Glaurung

[seemychess]

fine engine, great at playing position chess. What it is lacking is to be able to attack the king and win, i am sure if i could attack the opponents king Glaurung will go up about 100 ELO. so what do you think of Glaurung?

[Graham Banks]

fine engine, great at playing position chess. What it is lacking is to be able to attack the king and win, i am sure if i could attack the opponents king Glaurung will go up about 100 ELO. so what do you think of Glaurung?

Tord has said that Glaurung 2.0.1 is still in a basic stage and plays a less attractive game than Glaurung 1.2.1.
I therefore have no doubt that there is a lot of strength improvement to come.

[ernest]

At least, this is a program which was really built for 64-bit.
The only others I know are Rybka, Zappa (they are 1.6 to 1.9 times faster in 64-bit) and Crafty.

Glaurung 2.0.1 (Jim Ablett's compile) 64-bit ist 1.4 times faster than 32-bit.

Win XP Pro x64
1 thread (for reproducibility)
rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq -
Engine: Glaurung 2.0.1 64-bit (256 MB) by Tord Romstad
14.00 0:01 +0.23 1.e4 Nf6 2.Nc3 Nc6 3.Nf3 e6 4.Bb5 Bb4 5.d3 O-O 6.O-O d5 7.Bxc6 bxc6 8.Ne5 (1.896.048) 1663
15.01 0:02 +0.25 1.e4 Nf6 2.Nc3 Nc6 3.Nf3 e6 4.Bb5 Bb4 5.d3 O-O 6.O-O d5 7.Bxc6 bxc6 8.Ne5 Bb7 (3.838.438) 1671
16.01 0:07 +0.25 1.e4 e5 2.Nf3 Nf6 3.Nc3 Nc6 4.Bb5 Bc5 5.O-O O-O 6.Nxe5 Nxe5 7.d4 c6 8.dxe5 cxb5 9.exf6 Qxf6 10.Nxb5 (12.849.404) 1671
17.01 0:12 +0.25 1.e4 e5 2.Nf3 Nf6 3.Nc3 Nc6 4.Bb5 Bc5 5.O-O O-O 6.Nxe5 Nxe5 7.d4 c6 8.dxe5 cxb5 9.exf6 Qxf6 10.Nxb5 (20.820.260) 1676
18.01 0:29 +0.29 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Bc5 5.Nc3 O-O 6.Nxe5 Nxe5 7.d4 Bd6 8.dxe5 Bxe5 9.Bg5 d6 (49.616.889) 1677
19.01 0:53 +0.31 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Bc5 5.c3 O-O 6.d4 Be7 7.Re1 exd4 8.cxd4 d5 9.Bxc6 bxc6 10.Nc3 dxe4 11.Nxe4 Nd5 (90.450.482) 1682
20.01 1:43 +0.29 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Bc5 5.c3 O-O 6.d4 Be7 7.Re1 exd4 8.cxd4 d5 9.Bxc6 bxc6 10.Nc3 dxe4 11.Nxe4 Nxe4 12.Rxe4 Be6 (174.827.468) 1681
best move: e2-e4 time: 2:03.125 min n/s: 1.691.122 nodes: 208.140.000

