Lc0 question

[Rebel]

For my gambit tournament (see elsewhere) I want to include Lc0-v27. Hardware RTX 1060 6Gb.

Time control one : 40moves in 2 minutes --> which net?
Time control two : 40moves in 15 minutes --> which net?

[mclane]

On the fast time control i would use the net that comes with lc0. 703810.
With longer time controls j94-130 could maybe work.
How much ram has your 1060 ?! 3GB or 6 GB ?!

[Thomas Lagershausen]

In a gambit tournament i would prefer Leelenstein 15.0 at both timecontrols.

It is now free.

[Werewolf]

On the fast time control i would use the net that comes with lc0. 703810.
With longer time controls j94-130 could maybe work.
How much ram has your 1060 ?! 3GB or 6 GB ?!

He clearly states 6GB

[Modern Times]

He clearly states 6GB

Hmm... he also says RTX 1060. As far as I know that doesn't exist or am I wrong ? Either GTX 1060 or RTX 2060.

[Rebel]

It's indeed the RTX 1060 6Gb.

I am having bad results for v27.0. At first I thought it must be the net, but no, after inspection of the PGN it is the second time control (move 41-80), something is very wrong. An example playing 40m/120s (3 secs average) -

All goes well till move 60, then it starts moving faster and faster with a devastating result. Then at move 81 it suddenly uses 12 seconds, damage is already done. This is a repeating pattern in the other games.

[Black "Lc0-v27"]
41. Rb7 {+1.12/33 3.2s} Kg5 {-0.92/13 6.9s} 42. Rf7 {+1.22/30 1.8s}
Rd6 {-0.91/12 4.9s} 43. Rg7+ {+1.29/31 2.8s} Kf6 {-0.90/12 7.3s}
44. Rg8 {+1.38/31 2.2s} Kf7 {-0.89/12 13s} 45. Rb8 {+0.92/33 5.1s}
Kf6 {-0.85/10 11s} 46. Re8 {+1.00/29 1.9s} Rd4 {-0.91/11 8.9s}
47. Rf8+ {+0.86/35 5.6s} Kg6 {-0.88/11 7.9s} 48. Re8 {+0.82/34 3.1s}
Kf6 {-0.85/12 2.1s} 49. Rb8 {+0.74/32 2.0s} Rd6 {-0.78/11 4.6s}
50. Rb7 {+0.74/36 2.1s} Kg6 {-0.76/12 7.1s} 51. Re7 {+0.73/34 2.2s}
Kf6 {-0.74/12 2.9s} 52. Rh7 {+0.69/37 3.2s} Kg6 {-0.76/12 9.2s}
53. Rh8 {+0.72/40 16s} Kg5 {-0.74/10 4.3s} 54. Re8 {+0.66/35 1.2s}
Kf6 {-0.75/12 2.7s} 55. Rb8 {+0.66/38 5.5s} Kg6 {-0.75/10 6.5s}
56. Re8 {+0.65/31 1.7s} Kf6 {-0.80/13 0.86s} 57. Rf8+ {+0.59/33 1.9s}
Kg5 {-0.66/10 3.6s} 58. Rg8+ {+0.52/32 2.4s} Kf6 {-0.65/10 1.8s}
59. Rb8 {+0.71/34 3.0s} Kg5 {-0.68/10 4.0s} 60. Rf8 {+0.66/33 2.0s}
Rd4 {-0.67/10 0.93s} 61. Rg8+ {+0.51/36 6.2s} Kf6 {-0.65/10 1.1s}
62. Rh8 {+0.51/36 1.2s} Kg5 {-0.62/10 1.4s} 63. Kc3 {+0.51/37 5.4s}
Rd6 {-0.64/10 1.8s} 64. Rg8+ {+0.51/38 3.0s} Kf6 {-0.63/11 0.74s}
65. Kc2 {+0.51/37 1.2s} Kf7 {-0.60/11 1.5s} 66. Rh8 {+0.54/37 1.8s}
Kg6 {-0.63/11 1.3s} 67. Rf8 {+0.52/37 1.8s} Kg5 {-0.59/9 0.71s}
68. Rc8 {+0.62/38 2.2s} Kf6 {-0.58/9 0.89s} 69. f4 {+0.89/37 1.9s}
Kf7 {-0.64/11 0.81s} 70. Rc7+ {+0.45/38 6.1s} Kg6 {-0.65/10 0.46s}
71. Re7 {+0.62/39 5.5s} Kh6 {-0.67/10 0.28s} 72. Rf7 {+0.55/35 1.2s}
Kg6 {-0.64/10 0.22s} 73. Rf8 {+0.43/37 1.8s} Rd7 {-0.61/9 0.61s}
74. Rb8 {+0.42/36 1.7s} Rd6 {-0.60/10 0.21s} 75. Re8 {+0.47/34 3.9s}
Kg7 {-0.73/10 0.39s} 76. Re5 {+1.19/29 1.5s} Kg6 {-0.78/11 0.23s}
77. Rd5 {+0.95/35 2.1s} Rf6 {-0.77/10 0.23s} 78. Rd8 {+0.96/35 1.9s}
Re6 {-0.79/9 0.25s} 79. Kd2 {+2.87/26 0.90s} Kg7 {-1.27/10 0.53s}
80. Rd5 {+3.72/25 1.4s} Kf6 {-1.97/9 0.23s} 81. Kd3 {+5.74/27 2.9s}
Re1 {-3.58/12 12s}

