Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / SF vs. NNUE at VLTC
- - By cma6 (****) Date 2020-08-16 17:05
Almost all the testing of SF vs. SF-NNUE is done by self-play at what correspondence players would call very short time controls. On Talk Chess, one has seen rapturous tales of how SF-NNUE, unlike SF, finds innovative lines of play that lead to small edges for one side or the other.

This supposed to be in contrast to SF, which tends to home in on lines of play resulting in 0.00 even early in the middlegame and staying that way for the remainder of the sterile game.  So far, I have not seen much, if any, difference in how the two engines operate at VLTC.
Parent - - By Peter Grayson (****) Date 2020-08-18 12:32 Upvotes 1

> So far, I have not seen much, if any, difference in how the two engines operate at VLTC


How do you define VLTC?

> resulting in 0.00 even early in the middlegame and staying that way for the remainder of the sterile game.


My last 100 game match at Game in 10m +3s/move using a short line player games based openings books vs an established deep engine games based book produced the majority of games conforming to the 0.00 comment.

Currently running a 600 game match including the 160620 hybrid engine, the 060820 final pure nnue engine both with the 2257 Sergio net, Houdini 6.03 and Komodo 11.3.1. Alothough now dated, the Komodo 11.3.1 engine according to the CCRL 40/15 list that is closest to the match game time controls indicates it is just 23 Elo below Komodo 14. I am using x50 book lines I composed from 260816 based on popular player lines at that time. There is no direct influence by engines in the opening lines selection.

The very first game highlighted how much today's Stockfish engines have moved on with the Hybrid engine here playing some fascinating chess. Key points were tying in Black's King to h8 and White's King manoevre to e6. Komodo sees it is lost about 30 moves too late. The nature of the games produced is reflective of the opening line opportunities offered to the engne and of course its opponent.

12 threads + 4 Gb hash per engine.

[Event "Stockfish 160820 vs 060820 vs Houdini 6"]
[Site "Newport, South Wales"]
[Date "2020.08.16"]
[Round "1.1"]
[White "Stockfish 160820-H2257"]
[Black "Komodo 11.3.1 64-bit BMI2"]
[Result "1-0"]
[ECO "B90"]
[Annotator "0.62;0.13"]
[PlyCount "179"]
[EventDate "2020.08.16"]
[EventType "tourn"]
[SourceTitle "Fritz Engine Tournament"]
[Source "Grayson"]

{Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz 3192 MHz W=54.0 plies; 12,683kN/s;
842,497 TBAs B=34.7 plies; 14,660kN/s; 405,888 TBAs} 1. e4 c5 2. Nf3 d6 3. d4
cxd4 4. Nxd4 Nf6 5. Nc3 a6 6. Be3 e5 7. Nb3 Be6 8. f3 Be7 9. Qd2 O-O 10. O-O-O
Nbd7 11. g4 {0.62/35 71} Qc7 {0.13/30 60 (b5)} 12. g5 {0.53/31 22 (Kb1)} Nh5 {
0.06/30 28} 13. f4 {0.73/32 25 (Kb1)} exf4 {0.09/28 26} 14. Bxf4 {0.62/28 11}
Nxf4 {0.24/28 32} 15. Qxf4 {0.81/29 10} Rac8 {0.20/30 48 (Ne5)} 16. h4 {
1.05/28 10 (Kb1)} b5 {0.13/26 17} 17. a3 {1.09/32 20 (Qf2)} Ne5 {0.22/27 29}
18. Kb1 {0.74/30 16} Rfe8 {0.36/28 64 (Bg4)} 19. Rd2 {0.98/31 42 (Rh2)} Qa7 {
0.17/27 35 (Bd8)} 20. Nd5 {1.72/24 5} Nc4 {0.15/26 11 (Qb7)} 21. Rf2 {1.11/33
47 (Rd4)} Bd8 {-0.10/28 11} 22. Bh3 {0.91/34 9 (Bxc4)} Bxd5 {-0.16/27 15} 23.
exd5 {1.25/32 8 (Bxc8)} Bb6 {0.00/27 18} 24. Rf3 {1.15/35 11} Rc7 {0.20/28 23}
25. h5 {0.70/38 26} Be3 {0.00/28 11 (Qa8)} 26. Qg3 {1.19/28 8} h6 {0.09/31 28
(Bd4)} 27. g6 {2.70/30 9} f6 {0.00/29 7} 28. Be6+ {2.73/29 7} Kh8 {0.30/31 13}
29. Re1 {2.49/34 11} Bd4 {0.26/30 15} 30. Nxd4 {2.49/33 7} Qxd4 {0.39/32 17}
31. Rc3 {2.49/35 9} Rb8 {0.42/29 21 (Qb6)} 32. Qg4 {2.49/36 9} Qf2 {0.42/33 14}
33. Qe2 {2.49/41 7} Qb6 {0.38/32 23} 34. Ka2 {2.49/46 8} Ra7 {0.33/31 18 (Re7)}
35. Rf3 {2.49/40 8 (Rf1)} Qd4 {0.35/33 17 (Qd8)} 36. c3 {3.43/36 20 (Rb3)} Qc5
{0.23/33 10 (Qb6)} 37. Qf2 {3.88/38 6 (Rf4)} Qxf2 {0.29/37 8} 38. Rxf2 {
3.88/49 7} Ne5 {0.49/37 9} 39. Re4 {3.88/51 13 (Rd1)} a5 {0.50/35 6 (Rc7)} 40.
Rd4 {3.88/53 8 (Rfe2)} Rc7 {0.50/37 20} 41. Kb1 {3.88/55 8 (Re2)} Re8 {0.50/35
5 (Nc4)} 42. Kc2 {3.88/49 7 (Re2)} a4 {0.50/37 6} 43. Re4 {4.92/48 89 (Re2)}
Rc5 {0.50/36 5} 44. Kd2 {4.92/46 6 (Rb4)} Nc4+ {0.50/39 4} 45. Ke1 {4.92/50 11
(Kc1)} Ne5 {0.50/40 5 (Rc7)} 46. Rd4 {4.92/90 7 (Rfe2)} Re7 {0.50/38 4} 47. Kd2
{4.92/51 5 (Kd1)} Nc4+ {0.50/37 4} 48. Ke1 {4.92/53 6 (Kd1)} Re8 {0.50/40 5}
49. Re2 {4.92/50 8 (Kd1)} Rc7 {0.50/35 7 (Ne5)} 50. Kf1 {4.92/51 9 (Kf2)} Rb8 {
0.50/39 6 (Ne5)} 51. Rf2 {4.92/78 6 (Kf2)} Ne5 {0.50/39 9 (Rcb7)} 52. Kg2 {
4.92/49 10} Nc4 {0.50/39 3 (Re8)} 53. Re2 {4.92/48 5} Rc5 {0.50/39 4 (Re7)} 54.
Kg3 {4.92/48 5 (Kf2)} Re8 {0.50/37 5 (Rc7)} 55. Kf4 {5.63/44 33 (Kf3)} Rb8 {
0.51/33 19 (Re7)} 56. Kf5 {5.51/32 6} Ne5 {0.86/32 7 (Re8)} 57. Bf7 {6.22/33 14
} Rcc8 {0.98/35 18} 58. Rb4 {6.40/32 3} Nd3 {1.41/29 3} 59. Rbe4 {6.67/33 11}
Ne5 {1.49/32 3} 60. Ke6 {6.68/31 3} Rb6 {1.44/30 2} 61. Rb4 {6.63/30 6} Rf8 {
1.76/28 4 (Rd8)} 62. Re3 {7.01/31 9 (Ree4)} Ra8 {1.43/27 2} 63. Kf5 {6.94/32 7}
Rbb8 {1.44/28 3} 64. c4 {7.60/29 4 (Kf4)} bxc4 {1.77/28 3} 65. Rxb8+ {7.88/29 5
} Rxb8 {-0.05/5 0} 66. Ke6 {8.14/28 4 (Re2)} Rb3 {2.67/30 6 (Rxb2)} 67. Rc3 {
8.58/28 3 (Re2)} Rxb2 {2.95/32 4} 68. Kxd6 {9.45/31 5} Rb7 {3.68/30 3 (Rh2)}
69. Be6 {10.33/29 4 (Rc2)} Rb3 {4.83/31 3 (Ra7)} 70. Rc1 {11.13/30 5 (Rc2)} Rb7
{6.01/30 3} 71. Rc2 {11.81/28 5} Ra7 {6.61/32 3} 72. Kc5 {12.66/29 6} Re7 {
7.32/29 3} 73. Rc3 {13.39/28 4} Re8 {12.21/29 4} 74. Rc2 {13.46/33 14 (Kb6)} f5
{7.43/33 3 (Nd3+)} 75. Kd6 {14.15/26 1} Ng4 {14.84/29 3} 76. Rxc4 {14.58/27 2}
Nf6 {250.00/25 1} 77. Bxf5 {15.05/27 3 (Kc6)} Kg8 {9.25/29 5} 78. Be6+ {
15.98/27 2 (Rxa4)} Kf8 {9.02/27 3} 79. Rc7 {16.49/26 2 (Rxa4)} Rd8+ {250.00/25
2 (Ra8)} 80. Ke5 {#11/78 2 (Kc6)} Nxh5 {#11/24 0} 81. Rf7+ {#10/83 2} Ke8 {
#9/25 0} 82. d6 {#9/93 2} Nf6 {#8/25 0} 83. Rxg7 {#8/110 2} Rxd6 {#7/27 0} 84.
Kxd6 {#7/149 2} Kd8 {#6/34 0} 85. Ra7 {#6/202 2 (Rf7)} Ne8+ {#5/56 0} 86. Ke5 {
#5/245 0} h5 {#4/99 0} 87. g7 {#4/245 0} Nxg7 {#3/99 0} 88. Rxg7 {#3/245 0} h4
{#2/99 0} 89. Kd6 {#2/245 0} Ke8 {#1/99 0 (h3)} 90. Rg8# {#1/245 0} 1-0

Parent - By cma6 (****) Date 2020-08-18 18:03 Upvotes 1
Peter, thanks for the feedback. For VLTC, I mean at least 5 minutes/move AND at least 40 ply--the minimum settings I use in IDeA. My observation was based on middlegame and endgame positions. By the end of the year, after I start 2 new tourneys in late 9/2020, I should have a readout on SF-NNUE in the late opening phase.
Up Topic The Rybka Lounge / Computer Chess / SF vs. NNUE at VLTC

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill