Stockfish vs Stockfish-LazySMP 20 core 120 min + 30 sec

[zullil]

Code: Select all

Started game 1 of 2 (SF-master vs SF-lazySMP)
Finished game 1 (SF-master vs SF-lazySMP): 1/2-1/2 {Draw by 3-fold repetition}
Score of SF-master vs SF-lazySMP: 0 - 0 - 1  [0.500] 1
Started game 2 of 2 (SF-lazySMP vs SF-master)
Finished game 2 (SF-lazySMP vs SF-master): 1/2-1/2 {Draw by 3-fold repetition}
Score of SF-master vs SF-lazySMP: 0 - 0 - 2  [0.500] 2

[pgn]
[Event "Engine_Match"]
[Site "LZsT7910"]
[Date "2015.10.07"]
[Round "13"]
[White "SF-master"]
[Black "SF-lazySMP"]
[Result "1/2-1/2"]
[FEN "rnbqkb1r/pppp1ppp/7n/4p3/1P6/3P4/P1P1PPPP/RNBQKBNR w KQkq -"]
[PlyCount "101"]
[SetUp "1"]
[TimeControl "7200+30"]

1. c3 {-0.03/32 159s} d5 {+0.02/28 185s} 2. Nf3 {-0.03/29 15s}
Bd6 {0.00/30 359s} 3. e4 {0.00/35 436s} c6 {+0.01/32 175s} 4. Bg5 {0.00/35 112s}
Qc7 {-0.07/33 952s} 5. Be2 {0.00/34 170s} f5 {0.00/34 100s}
6. Bxh6 {+0.06/35 298s} gxh6 {0.00/36 101s} 7. Nh4 {+0.04/36 167s}
O-O {0.00/36 101s} 8. Nxf5 {+0.01/34 19s} Bxf5 {0.00/37 102s}
9. exf5 {0.00/38 297s} Rxf5 {0.00/37 106s} 10. Nd2 {0.00/36 32s}
Na6 {0.00/37 103s} 11. Bg4 {0.00/38 183s} Rg5 {0.00/36 244s}
12. O-O {0.00/36 37s} Qf7 {0.00/34 102s} 13. Qf3 {0.00/38 113s}
Qg7 {0.00/33 215s} 14. Be6+ {0.00/38 114s} Kh8 {0.00/36 102s}
15. Qh3 {-0.14/36 229s} Qf6 {0.00/37 132s} 16. Bc8 {0.00/39 178s}
Rg7 {0.00/38 103s} 17. Nb3 {0.00/41 125s} Rc7 {+0.02/33 262s}
18. Bg4 {-0.07/36 162s} Rg8 {+0.11/35 163s} 19. f3 {-0.09/38 114s}
Rcg7 {+0.17/36 582s} 20. Kh1 {-0.12/38 198s} Qg6 {+0.08/34 139s}
21. Rae1 {-0.12/34 16s} Qxd3 {+0.10/29 29s} 22. Qxh6 {-0.13/38 164s}
Rg6 {+0.07/34 95s} 23. Qe3 {-0.09/35 39s} Qxe3 {+0.05/35 96s}
24. Rxe3 {-0.12/38 119s} Rf8 {+0.03/34 514s} 25. Rc1 {-0.10/35 178s}
Nc7 {+0.06/33 169s} 26. g3 {-0.05/39 181s} e4 {+0.06/35 131s}
27. Nd4 {-0.05/37 18s} Be5 {+0.01/29 17s} 28. a4 {-0.05/40 121s}
Na8 {+0.01/35 98s} 29. a5 {-0.05/42 121s} Nc7 {+0.01/38 89s}
30. Kg2 {-0.01/38 121s} Rg7 {+0.01/37 89s} 31. Rd1 {0.00/42 120s}
Bxd4 {+0.01/39 154s} 32. cxd4 {0.00/44 120s} Nb5 {0.00/34 13s}
33. h3 {0.00/45 119s} exf3+ {0.00/41 100s} 34. Bxf3 {0.00/42 118s}
Nd6 {0.00/42 88s} 35. Re6 {0.00/45 158s} Rg6 {0.00/41 87s} 36. Re5 {0.00/43 25s}
Rgf6 {0.00/43 86s} 37. Rd3 {0.00/47 116s} Nb5 {0.00/44 85s}
38. h4 {0.00/48 115s} Kg8 {0.00/43 85s} 39. Rg5+ {0.00/46 149s}
Kh8 {0.00/56 86s} 40. Re5 {0.00/67 151s} Rg6 {0.00/41 82s}
41. Bh5 {0.00/42 110s} Rgf6 {0.00/39 84s} 42. Rd2 {0.00/44 194s}
a6 {0.00/40 80s} 43. Re7 {0.00/43 26s} Rf5 {0.00/40 79s} 44. g4 {0.00/44 103s}
Rf3 {0.00/40 78s} 45. Rxb7 {0.00/44 105s} Nxd4 {0.00/40 77s}
46. Rxd4 {0.00/45 180s} Rf2+ {0.00/49 75s} 47. Kg1 {0.00/57 23s}
Rf1+ {0.00/55 209s} 48. Kh2 {0.00/68 96s} R1f2+ {0.00/55 69s}
49. Kg1 {0.00/95 200s} Rf1+ {0.00/83 104s} 50. Kh2 {0.00/126 164s}
R1f2+ {0.00/63 214s} 51. Kg1 {0.00/104 138s, Draw by 3-fold repetition} 1/2-1/2