Engine: Glaurung 2.0.1 32-bit (256 MB) by Tord Romstad
14.02 0:01 +0.23 1.e4 Nf6 2.Nc3 Nc6 3.Nf3 e6 4.Bb5 Bb4 5.d3 O-O 6.O-O d5 7.Bxc6 bxc6 8.Ne5 (1.896.048) 1093
15.01 0:03 +0.25 1.e4 Nf6 2.Nc3 Nc6 3.Nf3 e6 4.Bb5 Bb4 5.d3 O-O 6.O-O d5 7.Bxc6 bxc6 8.Ne5 Bb7 (3.838.438) 1142
16.01 0:10 +0.25 1.e4 e5 2.Nf3 Nf6 3.Nc3 Nc6 4.Bb5 Bc5 5.O-O O-O 6.Nxe5 Nxe5 7.d4 c6 8.dxe5 cxb5 9.exf6 Qxf6 10.Nxb5 (12.849.404) 1179
17.01 0:17 +0.25 1.e4 e5 2.Nf3 Nf6 3.Nc3 Nc6 4.Bb5 Bc5 5.O-O O-O 6.Nxe5 Nxe5 7.d4 c6 8.dxe5 cxb5 9.exf6 Qxf6 10.Nxb5 (20.820.260) 1189
18.01 0:41 +0.29 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Bc5 5.Nc3 O-O 6.Nxe5 Nxe5 7.d4 Bd6 8.dxe5 Bxe5 9.Bg5 d6 (49.616.889) 1198
19.01 1:15 +0.31 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Bc5 5.c3 O-O 6.d4 Be7 7.Re1 exd4 8.cxd4 d5 9.Bxc6 bxc6 10.Nc3 dxe4 11.Nxe4 Nd5 (90.450.482) 1204
20.01 2:24 +0.29 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.O-O Bc5 5.c3 O-O 6.d4 Be7 7.Re1 exd4 8.cxd4 d5 9.Bxc6 bxc6 10.Nc3 dxe4 11.Nxe4 Nxe4 12.Rxe4 Be6 (174.827.468) 1207
best move: e2-e4 time: 2:49.109 min n/s: 1.212.886 nodes: 205.110.000

[Mike S.]

In the following game, Glaurung played successfully on both wings, which is even better

[Event "Glaurung201-Test_A"]
[Site "Schrotty"]
[Date "2007.12.09"]
[Round "25"]
[White "Glaurung 2.0.1"]
[Black "Spike 1.2 Turin"]
[Result "1-0"]
[ECO "D37/22"]
[PlyCount "57"]

1. d4 Nf6 2. c4 e6 3. Nf3 d5 4. Nc3 Be7 5. Bf4 O-O 6. e3 c5 7. dxc5 Bxc5 8. a3
Nc6 9. Qc2 Qa5 10. O-O-O Be7 11. Kb1 Bd7 12. g4 dxc4 13. Bxc4 Rad8 14. h3 Qb6
15. g5 Ne8 16. b4 a6 17. Bd3 g6 18. Be2 Qa7 19. h4 a5 20. b5 Nb8 21. h5 b6 22.
hxg6 fxg6 23. Be5 Bc6 24. Rxd8 Bxd8 25. bxc6 Nxc6 26. Nb5 Qd7 27. Rd1 Qb7 28.
Qe4 Be7 29. Bc4 1-0

(last book move was 13.Bxc4)

It was Glaurung's only quick win (<35) in a 40 games match, nevertheless I hesitate to agree. Against strong opponents, maybe the king attacks you are missing just were inappropriate, most of the time? - I need to collect more Glaurung impressions though, to get a picture of it's style. The tactical speed seems very good though (as a basic requirement for aggressive play).

The match result was Glaurung 2.0.1 vs. Spike 1.2, 20.5-19.5 (blitz with typical depths of ~13, both on 1 CPU).

[Tord Romstad]

fine engine,

Thanks.

great at playing position chess.

Actually, there isn't much positional knowledge in the program, but the huge search depths you reach on modern computers often create an illusion of deep positional understanding.

What it is lacking is to be able to attack the king and win,

This is very true. The explanation is that my program still knows almost nothing about king safety. It will not attack the opponent's king unless it searches sufficiently deeply to se a forced win.

If you want king attacks, give the old Glaurung 1.2.1 a try. It plays very aggressively with the default settings, and if you want something even wilder, adjust the "Aggressiveness" parameter to the maximum value.

i am sure if i could attack the opponents king Glaurung will go up about 100 ELO.

Nowhere near that much, I think -- but it would certainly make the program a lot more interesting.

so what do you think of Glaurung?

My opinion is hardly the most objective, but I would say that it is still somewhat incomplete, but a reasonably solid foundation for building a good chess engine.

Tord has said that Glaurung 2.0.1 is still in a basic stage and plays a less attractive game than Glaurung 1.2.1.
I therefore have no doubt that there is a lot of strength improvement to come.

Thanks for the confidence, but I must admit that I currently have very little time and motivation to keep working on Glaurung. Perhaps this will change, I don't know.

At least, this is a program which was really built for 64-bit.

Yes, the current version is definitely built for high-end machines: 64-bit, and as many CPUs as possible. I don't think it will do well on single-CPU 32-bit computers.

Tord

[Dann Corbit]

At least, this is a program which was really built for 64-bit.

Yes, the current version is definitely built for high-end machines: 64-bit, and as many CPUs as possible. I don't think it will do well on single-CPU 32-bit computers.
Tord

I guess that it depends on your definition of 'well'. It is the 15th strongest engine in the world at 40/20:
http://www.husvankempen.de/nunn/40_40%2 ... liste.html

Code: Select all

no Program Elo + - Games Score Av.Op. Draws 
1 Rybka 2.3.2a w32 1CPU 2971 12 12 2186 67.8% 2841 37.6% 
2 Fritz 11 2921 10 10 2602 59.9% 2851 40.6% 
3 Deep Shredder 11 w32 1CPU 2889 15 15 1382 59.1% 2826 34.7% 
4 Fruit 2.3.3f Beta 2839 16 16 1164 48.5% 2850 36.9% 
5 Toga II 1.3.4 egbb 2829 16 16 1129 47.8% 2845 38.9% 
6 Hiarcs 11.1 1CPU 2828 13 13 1564 51.6% 2817 39.1% 
7 Naum 2.2 w32 1CPU 2827 17 17 998 52.6% 2809 41.2% 
8 Zappa Mexico w32 1CPU 2822 19 19 774 47.9% 2836 38.8% 
9 Loop 10.32f 2814 11 11 2193 48.8% 2823 42.0% 
10 Zap!Chess Zanzibar w32 1CPU 2792 17 17 1006 46.8% 2815 41.1% 
11 Deep Sjeng 2.7 1CPU 2772 14 14 1439 52.7% 2753 35.2% 
12 Spike 1.2 Turin 2770 10 10 2948 51.9% 2757 38.1% 
13 Junior 10.1 2763 24 24 539 42.9% 2812 33.2% 
14 Ktulu 8.0 2759 11 11 2591 49.4% 2763 34.8% 
15 Glaurung 2.0.1 w32 1CPU 2736 30 30 353 44.1% 2777 33.7% 
16 Chess Tiger 2007.1 2726 12 12 1806 46.4% 2751 40.4% 
17 SmarThink 1.00 x64 2710 13 13 1734 46.1% 2737 37.1% 
18 Alaric 707 2701 16 16 1326 49.4% 2705 30.5% 
19 CM10th R130 2692 28 28 385 48.7% 2701 34.0% 
20 Bright 0.2a 2684 25 25 515 51.7% 2672 32.6% 
21 The King 3.5 1CPU 2676 19 19 862 43.4% 2721 32.4% 
22 Scorpio 1.8 1CPU 2667 19 19 905 46.1% 2694 30.2% 
23 Movei 0.08.438 P10 2663 29 29 350 46.6% 2687 36.6% 
24 Gandalf 6.0 2659 7 7 6158 47.6% 2676 33.3% 
25 Delfi 5.1 2658 16 16 1307 49.1% 2665 30.8% 

But clearly added CPUs *are* a benefit as the CCRL 40/40 list shows:

