf6 2. Nf3 g6 3. Bg5 Bg7 4. Nbd2 d5 5. e3 O-O 6. c3 h6 7. Bxf6 exf6 8.
Qb3 Nc6 9. c4 dxc4 10. Bxc4 Na5 11. Qc3 Nxc4 12. Qxc4 Be6 13. Qb5 b6 14. O-O f5
15. Rac1 Re8 16. Nc4 a6 17. Qb4 Bd5 18. Rfe1 Bxf3 19. gxf3 f4 20. Qd2 Qg5+ 21.
Kh1 Rad8 22. Rg1 Qh5 23. Rg4 b5 24. Na5 Qh3 25. Qe2 Rxd4 26. Nc6 Rd6 27. Na5
Bxb2 28. Rc2 Bd4 29. Nb3 c5 30. Nxc5 Bxc5 31. Rxc5 Red8 32. Rg1 Rd2 33. Qf1
Qxf3+ 0-1

I had not found any free engine that could beat Thinker 5.4 Inert in the last two weeks, but Rybka v2.2 32 is very close.

[pichy]

[quote="Mike S."]I found the following ratings (32-bit single cpu):

Code: Select all

           Col.2008b  Thi.54a/i   Diff. ~Exp.
---------------------------------------------
CCRL 40/4    2755       2938(*)   +183   74%
CEGT 40/20   2679       2898(**)  +219   78%
CCRL 40/40   2742       2938      +196   75%

*) from version 5.3b inert
**) 64 bit version; Thinker doesn't seem to gain much from it

Colossus is a good engine! Next to the ratings of version 2008b we find good old engines, like various Chessmaster 10 settings, Tiger 15 or -2004, or SlowChessBlitz WV2.1. Nevertheless...

6m+4s, Intel D945 3.4 GHz (~P4)
128 MB hash each, pondering off
Windows Vista, Fritz 10 interface
generic opening book from GM games
set to max. 15 moves, min. 5 games
3/4 + some 5 piece tbs. (for Colossus)
files on flash mem., 128 MB tbs. cache

Thanks to Olivier Deville for his collection of recent GM games,
http://www.open-aurec.com/chesswar/download.html

(I started a match between these two, because I couldn't includ them in a test on my other computer, where I still have Win98 and they cannot run there.)

Code: Select all

6m+4s, D945/3.4, 2009
                        
1   Thinker 5.4a inert  +23/-1/=6 86.67%   26.0/30 --> perf. +318 Elo
2   Colossus 2008b      +1/-23/=6 13.33%    4.0/30

Brutal. In terms of Elo relative to Colossus' ratings, this is a performance above Rybka 2.x level. Maybe Thinker is an extra good Anticolossus? As far as I could see, everything was technically ok (CPU usage, hash memory allocation, time consumption). Thinker's white score was 14.0/15, +13=2-0. Fabulous!

I had planned to play a 40 games match, but decided to stop the slaughtering after 30 games. Colossus won only once, in the second game:


I usually play matches up to 100 games, but decided to play only 50 games instead:

Thinker 5.4 Inert has not played against Rybka V2.2 yet, but I believe that it will come ahead of the free one cpu engines once graham match one vs the other.

http://www.computerchess.org.uk/ccrl/40 ... ons_only=1
Engine Score SP Gl S-B
1: SP-x86-Inert---Thinker 31.0/50 ·················································· 1=1=1==110=1=1=0111=0=1=000001=1=1111=11011=100110 589.00
2: Glaurung-w32 19.0/50 0=0=0==001=0=0=1000=1=0=111110=0=0000=00100=011001 ·················································· 589.00

50 games played / Tournament is finished
Name of the tournament: Arena tournament
Site/ Country: JORGE, United States
Level: Tournament Game in 5 Minutes
Hardware: AMD Sempron 3800+ AM2 with 1GB DDR2 RAM Memory
Operating system: Microsoft Windows XP Professional Service Pack 2 (Build 2600)
PGN-File: C:\Program Files\Arena\Arena.pgn
Website:
E-Mail Address:

[Mike S.]

I just saw that at CEGT, single Thinker 5.4a/i is rated approx. at Shredder 11 level. There is not a long way to go until Thinker could be a successful commercial engine, interesting for every company, if it is enhanced by:

1. standard information output
2. UCI
3. multi-pv

(and desireable also, Nalimov access).

I would not be surprised if Rybka, Naum and Thinker will be the top-3 (all categories), in 2010. It is not unusual that the "engine landscape" changes dramatically, over the years. Not just considering Shredder, Fritz etc., but before them, programs like Chess Genius or Mchess ruled, were superseeded and disappeared from top competition. The history repeats itself.