[Event "Engine_Match"]
[Site "LZsT7910"]
[Date "2015.10.07"]
[Round "14"]
[White "SF-lazySMP"]
[Black "SF-master"]
[Result "1/2-1/2"]
[FEN "rnbqkb1r/pppp1ppp/7n/4p3/1P6/3P4/P1P1PPPP/RNBQKBNR w KQkq -"]
[PlyCount "72"]
[SetUp "1"]
[TimeControl "7200+30"]

1. Nf3 {0.00/31 246s} Bxb4+ {+0.02/34 174s} 2. c3 {0.00/32 159s}
Bd6 {+0.09/35 167s} 3. Bg5 {-0.12/32 317s} f6 {+0.02/31 26s}
4. Bxh6 {-0.07/32 267s} gxh6 {0.00/36 163s} 5. e4 {-0.12/31 280s}
O-O {0.00/35 172s} 6. d4 {-0.11/32 585s} exd4 {+0.14/36 161s}
7. Bc4+ {-0.12/32 326s} Kh8 {+0.11/32 19s} 8. cxd4 {-0.01/29 23s}
Bb4+ {+0.08/36 369s} 9. Nbd2 {-0.09/33 414s} c6 {+0.16/36 154s}
10. d5 {0.00/33 188s} b5 {0.00/33 62s} 11. Be2 {0.00/33 389s} a5 {+0.06/37 306s}
12. O-O {0.00/34 91s} Na6 {0.00/37 220s} 13. a3 {0.00/33 91s} Bc3 {0.00/35 19s}
14. Ra2 {0.00/36 202s} Bb7 {+0.01/38 120s} 15. dxc6 {-0.05/31 28s}
Bxc6 {0.00/40 130s} 16. Nb1 {0.00/37 225s} Be5 {0.00/41 117s}
17. Nxe5 {0.00/39 275s} fxe5 {0.00/42 118s} 18. Rb2 {0.00/36 13s}
b4 {0.00/42 119s} 19. Qd6 {0.00/38 91s} Qf6 {0.00/42 162s}
20. Qxf6+ {0.00/42 92s} Rxf6 {0.00/42 16s} 21. f3 {0.00/38 93s}
h5 {0.00/43 122s} 22. Bxa6 {0.00/42 196s} Rxa6 {0.00/47 123s}
23. axb4 {0.00/43 92s} Rb6 {0.00/47 124s} 24. Nd2 {0.00/45 96s}
Rxb4 {0.00/48 125s} 25. Rxb4 {0.00/47 226s} axb4 {0.00/50 125s}
26. Rb1 {0.00/49 143s} Rd6 {0.00/50 126s} 27. Nc4 {0.00/51 91s}
Rd4 {0.00/50 126s} 28. Nxe5 {0.00/50 91s} Bb5 {0.00/50 126s}
29. Kf2 {0.00/48 109s} d6 {0.00/50 126s} 30. Nf7+ {0.00/50 91s}
Kg7 {0.00/51 126s} 31. Ng5 {0.00/52 91s} Kf6 {0.00/50 126s}
32. Nxh7+ {0.00/51 746s} Kg6 {0.00/51 298s} 33. Nf8+ {0.00/53 75s}
Kf7 {0.00/50 18s} 34. Nh7 {0.00/54 105s} Kg6 {0.00/57 177s}
35. Nf8+ {0.00/61 100s} Kf7 {0.00/70 180s} 36. Nh7 {0.00/60 230s}
Kg6 {0.00/59 168s, Draw by 3-fold repetition} 1/2-1/2
[/pgn]