[Rebel]

[pgn][Event "CCRL 40/15"]
[Site "CCRL"]
[Date "2021.03.14"]
[Round "763.2.875"]
[White "Lc0 0.27.0 w703810 64-bit 4CPU"]
[Black "Wasp 4.50 64-bit 4CPU"]
[Result "1/2-1/2"]
[ECO "D04"]
[Opening "Queen's pawn game"]
[PlyCount "171"]
[WhiteElo "3334"]
[BlackElo "3234"]

1. d4 {book} d5 {book} 2. Nf3 {book} Nf6 {book} 3. e3 {book} c5 {book} 4. b3
{book} Bg4 {book} 5. Be2 {book} e6 {book} 6. O-O {+0.01/10 59s} Nc6 {+0.00/24
22s} 7. Bb2 {+0.00/11 19s} cxd4 {-0.01/24 22s} 8. Nxd4 {+0.03/11 14s} Bxe2
{-0.05/26 16s} 9. Qxe2 {+0.04/10 4.9s} Nxd4 {-0.01/25 16s} 10. Bxd4 {+0.04/9
17s} Rc8 {-0.14/24 16s} 11. Rd1 {+0.07/10 33s} Be7 {-0.16/24 22s} 12. c4
{+0.06/10 3.1s} O-O {-0.25/25 17s} 13. Nd2 {+0.06/9 7.9s} Qc7 {-0.16/24 17s}
14. Rac1 {+0.06/8 28s} b6 {-0.17/25 23s} 15. Nf3 {+0.06/9 33s} Qb7 {-0.09/24
23s} 16. Bxf6 {+0.13/11 12s} Bxf6 {-0.14/28 17s} 17. cxd5 {+0.14/11 2.2s} Rxc1
{+0.00/29 17s} 18. Rxc1 {+0.13/10 3.2s} Qxd5 {-0.04/29 17s} 19. Qc4 {+0.15/8
5.9s} Qb7 {-0.13/26 23s} 20. Qc7 {+0.18/9 20s} Qa6 {-0.33/28 23s} 21. Rc2
{+0.18/11 11s} Qd3 {-0.21/26 23s} 22. g3 {+0.21/10 37s} g5 {+0.00/27 17s} 23.
h3 {+0.21/11 4.1s} h5 {-0.03/27 18s} 24. Qc4 {+0.21/11 8.4s} Qd7 {-0.16/26 24s}
25. Kg2 {+0.29/10 24s} Rd8 {-0.25/28 24s} 26. g4 {+0.30/9 11s} h4 {-0.16/28
18s} 27. Qc7 {+0.29/9 17s} Kg7 {-0.10/27 18s} 28. Rc4 {+0.28/9 37s} Kg6
{+0.00/31 19s} 29. Rc2 {+0.26/10 33s} Qxc7 {+0.00/34 19s} 30. Rxc7 {+0.22/11
13s} a5 {+0.00/35 19s} 31. Rc6 {+0.17/11 126s} b5 {+0.00/35 20s} 32. Rc5
{+0.16/12 1.3s} Rd5 {+0.00/35 20s} 33. Rxd5 {+0.14/10 11s} exd5 {+0.00/35 21s}
34. Kf1 {+0.14/11 1.1s} Bc3 {+0.14/32 29s} 35. Ke2 {+0.10/12 38s} f5 {+0.07/33
21s} 36. Kd3 {+0.10/12 0.20s} fxg4 {+0.00/33 33s} 37. hxg4 {+0.08/12 13s} Bf6
{+0.00/34 22s} 38. Ng1 {+0.11/10 16s} Be5 {+0.00/36 23s} 39. Nh3 {+0.13/9 19s}
a4 {+0.00/36 25s} 40. f4 {+0.10/9 19s} Bf6 {+0.00/37 28s} 41. Ke2 {+0.07/10
116s} a3 {+0.00/34 15s} 42. Kd3 {+0.14/10 56s} Kg7 {+0.00/35 15s} 43. e4
