Glaurung 2.1 : 2582

[Le Fou numerique]

Hi,

Glaurung 2.1 first rating in the UEL: 2582.

PGN on the Division Elite page.

Regards,
Patrick

[Mike S.]

In a blitz match against Fruit 05/11/03, Glaurung 2.1 won 26-24.

Details: http://rybkaforum.net/cgi-bin/rybkaforu ... ?pid=57541
(see also posting above with more results)

[Martin Thoresen]

Patrick,

2582 seems very low?

[Tord Romstad]

Patrick,

2582 seems very low?

Hard to say. It isn't that far behind Glaurung 2.0.1, and if you check the UEL again now, Glaurung 2.1 has already climbed above 2600 again. The UEL results are still far worse than most other results I have seen, but perhaps this is caused by a very different pool of opponents. Perhaps the risky and speculative style of Glaurung 2.1 results in too many unnecessary draws and losses against weaker programs?

I could be wrong, but I think Patrick is also testing on a 32-bit CPU, which is a considerable disadvantage for Glaurung.

At any rate, this is a very interesting result. Thanks, Patrick!

Tord

[Le Fou numerique]

Patrick,

2582 seems very low?

Hard to say. It isn't that far behind Glaurung 2.0.1, and if you check the UEL again now, Glaurung 2.1 has already climbed above 2600 again. The UEL results are still far worse than most other results I have seen, but perhaps this is caused by a very different pool of opponents. Perhaps the risky and speculative style of Glaurung 2.1 results in too many unnecessary draws and losses against weaker programs?

I could be wrong, but I think Patrick is also testing on a 32-bit CPU, which is a considerable disadvantage for Glaurung.

At any rate, this is a very interesting result. Thanks, Patrick!

Tord

Hi Tord,

You are right.

The "first rating" is calculing about the results of the "tournoi d'entrée" (126 games) and the opponents have rating between 2693 and 2138. I choose an offset to not have a 3000 rating for no engines.
The most important is the difference between two ratings and not the number that would say nothing.

No Glaurung 2.1 has reach the same place in the list as the version 2.01.

Regards,
Patrick

[Eelco de Groot]

Maybe Glaurung 2.1 does not like the architecture of the Intel PIV very much. Glaurung scored 73.4% out of 126 games if I calculated somewhat correctly but the average opposition was not so high and there are some unexpected results against the engines in the lower half of the table. I think the search depths on more modern processors are way better than on a P IV.
For instance in the following game:


[Event "Tournoi d'entrée HS-Book.abk"]
[Site "Le Fou numerique"]
[Date "2008.05.17"]
[Round "1"]
[White "Glaurung 2.1"]
[Black "Cyrano 0.4"]
[Result "1/2-1/2"]
[ECO "A60"]
[Opening "Benoni"]
[Time "14:08:09"]
[Variation "4.Nf3"]
[TimeControl "300+0"]
[Termination "normal"]
[PlyCount "111"]
[WhiteType "program"]
[BlackType "program"]