[zullil]

Code: Select all

Started game 1 of 2 (SF-master vs SF-lazySMP)
Finished game 1 (SF-master vs SF-lazySMP): 1/2-1/2 {Draw by 3-fold repetition}
Score of SF-master vs SF-lazySMP: 0 - 0 - 1  [0.500] 1
Started game 2 of 2 (SF-lazySMP vs SF-master)
Finished game 2 (SF-lazySMP vs SF-master): 1/2-1/2 {Draw by 3-fold repetition}
Score of SF-master vs SF-lazySMP: 0 - 0 - 2  [0.500] 2

[pgn]
[Event "Engine_Match"]
[Site "LZsT7910"]
[Date "2015.10.08"]
[Round "15"]
[White "SF-master"]
[Black "SF-lazySMP"]
[Result "1/2-1/2"]
[FEN "rnbqkb1r/pp1ppppp/2p4n/8/5P2/8/PPPPP1PP/RNBQKBNR w KQkq -"]
[PlyCount "201"]
[SetUp "1"]
[TimeControl "7200+30"]

1. Nf3 {-0.08/36 173s} e6 {+0.06/31 264s} 2. Nc3 {-0.07/37 222s}
Be7 {0.00/35 249s} 3. e4 {-0.06/33 15s} d5 {+0.11/34 157s} 4. d4 {-0.10/37 115s}
O-O {0.00/33 183s} 5. Bd3 {-0.05/35 168s} c5 {+0.05/35 219s}
6. dxc5 {0.00/36 180s} Nd7 {0.00/34 151s} 7. Be3 {0.00/34 25s}
Nxc5 {0.00/33 81s} 8. exd5 {0.00/38 197s} exd5 {+0.03/35 135s}
9. O-O {0.00/33 16s} Ng4 {+0.04/36 112s} 10. Nxd5 {-0.03/40 121s}
Nxe3 {0.00/38 191s} 11. Nxe3 {-0.06/40 119s} Nxd3 {0.00/37 151s}
12. Qxd3 {0.00/40 168s} Bc5 {0.00/37 16s} 13. Rfe1 {0.00/39 20s}
Be6 {0.00/38 114s} 14. Qxd8 {0.00/41 122s} Raxd8 {0.00/37 115s}
15. b3 {0.00/41 123s} f6 {+0.03/36 158s} 16. Kf1 {-0.01/41 124s}
g5 {+0.03/40 118s} 17. Rad1 {-0.03/42 227s} gxf4 {+0.03/39 134s}
18. Rxd8 {0.00/38 26s} Rxd8 {+0.03/43 118s} 19. Rd1 {-0.04/45 126s}
Rxd1+ {+0.03/41 119s} 20. Nxd1 {-0.04/46 127s} Bf5 {+0.03/40 119s}
21. Ne1 {-0.04/44 128s} Kf7 {+0.03/39 120s} 22. Nc3 {-0.02/42 129s}
Ke6 {+0.05/44 201s} 23. Ke2 {0.00/43 130s} f3+ {0.00/39 167s}
24. gxf3 {0.00/45 131s} Bd6 {0.00/41 55s} 25. Nb5 {0.00/49 143s}
Bxh2 {0.00/45 122s} 26. Nxa7 {0.00/49 136s} h5 {0.00/44 122s}
27. Nb5 {0.00/49 132s} Kd5 {0.00/47 160s} 28. c4+ {0.00/51 132s}
Kc5 {0.00/52 127s} 29. Nd3+ {0.00/51 132s} Bxd3+ {0.00/60 122s}
30. Kxd3 {0.00/55 132s} h4 {0.00/63 122s} 31. a3 {0.00/56 132s}
Bg1 {0.00/54 221s} 32. b4+ {0.00/58 131s} Kc6 {0.00/57 119s}
33. Ke2 {0.00/59 130s} h3 {0.00/58 326s} 34. Kf1 {0.00/60 129s}
Be3 {0.00/60 112s} 35. a4 {0.00/59 128s} Kd7 {0.00/60 111s}
36. Na3 {0.00/50 158s} Ke6 {0.00/58 110s} 37. Nc2 {0.00/55 125s}
Ba7 {0.00/56 109s} 38. c5 {0.00/55 123s} Kd5 {0.00/56 108s}
39. Ne3+ {0.00/51 121s} Kd4 {0.00/55 234s} 40. Nf5+ {0.00/52 120s}
Kc3 {0.00/57 16s} 41. Ne7 {0.00/52 118s} Kc4 {0.00/56 242s}
42. c6 {0.00/53 118s} bxc6 {0.00/63 276s} 43. Nxc6 {0.00/55 114s}