Code: Select all

Rank Name Rating Score Average Opponent Draws Games LOS  ELO + − 
1 Rybka 2.3.2a 64-bit 4CPU 3119 +20 −19 71.3% −142.6 39.9% 903   100.0% 
2 Zappa Mexico 64-bit 4CPU 3064 +20 −20 64.6% −96.7 43.7% 813 99.9% 
3 Deep Shredder 11 64-bit 4CPU 3020 +19 −19 53.3% −22.8 43.7% 900 81.9% 
4 Naum 2.2 64-bit 4CPU 3008 +19 −19 54.7% −29.1 47.3% 833 84.3% 
5 Deep Fritz 10.1 4CPU 2993 +25 −25 53.4% −22.9 42.8% 512 62.2% 
6 Hiarcs 11.1 4CPU 2988 +18 −18 53.1% −21.2 43.6% 922 99.5% 
7 Loop M1-T 64-bit 4CPU 2956 +17 −17 49.0% +5.8 46.0% 1045 85.6% 
8 Glaurung 2.0.1 64-bit 4CPU 2935 +34 −34 50.6% −4.8 42.0% 264 88.4% 
9 Deep Junior 10 4CPU 2912 +17 −17 46.5% +26.0 35.6% 1128 92.1% 
10 Toga II 1.3.1 2897 +12 −12 49.0% +4.7 41.7% 2354 97.4% 
11 Deep Sjeng 2.7 4CPU 2872 +23 −23 40.3% +64.7 39.4% 642 91.0% 
12 Spike 1.2 Turin 2855 +10 −9 47.7% +13.5 41.3% 3833 100.0% 
13 Ktulu 8.0 2804 +12 −13 40.3% +63.9 37.4% 2250 53.0% 
14 Bright 0.2b 2803 +26 −26 55.1% −39.9 32.8% 494 69.0% 
15 Chess Tiger 2007 2795 +16 −16 42.5% +51.2 40.0% 1359 83.7% 
16 Scorpio 1.84 2777 +32 −32 45.4% +21.4 28.7% 355 57.9% 
17 Chessmaster 11 2773 +21 −21 43.2% +44.3 35.8% 743 88.7% 
18 Smarthink 1.00 32-bit 2757 +15 −15 43.5% +42.6 38.7% 1459 61.8% 
19 Alaric 707 2753 +22 −22 44.1% +41.7 35.7% 705 58.8% 
20 Movei 00.8.438 2749 +20 −20 53.1% −19.4 38.3% 772 69.4% 

[Tord Romstad]

But clearly added CPUs *are* a benefit as the CCRL 40/40 list shows:

Yes, clearly they are. I just had a look at how Glaurung compares to the best few engines with multi-CPU support which are included on both lists, and calculated how much Glaurung gains compared to the others when going from a single 32-bit CPU to four 64-bit CPUs. For instance, Glaurung is 235 points behind Rybka 2.3.2a on the first list, and only 184 points behind the same Rybka version on the second list, so Glaurung gains 235-184=51 rating points more than Rybka from the four 64-bit CPUs.

The results of my comparison:

Code: Select all

Compared to Rybka 2.3.2a:  +51
Compared to Shredder 11:   +68
Compared to Zappa Mexico:  -43
Compared to Hiarcs 11.1:   +39
Compared to Naum 2.2:      +18
Compared to Sjeng 2.7:     +99
Compared to Junior 10.1:   +50

As you can see, Glaurung gains more than all programs except Zappa, which is known for being particularly well optimized for multiple CPUs.

Tord

[Martin Thoresen]

Impressive Tord!

Please continue your work on Glaurung, it's what I live for when it comes to computer chess.

[Tony Thomas]

Tord, thanks for the comparison list. It looks like many of the commercial programs can gain a rapid 30-50 point increase in strength on a 4 CPU by simply stealing your SMP code. I hope that I am not giving anyone any ideas. Wow, I should just deposit my pay check and buy that core2quad from best buy instead of paying my bills. I sure wish you were competitive (you said a while back that you werent), if you were we would have had a new number one in few months.

[Erik Roggenburg]

Glaurung 2.0.1 is a 30+ Elo improvement over Glaurung 1.2 SMP from what I can tell. I haven't checked out playing style yet, however.