1.d4 Nf6 2.c4 e6 3.Nf3 c5 4.d5 d6 5.Nc3 exd5 6.Nxd5 Nxd5 7.Qxd5 Nc6 8.Ng5 Qe7
9.Bd2 h6 10.Nf3 Be6 11.Qe4 O-O-O 12.e3 d5 13.cxd5 Bxd5 {+0.46/12 13} 14.Qg4+
{-0.25/11 9} Rd7 {+0.36/12 37} 15.Qa4 {-0.05/12 11} f5 {+0.48/11 26} 16.Bc3 {-0.17/12 9}
Qe4 {+0.56/10 4} 17.Qxe4 {+0.19/13 6} fxe4 {+0.30/11 3} 18.Nd2 {+0.19/12 6} Bd6
{+0.30/11 6} 19.Bc4 {+0.21/13 7} Bxc4 {+0.32/12 6} 20.Nxc4 {+0.29/13 4} Bc7 {+0.24/11 3}
21.Rd1 {+0.35/13 4} Rxd1+ {+0.20/10 3} 22.Kxd1 {+0.58/9} Rd8+ {+0.14/11 14} 23.Ke2
{+0.66/14 6} g6 {-0.16/11 5} 24.Rc1 {+0.70/14 15} Re8 {+0.26/10 5} 25.h4 {+0.68/13 10}
Re6 {+0.34/10 4} 26.Bg7 {+0.86/13 5} h5 {-0.06/11 10} 27.Nd2 {+1.19/13 3} Ne7
{+0.12/11 3} 28.Bf8 {+1.05/13 5} Nf5 {-0.08/11 4} 29.Bxc5 {+1.19/13 4} Nxh4 {-0.75/13 19}
30.g3 {+1.21/12 3} Nf5 {-0.88/13 9} 31.Rc4 {+1.09/12 4} a5 {-0.88/12 5} 32.Rxe4
{+1.41/13 3} Rc6 {-0.88/11 2} 33.Rc4 {+1.29/12 3} b6 {-1.03/12 10} 34.Bd4 {+1.31/13 4}
Nxd4+ {-1.00/11 2} 35.exd4 {+1.43/14 3} Rxc4 {-1.02/13 6} 36.Nxc4 {+1.62/10} g5
{-1.02/12 4} 37.Ne5 {+1.49/13 4} Bd8 {-0.94/12 3} 38.Kf3 {+1.15/12 3} Kc7 {-0.94/12 1}
39.b3 {+0.84/13 5} Be7 {-0.94/12 2} 40.Nc4 {+0.78/12 5} b5 {-0.90/11 2} 41.Ne3
{+0.92/13 3} Bd8 {-0.94/12 4} 42.Nf5 {+0.64/11 3} Kc6 {-0.88/13 1} 43.Ng7 {+0.60/13 5}
Kd5 {-0.74/14 3} 44.Nxh5 {+0.68/13 2} Kxd4 {-0.54/14 1} 45.Ng7 {+0.35/12 2} Kc3
{+0.16/14 1} 46.Ne6 {-0.01/13 3} Be7 {+0.20/14 1} 47.Nc7 {+0.13/12 2} Kb2 {0.00/15 2}
48.Nxb5 {-0.09/12 2} Bc5 {+0.22/14 2} 49.b4 {0.00/13 3} axb4 {0.00/15 2} 50.Nc7
{0.00/14 1} Kxa2 {0.00/17 2} 51.Na6 {0.00/16 2} Bd6 {0.00/17 2} 52.Nxb4+ {0.00/16 2}
Bxb4 {0.00/20 2} 53.Kg4 {0.00/19 2} Bd2 {0.00/20 1} 54.f4 {0.00/19 1} Bxf4 {0.00/23}
55.gxf4 {0.00/25 1} gxf4 {0.00/32 1} 56.Kxf4 {0.00/46 2} {Matériel insuffisant} 1/2-1/2

Comparing depths at move thirty, on my Athlon; 32 bit build

[d]2k5/ppb5/4r1p1/2B4p/4p2n/4P3/PP1NKPP1/2R5 w - -


2k5/ppb5/4r1p1/2B4p/4p2n/4P3/PP1NKPP1/2R5 w - -

Engine: Glaurung 2.1 (JA) (256 MB)
by Tord Romstad

2.00 0:00 +0.84 30.f4 exf3+ 31.gxf3 (208)

2.00 0:00 +1.62 30.g3 Nf5 31.Bxa7 (375)

3.00 0:00 +1.62 30.g3 Nf5 31.Bxa7 (624)

4.00 0:00 +0.92 30.g3 Ng2 31.g4 b6 (1.850)

4.00 0:00 +0.96 30.f3 exf3+ 31.gxf3 Ra6 32.Be7 (2.859)

5.00 0:00 +0.64 30.f3 exf3+ 31.gxf3 Ra6 32.Be7 Nf5 (4.229)