Be3 {0.00/64 93s} 44. Ne7 {0.00/55 112s} Bd4 {0.00/62 90s} 45. a5 {0.00/57 122s}
Kb5 {0.00/62 88s} 46. Nd5 {0.00/58 108s} f5 {0.00/63 87s}
47. Nf4 {+0.06/48 107s} Kxb4 {0.00/41 196s} 48. Nxh3 {+0.06/49 105s}
Kxa5 {0.00/45 81s} 49. Nf4 {+0.06/49 103s} Kb4 {0.00/45 153s}
50. Ng2 {+0.06/48 118s} Kc3 {0.00/39 77s} 51. Nh4 {+0.06/53 99s}
f4 {0.00/51 76s} 52. Kg2 {+0.06/55 97s} Be3 {0.00/49 74s}
53. Nf5 {+0.06/55 101s} Kd3 {0.00/44 73s} 54. Kh3 {+0.06/56 94s}
Kc4 {0.00/55 72s} 55. Kg4 {+0.06/55 92s} Kd5 {0.00/54 71s}
56. Kg5 {+0.06/56 91s} Bd2 {0.00/54 70s} 57. Kh5 {+0.06/58 89s}
Bc3 {0.00/57 69s} 58. Kg4 {+0.06/59 88s} Bd2 {0.00/57 68s}
59. Kg5 {+0.06/59 86s} Ke5 {0.00/51 67s} 60. Kg6 {+0.06/56 85s}
Bc3 {0.00/61 66s} 61. Nh6 {+0.06/62 83s} Ke6 {0.00/47 65s}
62. Ng4 {+0.06/51 82s} Bd4 {0.00/56 64s} 63. Kg5 {+0.06/47 81s}
Be3 {0.00/54 63s} 64. Nf6 {+0.06/47 79s} Bc5 {0.00/51 62s}
65. Ne4 {+0.06/50 86s} Be3 {0.00/56 61s} 66. Kg6 {+0.06/53 76s}
Ke5 {0.00/57 60s} 67. Nf6 {+0.06/58 75s} Bd4 {0.00/54 59s}
68. Nd7+ {+0.06/60 74s} Ke6 {0.00/57 118s} 69. Nf8+ {+0.06/61 73s}
Ke7 {0.00/56 56s} 70. Nh7 {+0.06/59 72s} Ke6 {0.00/61 55s}
71. Ng5+ {+0.06/51 70s} Ke5 {0.00/60 54s} 72. Nh3 {0.00/54 102s}
Be3 {0.00/57 53s} 73. Kg5 {0.00/53 26s} Bd2 {0.00/71 53s} 74. Nf2 {+0.06/60 91s}
Be3 {0.00/41 92s} 75. Nd3+ {+0.06/59 10s} Kd4 {0.00/52 50s}
76. Nb4 {+0.06/58 66s} Bd2 {0.00/49 49s} 77. Nc6+ {+0.06/55 65s}
Kd5 {0.00/53 363s} 78. Ne7+ {+0.06/53 64s} Kd6 {0.00/57 37s}
79. Ng6 {+0.06/58 63s} Ke6 {0.00/53 37s} 80. Nf8+ {+0.06/57 62s}
Ke7 {0.00/66 48s} 81. Nh7 {+0.06/43 61s} Ke6 {0.00/71 200s}
82. Nf6 {+0.06/56 61s} Bc1 {0.00/71 51s} 83. Kg6 {+0.06/56 60s}
Bb2 {0.00/56 27s} 84. Ng8 {+0.06/45 59s} Ba1 {0.00/57 28s}
85. Kh5 {+0.06/46 58s} Bc3 {0.00/48 28s} 86. Kg5 {+0.06/64 57s}
Bd2 {0.00/61 108s} 87. Kg6 {+0.06/65 56s} Bc1 {0.00/89 61s}
88. Nh6 {+0.06/50 56s} Be3 {0.00/62 35s} 89. Nf5 {+0.06/53 55s}
Ke5 {0.00/55 17s} 90. Ne7 {+0.06/48 54s} Bc1 {0.00/53 21s}
91. Kg5 {+0.06/53 53s} Be3 {0.00/54 22s} 92. Nc6+ {+0.06/53 53s}
Kd6 {0.00/53 22s} 93. Nb4 {+0.06/55 52s} Ke5 {0.00/51 23s}
94. Kg4 {+0.06/51 51s} Bd2 {0.00/48 24s} 95. Nd3+ {+0.06/51 50s}
Ke6 {0.00/47 24s} 96. Kg5 {+0.06/47 100s} Be3 {0.00/50 74s}
97. Kg6 {+0.06/56 48s} Bd2 {0.00/54 21s} 98. Kg7 {0.00/36 72s} Be3 {0.00/36 52s}
99. Kg6 {0.00/58 83s} Bd2 {0.00/80 25s} 100. Kg7 {0.00/60 109s}
Be3 {0.00/79 36s} 101. Kg6 {0.00/94 264s, Draw by 3-fold repetition} 1/2-1/2