{+0.13/11 53s} dxe4+ {+0.00/37 15s} 44. Kxe4 {+0.12/12 12s} Kg6 {+0.00/38 15s}
45. f5+ {+0.12/11 8.6s} Kf7 {+0.00/39 16s} 46. Kd5 {+0.09/11 140s} b4 {+0.00/40
16s} 47. Kc4 {+0.08/10 65s} Be7 {+0.00/39 16s} 48. Kd4 {+0.07/10 45s} Kf6
{+0.00/40 16s} 49. Nf2 {+0.07/10 46s} Kf7 {+0.00/40 16s} 50. Ke5 {+0.07/10 31s}
Bf6+ {+0.00/38 16s} 51. Ke4 {+0.07/10 56s} Bc3 {+0.00/41 16s} 52. Kd5 {+0.10/9
26s} Bf6 {+0.00/41 16s} 53. Nh3 {+0.07/9 35s} Be7 {+0.00/41 17s} 54. Ke5
{+0.06/9 13s} Bf6+ {+0.00/42 17s} 55. Ke4 {+0.06/9 15s} Ke7 {+0.00/43 17s} 56.
Kd5 {+0.08/9 7.1s} Kf7 {+0.00/43 17s} 57. Kc5 {+0.05/9 33s} Be7+ {+0.00/40 17s}
58. Kc4 {+0.05/9 12s} Kf6 {+0.00/42 17s} 59. Kd4 {+0.04/9 11s} Bd6 {+0.00/39
18s} 60. Ke4 {+0.05/8 14s} Bg3 {+0.00/40 18s} 61. Kd5 {+0.08/8 7.2s} Bf4
{+0.00/40 18s} 62. Kd4 {+0.06/7 8.0s} Bd6 {+0.00/40 18s} 63. Ke4 {+0.05/7 4.8s}
Bg3 {+0.00/41 18s} 64. Kd5 {+0.07/6 3.7s} Bf4 {+0.00/41 19s} 65. Nf2 {+0.03/6
12s} Kf7 {+0.00/42 19s} 66. Kd4 {+0.06/7 6.4s} Bd2 {+0.00/38 19s} 67. Nh3
{+0.07/7 3.8s} Bf4 {+0.00/39 20s} 68. Ke4 {+0.08/6 2.8s} Bd2 {+0.00/41 20s} 69.
Kd3 {+0.07/6 2.0s} Bf4 {+0.00/41 20s} 70. Kc4 {+0.07/6 1.6s} Bd2 {+0.00/39 21s}
71. Nf2 {+0.06/6 1.4s} Be3 {+0.00/39 21s} 72. Nh3 {+0.07/7 1.0s} Bd2 {+0.00/40
22s} 73. Kd3 {+0.06/6 1.2s} Bf4 {+0.00/40 22s} 74. Ke4 {+0.05/5 1.3s} Bc1
{+0.00/37 23s} 75. Nf2 {+0.04/6 4.1s} Bd2 {+0.00/40 24s} 76. Kd3 {+0.08/5
0.89s} Bf4 {+0.00/40 25s} 77. Nh3 {+0.04/5 0.83s} Kf6 {+0.00/40 26s} 78. Kc4
{+0.05/5 0.94s} Bd6 {+0.00/39 28s} 79. Kd4 {+0.04/5 0.83s} Be7 {+0.00/39 30s}
80. Nf2 {+0.05/5 0.72s} Kf7 {+0.00/43 34s} 81. Ke5 {+0.02/8 89s} Bf6+ {+0.00/42
15s} 82. Ke4 {+0.02/9 59s} Bc3 {+0.00/41 15s} 83. Nh3 {+0.03/9 98s} Bf6
{+0.00/43 15s} 84. Kd5 {+0.02/8 36s} Be7 {+0.00/41 16s} 85. Ke5 {+0.01/8 38s}
Bd8 {+0.00/40 16s} 86. Kd6 {+0.01/7 39s, Draw by adjudication} 1/2-1/2
[/pgn]

CCRL games show the same behavior.