5.00 0:00 +1.25 30.g3 Nf5 31.g4 hxg4 32.Bxa7 (6.421)

6.00 0:00 +1.33 30.g3 Nf5 31.Rc4 b6 32.Ba3 b5
33.Rxe4 (12.934) 808

7.00 0:00 +1.33 30.g3 Nf5 31.Rc4 b6 32.Ba3 b5
33.Rxe4 (18.436) 594

8.00 0:00 +1.21 30.g3 Nf5 31.Rc4 b6 32.Bb4 a5 33.Ba3 b5
34.Rxe4 (29.027) 617

9.00 0:00 +1.05 30.g3 Nf5 31.Rc4 Ra6 32.Ba3 b5
33.Rb4 Bxg3 34.fxg3 Nxg3+ 35.Kf2 (64.812) 689

10.00 0:00 +1.05 30.g3 Nf5 31.Rc4 Ra6 32.Ba3 b5
33.Rb4 Bxg3 34.fxg3 Nxg3+ 35.Kf2 (116.462) 677

11.00 0:00 +1.31 30.g3 Nf5 31.Rc4 b6 32.Bf8 b5
33.Rxe4 Ra6 34.Rb4 Kd7 35.Rxb5 Rxa2 (248.208) 721

12.00 0:00 +1.33 30.g3 Nf5 31.Rc4 b6 32.Bf8 b5
33.Rxe4 Kd7 34.Rxe6 Kxe6 35.Ne4 Be5
36.Nc5+ Kf6 (406.221) 742

13.00 0:00 +1.27 30.g3 Nf5 31.Rc4 b6 32.Bf8 b5
33.Rxe4 Kd7 34.Rxe6 Kxe6 35.Ne4 Be5
36.Nc5+ Kf6 37.Nd3 (751.150) 762

14.01 0:02 +1.09 30.g3 Nf5 31.Rc4 Kd7 32.Ba3 Bd6
33.Nxe4 Bxa3 34.bxa3 Rc6 35.Rxc6 Kxc6
36.Kd3 Nd6 37.f3 Nxe4 38.Kxe4 (2.238.205) 753

15.01 0:05 +1.29 30.g3 Nf5 31.Rc4 Kd7 32.Rxe4 Ra6
33.g4 hxg4 34.Rxg4 Rxa2 35.Rxg6 Be5
36.b3 b6 37.f4 bxc5 38.fxe5 (4.350.840) 764

16.01 0:09 +1.01 30.g3 Nf5 31.Rc4 Kd7 32.Rxe4 Ra6
33.g4 hxg4 34.Rxg4 Rxa2 35.Rxg6 Be5
36.f4 Bxb2 37.e4 Bd4 38.Bxd4 Nxd4+
39.Kd3 (7.382.782) 768

17.01 0:26 +1.15 30.g3 Nf5 31.Rc4 Kd7 32.Rxe4 Ra6
33.g4 hxg4 34.Rxg4 Rxa2 35.Rxg6 Be5
36.Rg5 Ke6 37.e4 Nd4+ 38.Kd3 Ra4
39.Rg6+ Bf6 (20.123.188) 771

18.01 0:48 +0.94 30.g3 Nf5 31.Rc4 Kd7 32.Ba3 Bd6
33.Nxe4 Bxa3 34.bxa3 Rc6 35.Nc5+ Kc8
36.Rc3 b6 37.Ne4 Rxc3 38.Nxc3 Kd7
39.Ne4 Nd6 40.Kd3 Nxe4 41.Kxe4 (38.026.911) 779