[Event "Engine_Match"]
[Site "LZsT7910"]
[Date "2015.10.08"]
[Round "16"]
[White "SF-lazySMP"]
[Black "SF-master"]
[Result "1/2-1/2"]
[FEN "rnbqkb1r/pp1ppppp/2p4n/8/5P2/8/PPPPP1PP/RNBQKBNR w KQkq -"]
[PlyCount "119"]
[SetUp "1"]
[TimeControl "7200+30"]

1. Nf3 {-0.09/32 434s} e6 {+0.12/34 178s} 2. e4 {-0.04/35 979s}
d5 {+0.09/35 172s} 3. Nc3 {0.00/33 159s} Be7 {0.00/37 186s}
4. d4 {-0.05/34 142s} O-O {0.00/35 22s} 5. Bd3 {0.00/33 152s} Na6 {0.00/38 135s}
6. a3 {0.00/33 370s} c5 {0.00/39 118s} 7. O-O {0.00/35 95s} dxe4 {0.00/39 136s}
8. Nxe4 {0.00/36 166s} Nf5 {0.00/39 117s} 9. Bxa6 {0.00/36 21s}
bxa6 {0.00/41 117s} 10. Nxc5 {0.00/39 97s} Nxd4 {0.00/43 118s}
11. Qxd4 {0.00/41 98s} Qxd4+ {0.00/42 119s} 12. Nxd4 {0.00/42 98s}
Bxc5 {0.00/45 120s} 13. Be3 {0.00/39 99s} Bb7 {0.00/44 120s}
14. b4 {+0.05/36 184s} Be7 {-0.05/41 162s} 15. Nb3 {+0.03/37 99s}
Bf6 {-0.06/37 17s} 16. Rac1 {+0.07/36 100s} Bb2 {-0.06/40 124s}
17. Rb1 {+0.05/37 101s} Bc3 {-0.10/41 125s} 18. Rbd1 {+0.06/37 102s}
Rfd8 {-0.06/38 176s} 19. Rxd8+ {+0.06/39 102s} Rxd8 {-0.06/1 0s}
20. Bxa7 {+0.04/38 225s} Be4 {0.00/41 129s} 21. Be3 {0.00/40 102s}
Bxc2 {0.00/45 130s} 22. Nc5 {0.00/42 103s} Bb2 {0.00/45 131s}
23. Nxa6 {0.00/42 144s} Bxa3 {-0.01/43 236s} 24. b5 {0.00/43 103s}
Be4 {0.00/36 20s} 25. b6 {0.00/43 111s} Rd3 {0.00/45 134s}
26. Bc5 {0.00/42 104s} Rd2 {0.00/47 134s} 27. Re1 {0.00/43 105s}
Bxc5+ {0.00/47 134s} 28. Nxc5 {0.00/46 170s} Bc6 {0.00/49 134s}
29. b7 {0.00/44 169s} Rxg2+ {0.00/50 134s} 30. Kf1 {0.00/45 16s}
Bxb7 {0.00/51 134s} 31. Rb1 {0.00/47 104s} Rxh2 {0.00/53 133s}
32. Rxb7 {0.00/46 104s} g6 {0.00/54 133s} 33. Nd3 {0.00/48 104s}
Rh1+ {0.00/54 132s} 34. Ke2 {0.00/50 103s} Rh5 {0.00/57 131s}