[Guenther]

Code: Select all

[Event "CCRL 40/15"]
[Site "CCRL"]
[Date "2021.03.14"]
[Round "763.2.875"]
[White "Lc0 0.27.0 w703810 64-bit 4CPU"]
[Black "Wasp 4.50 64-bit 4CPU"]

Code: Select all

41. Ke2 {+0.07/10 116} a3 {+0.00/34 15}
42. Kd3 {+0.14/10 56} Kg7 {+0.00/35 15}
43. e4 {+0.13/11 53} dxe4+ {+0.00/37 15}
44. Kxe4 {+0.12/12 12} Kg6 {+0.00/38 15}
45. f5+ {+0.12/11 8.6} Kf7 {+0.00/39 16}
46. Kd5 {+0.09/11 140} b4 {+0.00/40 16}
47. Kc4 {+0.08/10 65} Be7 {+0.00/39 16}
48. Kd4 {+0.07/10 45} Kf6 {+0.00/40 16}
49. Nf2 {+0.07/10 46} Kf7 {+0.00/40 16}
50. Ke5 {+0.07/10 31} Bf6+ {+0.00/38 16}
51. Ke4 {+0.07/10 56} Bc3 {+0.00/41 16}
52. Kd5 {+0.10/9 26} Bf6 {+0.00/41 16}
53. Nh3 {+0.07/9 35} Be7 {+0.00/41 17}
54. Ke5 {+0.06/9 13} Bf6+ {+0.00/42 17}
55. Ke4 {+0.06/9 15} Ke7 {+0.00/43 17}
56. Kd5 {+0.08/9 7.1} Kf7 {+0.00/43 17}
57. Kc5 {+0.05/9 33} Be7+ {+0.00/40 17}
58. Kc4 {+0.05/9 12} Kf6 {+0.00/42 17}
59. Kd4 {+0.04/9 11} Bd6 {+0.00/39 18}
60. Ke4 {+0.05/8 14} Bg3 {+0.00/40 18}
61. Kd5 {+0.08/8 7.2} Bf4 {+0.00/40 18}
62. Kd4 {+0.06/7 8.0} Bd6 {+0.00/40 18}
63. Ke4 {+0.05/7 4.8} Bg3 {+0.00/41 18}
64. Kd5 {+0.07/6 3.7} Bf4 {+0.00/41 19}
65. Nf2 {+0.03/6 12} Kf7 {+0.00/42 19}
66. Kd4 {+0.06/7 6.4} Bd2 {+0.00/38 19}
67. Nh3 {+0.07/7 3.8} Bf4 {+0.00/39 20}
68. Ke4 {+0.08/6 2.8} Bd2 {+0.00/41 20}
69. Kd3 {+0.07/6 2.0} Bf4 {+0.00/41 20}
70. Kc4 {+0.07/6 1.6} Bd2 {+0.00/39 21}
71. Nf2 {+0.06/6 1.4} Be3 {+0.00/39 21}
72. Nh3 {+0.07/7 1.0} Bd2 {+0.00/40 22}
73. Kd3 {+0.06/6 1.2} Bf4 {+0.00/40 22}
74. Ke4 {+0.05/5 1.3} Bc1 {+0.00/37 23}
75. Nf2 {+0.04/6 4.1} Bd2 {+0.00/40 24}
76. Kd3 {+0.08/5 0.89} Bf4 {+0.00/40 25}
77. Nh3 {+0.04/5 0.83} Kf6 {+0.00/40 26}
78. Kc4 {+0.05/5 0.94} Bd6 {+0.00/39 28}
79. Kd4 {+0.04/5 0.83} Be7 {+0.00/39 30}
80. Nf2 {+0.05/5 0.72} Kf7 {+0.00/43 34}
81. Ke5 {+0.02/8 89} Bf6+ {+0.00/42 15}
82. Ke4 {+0.02/9 59} Bc3 {+0.00/41 15}
83. Nh3 {+0.03/9 98} Bf6 {+0.00/43 15}
84. Kd5 {+0.02/8 36} Be7 {+0.00/41 16}
85. Ke5 {+0.01/8 38} Bd8 {+0.00/40 16}
86. Kd6 {+0.01/7 39}
{Draw by adjudication}
1/2-1/2

CCRL games show the same behavior.

This looks like a serious regression. Did you check if they changed the time manager code at some time?
(I haven't tested LC0 for very long time, but I am sure it did not behave like that in the past.)
This should cost quite some rating points. It looks like weighting of average game length goes bonker.
OTH I am quite sure they nearly never test mps tc.

[Rebel]

Like you I hardly look at Lc0 and yes this is an expensive bug.

[Guenther]

Like you I hardly look at Lc0 and yes this is an expensive bug.

It seems this is happening since quite long!, OTH LC0 is very neglected in the 40/15 rating list anyways,
only cpu versions are tested there since ever. Not a single test for a version with a GPU and most of them with
not much games, so it seems no one cared for from the LC0 team, or it went unnoticed.

(In a short check for a few versions at CCRL I immediately noticed several losses due to that crappy time management)

http://ccrl.chessdom.com/ccrl/4040/cgi/ ... +opponents