18.06 1:36 +0.98 30.g4 hxg4 31.Nxe4 Rxe4 32.Bd6 Kd7
33.Bxc7 Rb4 34.b3 Nf3 35.Bb8 a6
36.Bf4 g5 37.Bc7 Ke6 38.Rc5 Re4
39.Kd3 (75.826.220) 784

best move: g2-g4 time: 2:16.844 min n/s: 789.439 nodes: 108.030.000

Three seconds here to reach depth twelve in the game is really not good,

Code: Select all

30.g3 {+1.21/12 3}

but Cyrano seems to be doing about the same depths, although there also seems to be considerable variance here. You would expect Glaurung to get better depths than Cyrano, although this is also dependant again on how fast a searcher Cyrano is. The variance in times per depth is really not such a good sign but I have not made any analysis of it, you would have to do a statistical analysis of it. A possible reason for the low depths is that Glaurung just does not like the PIV for games in five minutes, at least that would be my guess. Maybe a specific build for this processor would do better?

Regards, Eelco

[Ovyron]

I wanted to post this game from my private tourneys in where Glaurung defeated Rybka:

[Date "2008.05.18"]
[Round "1.10"]
[White "Rybka 2.3.2a 32-bit Slightly Optimistic"]
[Black "Glaurung 2.1 (JA)"]
[Result "0-1"]
[TimeControl "60+7"]

{Pondering=Off} 1. Nc3 {0.11/13 7} Nf6
{0.25/15 10} 2. Nf3 {0.11/12 4} d5 {0.17/14 8} 3. d4 {0.10/12 7} e6 {0.15/14 9} 4. g3 {0.16/12 6}
Bb4 {-0.31/13 8} 5. Bg2 {0.26/13 8} O-O {-0.09/13 10} 6. O-O {0.24/12 6} Nc6 {-0.11/14 11}
7. Qd3 {0.21/11 9} a5 {-0.25/13 10} 8. Nb5 {0.24/11 7} b6 {-0.21/13 8} 9. Be3 {0.21/11 7} Ba6
{-0.62/14 8} 10. c3 {0.08/10 12} Bd6 {-0.78/14 8} 11. b3 {0.06/9 2} e5 {-0.80/13 5} 12. dxe5
{-0.19/11 6} Nxe5 {-0.82/14 6} 13. Nxe5 {-0.24/12 3} Bxe5 {-0.70/14 8} 14. Rae1 {-0.26/12 6}
Bxc3 {-0.56/13 6} 15. Qxc3 {-0.26/12 2} Bxb5 {-0.54/15 6} 16. a4 {-0.25/12 5} Ba6 {-0.56/15 8}
17. Bd4 {-0.42/12 11} Re8 {-0.01/15 8} 18. Bxf6 {-0.36/13 7} Qxf6 {-0.25/16 5} 19. Qxf6
{-0.51/14 1} gxf6 {-0.41/18 7} 20. Bxd5 {-0.58/15 2} Rad8 {-0.37/17 6} 21. Bf3 {-0.47/15 4}
Rd2 {-0.33/16 6} 22. Rc1 {-0.30/14 6} c5 {-0.43/17 8} 23. Rfd1 {-0.32/15 2} Rxd1+ {-0.31/17 8}
24. Rxd1 {-0.29/14 0} Bxe2 {-0.31/18 5} 25. Re1 {-0.39/16 2} Bb5 {-0.27/18 8} 26. Rd1
{-0.49/16 6} Ba6 {-0.23/17 5} 27. Kg2 {-0.40/15 16} Kf8 {-0.33/16 6} 28. Rd7 {-0.32/15 20} Re7
{-0.17/16 8} 29. Rd8+ {-0.27/14 2} Kg7 {0.00/17 8} 30. Rd6 {-0.39/14 2} Re1 {-0.23/17 7}
31. Kh3 {-0.57/14 7} Re6 {-0.58/15 5} 32. Rd8 {-0.56/13 2} f5 {-0.66/15 5} 33. Rd7
{-0.63/15 9} f4 {-0.66/15 7} 34. Kg2 {-0.60/15 9} fxg3 {-0.72/16 8} 35. fxg3 {-0.62/15 19} Be2
{-0.80/15 6} 36. Bd5 {-0.71/16 16} Rf6 {-0.98/16 8} 37. Bc4 {-0.75/16 9} Bf3+ {-1.00/17 6}
38. Kg1 {-0.67/15 0} Bg4 {-0.98/18 7} 39. Rd8 {-0.75/16 11} Bh3 {-0.90/17 7} 40. Rd2
{-0.74/17 8} Kg6 {-1.00/16 5} 41. Rd1 {-0.74/16 5} Be6 {-1.01/16 7} 42. Kg2 {-0.87/14 9} Kf5
{-1.07/17 7} 43. Bxe6+ {-0.85/15 8} Kxe6 {-1.35/16 6} 44. Rd8 {-0.85/16 2} Ke5 {-1.39/17 5}
45. Rd7 {-0.90/14 4} h6 {-1.49/18 7} 46. h3 {-1.14/16 14} h5 {-1.60/18 5} 47. h4 {-1.07/17 6}
Ke4 {-1.80/19 8} 48. Rd8 {-1.05/16 11} Ke3 {-1.88/18 5} 49. Re8+ {-1.16/15 9} Kd2 {-1.94/18 8}
50. Re7 {-1.85/16 8} Kc2 {-2.27/19 6} 51. Re3 {-1.56/16 1} Rd6 {-2.47/18 5} 52. Kf3
{-1.32/17 5} Rd4 {-2.27/18 8} 53. Ke2 {-1.32/16 1} f5 {-2.43/18 8} 54. Rf3 {-1.56/16 6} Re4+
{-3.23/17 8} 55. Kf2 {-1.30/15 14} Rb4 {-3.49/18 7} 56. Ke2 {-2.66/15 7} Rxb3 {-4.17/18 7}
57. Rf4 {-2.69/15 1} Rb4 {-4.31/17 7} 58. Rxf5 {-3.06/14 3} c4 {-5.05/15 7} 59. Rf6
{-4.31/13 3} c3 {-6.78/16 6} 60. Rc6 {-5.54/13 12} Re4+ {-8.29/15 7} 61. Kf3 {-5.53/13 14}
Rxa4 {-9.82/16 5} 62. Ke3 {-6.91/13 12} b5 {-11.41/14 7} 63. Ke2 {-7.96/11 10} b4
{-14.25/16 7} 64. Rc4 {-8.79/12 9} Ra1 {-17.01/13 7} 65. Rc7 {-10.43/10 6} b3 {-25.00/13 6}
66. Rb7 {-10.43/8 9} b2 {-82.77/14 6} 67. Ke3 {-17.33/8 8} b1Q {-95.47/15 6} 68. Rxb1
{-17.77/9 4} Kxb1 {-95.47/14 7} 69. Kf4 {-17.79/8 3} c2 {-#10/13 1} 70. Kg5 {-18.07/8 2} c1Q+
{-#8/12 1} 71. Kf5 {-25.62/8 11} Qc6 {-#6/9 0} 72. Ke5 {-#7/7 4} Ra4 {-#5/7 0} 73. Kf5 {-#5/3 0}
Re4 {-#4/6 0} 74. g4 {-#3/3 0} Qe6+ {-#3/6 0} 75. Kg5 {-#2/3 0} Rxg4+ {-#2/6 0} 76. Kxh5
{-#1/3 0} Qg6# {-#1/6 0} {Checkmate !!} 0-1

[Eastendboy]

For what it's worth, I run Glaurung 2.1 on two different P4 systems, one of which is newer than the other. The newer P4 is hyper-threaded and Glaurung 2.1 appears to run much better on this system than it does on the older P4 even though the clock speeds are not that far apart. Note that I make that statement based on empirical data and have nothing concrete with which I can back it up.

I'm not familiar with the in's and out's of the P4 architecture but there does seem to be some value in compiling versions that are specific to both P4 system types. I say this because there are several different Toga 1.4.1SE versions compiled specifically for use with P4's and I noticed substantial performance increases on both P4 systems when I started using a version compiled specifically for my hardware. Something to do with the SSE implementation perhaps?