35. Kf2 {0.00/53 102s} Rf5 {0.00/59 134s} 36. Kg3 {0.00/50 101s}
g5 {0.00/51 128s} 37. fxg5 {0.00/56 113s} Rxg5+ {0.00/58 126s}
38. Kf2 {0.00/55 99s} h6 {0.00/57 125s} 39. Nf4 {0.00/54 98s} h5 {0.00/53 123s}
40. Nh3 {0.00/46 96s} Rf5+ {0.00/53 121s} 41. Ke2 {0.00/45 95s}
Rd5 {0.00/57 119s} 42. Kf1 {0.00/53 106s} Rd1+ {0.00/58 167s}
43. Ke2 {0.00/66 92s} Rd5 {0.00/56 523s} 44. Ng1 {0.00/59 91s}
Ra5 {0.00/53 102s} 45. Nf3 {0.00/53 89s} Kg7 {0.00/55 101s}
46. Ke3 {0.00/52 88s} Ra3+ {0.00/55 99s} 47. Kf4 {0.00/55 86s} Kf6 {0.00/56 97s}
48. Rb5 {0.00/48 85s} Kg6 {0.00/52 96s} 49. Ne5+ {0.00/50 84s} Kf6 {0.00/53 94s}
50. Nd7+ {0.00/49 82s} Kg6 {0.00/54 92s} 51. Ne5+ {0.00/53 162s}
Kf6 {0.00/77 151s} 52. Nf3 {0.00/63 142s} Kg6 {0.00/73 121s}
53. Re5 {0.00/40 74s} Kh6 {0.00/51 85s} 54. Rb5 {0.00/49 73s} Ra4+ {0.00/59 84s}
55. Ke5 {0.00/46 72s} Kg6 {0.00/60 83s} 56. Rb2 {0.00/45 71s} Ra5+ {0.00/54 83s}
57. Kf4 {0.00/52 70s} Ra4+ {0.00/67 80s} 58. Ke5 {0.00/68 126s}
Ra5+ {0.00/63 108s} 59. Kf4 {0.00/60 89s} Ra4+ {0.00/81 104s}
60. Ke5 {0.00/70 88s, Draw by 3-fold repetition} 1/2-1/2
[/pgn]

[zullil]

Since SF-lazySMP is now competing in TCEC Stage 3, I've ended the match. Based on 16 games, I'm prepared to say that SF-lazySMP is probably not amazingly stronger or shockingly weaker than SF-master. Also prepared to say that saying more would require thousands of hours of computer time.

[Dann Corbit]

I ran my own contest of 7 games.

My time control was 1/2 hour for the first move, then 40 moves in 40 minutes then game in 40.

I chose the 1/2 hour first move to fill up the hash table (to see if that would generate problems).

The match finished 0-0-7 (all 7 games were draws).