Code: Select all

    Program                          Elo    +   -   Games   Score   Av.Op.  Draws

  1 Rybka 2.3.1 mp 32-bit 2T       : 2824   13  13  2140    73.3 %   2649   29.3 %
  2 Rybka 2.2 mp 32-bit 2T         : 2816   15  15  1580    72.7 %   2646   30.4 %
  3 Rybka 2.3 LK mp 32-bit 2T      : 2811   18  18  1140    73.1 %   2637   28.8 %
  4 Rybka 2.3 mp 32-bit 2T         : 2810   18  18  1140    73.0 %   2637   30.4 %
  5 Deep Shredder 11 UCI 2T        : 2739   19  18   990    64.3 %   2636   30.3 %
  6 Zappa Mexico 2T                : 2732   18  18   960    63.2 %   2638   34.5 %
  7 Fritz 11                       : 2729   18  18   930    62.9 %   2638   35.5 %
  8 HIARCS 11.1 MP UCI 2T          : 2727   12  12  2210    60.2 %   2655   36.7 %
  9 Deep Fritz 10 2T               : 2723   12  12  2240    59.5 %   2657   31.2 %
 10 Loop 13.5.32 2T                : 2723   12  12  2010    61.0 %   2646   37.2 %
 11 Naum 2.2 2T                    : 2708   18  18   900    59.8 %   2639   40.6 %
 12 LoopMP 12.32 2T                : 2703   13  13  1680    56.9 %   2655   36.5 %
 13 HIARCS 11 MP UCI 2T            : 2696   14  14  1480    58.1 %   2639   37.2 %
 14 Deep Shredder 10 UCI 2T        : 2673   12  12  2190    52.0 %   2659   30.7 %
 15 Naum 2.1 MP 2T                 : 2665   12  12  1980    49.8 %   2666   39.1 %
 16 HIARCS 11 UCI                  : 2653   15  15  1380    52.7 %   2634   35.7 %
 17 Fruit 2.3.1                    : 2649   18  18   960    51.8 %   2637   35.3 %
 18 Toga II 1.2.1a                 : 2643   12  12  2160    47.4 %   2661   34.6 %
 19 Toga II 1.3x4                  : 2641   18  18   930    50.5 %   2638   36.5 %
 20 Deep Junior 10.1 2T            : 2638   12  12  2240    46.9 %   2659   30.6 %
 21 Spike 1.2 Turin                : 2635   12  12  2190    46.3 %   2660   36.9 %
 22 Hiarcs X54 UCI                 : 2629   16  16  1140    51.2 %   2621   35.7 %
 23 Hiarcs X50 UCI                 : 2627   16  16  1140    51.0 %   2621   36.7 %
 24 Deep Sjeng 2.7 2T              : 2624   18  18   990    47.2 %   2643   33.4 %
 25 Fritz 9                        : 2623   15  15  1410    48.2 %   2635   29.6 %
 26 Fruit 2.2.1                    : 2617   12  12  2160    43.7 %   2661   33.4 %
 27 Glaurung 2.0.1 2T              : 2617   27  27   420    46.1 %   2644   35.5 %
 28 Shredder 7.04 2T               : 2606   26  26   500    44.7 %   2643   29.4 %
 29 Glaurung 1.2 SMP 2T            : 2586   12  12  2220    39.3 %   2661   28.9 %
 30 Scorpio 2.0 2T                 : 2573   19  19   900    40.6 %   2639   31.7 %
 31 Chess Tiger 2007.1 UCI         : 2564   12  12  2270    36.6 %   2659   32.9 %
 32 Naum 2.0                       : 2548   16  16  1200    38.9 %   2627   36.2 %
 33 Scorpio 1.91 2T                : 2544   13  13  1920    32.1 %   2674   31.8 %
 34 Deep Pharaon 3.5.1 2T          : 2526   13  13  2250    31.3 %   2662   29.7 %
 35 Fritz 8 Bilbao                 : 2525   76  76    60    49.2 %   2531   28.3 %
 36 Chess Tiger 15.0               : 2525   17  17  1140    35.8 %   2626   34.5 %
 37 Scorpio 1.8 2T                 : 2503   16  16  1380    31.1 %   2641   30.0 %
 38 Deep Frenzee 3.0 2T            : 2500   13  13  2270    28.3 %   2661   24.8 %