[Laskos]

Since SF-lazySMP is now competing in TCEC Stage 3, I've ended the match. Based on 16 games, I'm prepared to say that SF-lazySMP is probably not amazingly stronger or shockingly weaker than SF-master. Also prepared to say that saying more would require thousands of hours of computer time.

I got some bad signs. Can you check the rough NPS of lazySMP compared to master in some 3 minute searches on 20 threads? What I get not having 20 core hardware: I checked at fixed nodes on 20 threads, to separate the NPS issue (unreachable to me) from scaling issue for effective speedup. If NPS are comparable, fixed nodes tests don't look good, lazySMP loses badly on 20 threads to master. And the problem is very nasty, fixed time tests show that lazySMP kicks in much faster, with over-performance on 20 threads at shorter time controls (fixed time).

[zullil]

Can you check the rough NPS of lazySMP compared to master in some 3 minute searches on 20 threads?

Code: Select all

Stockfish-master
Stockfish-77b4

r1bqk2r/1p2nppp/p1npp3/8/2P1P3/1NNBb3/PP2QPPP/R3K2R w KQkq - 0 1
info depth 38 seldepth 47 multipv 1 score cp 43 upperbound nodes 4351334705 nps 25099702 hashfull 974 tbhits 0 time 173362 pv e2e3 e8g8
info depth 38 currmove e2e3 currmovenumber 1
info nodes 4527074525 time 180001
bestmove e2e3 ponder e8g8

3qr1k1/1b1n2pp/2n2p2/p3p3/Pr2P3/1P1P2P1/1B1N1QBP/R4RK1 w - - 0 1
info depth 37 seldepth 53 multipv 1 score cp 156 nodes 4367250618 nps 26216192 hashfull 977 tbhits 0 time 166586 pv b2c3 d7b8 c3b4 c6b4 d2f3 b8c6 a1d1 b7a6 d1d2 c6d$
info depth 38 currmove b2c3 currmovenumber 1
info nodes 4744783656 time 180002
bestmove b2c3 ponder d7b8

8/8/1p1p4/4k3/1r2p3/2N1K3/1P4R1/8 w - - 0 1
info depth 41 seldepth 72 multipv 1 score cp 470 lowerbound nodes 3262421221 nps 18683296 hashfull 190 tbhits 0 time 174617 pv g2g5
info depth 41 currmove g2g5 currmovenumber 1
info nodes 3358822719 time 180002
bestmove g2g5


Stockfish-lazySMP
Stockfish-9f5c

r1bqk2r/1p2nppp/p1npp3/8/2P1P3/1NNBb3/PP2QPPP/R3K2R w KQkq - 0 1
info depth 33 seldepth 0 multipv 1 score cp 43 nodes 3152664431 nps 29282438 hashfull 937 tbhits 0 time 107664 pv e2e3 e8g8 a1d1 d8c7 d3e2 f8d8 e1g1 c8d7 d1d2 c6e5 $
info depth 34 currmove e2e3 currmovenumber 1
info depth 34 currmove f2e3 currmovenumber 2
info depth 34 currmove a1d1 currmovenumber 3
info nodes 5306929269 time 180003
bestmove e2e3 ponder d8c7

3qr1k1/1b1n2pp/2n2p2/p3p3/Pr2P3/1P1P2P1/1B1N1QBP/R4RK1 w - - 0 1
info depth 31 seldepth 0 multipv 1 score cp 177 nodes 5583210073 nps 34081578 hashfull 986 tbhits 0 time 163819 pv b2c3 b7a6 c3b4 c6b4 f1d1 b4d3 f2e3 d7c5 d2c4 a6c4$
info depth 32 currmove b2c3 currmovenumber 1
info nodes 6153567907 time 180000
bestmove b2c3 ponder b7a6

8/8/1p1p4/4k3/1r2p3/2N1K3/1P4R1/8 w - - 0 1
info depth 38 seldepth 0 multipv 1 score cp 608 lowerbound nodes 11125819123 nps 62988694 hashfull 329 tbhits 0 time 176632 pv g2g5
info depth 38 currmove g2g5 currmovenumber 1
info nodes 11345769125 time 180004
bestmove g2g5

[Laskos]

Can you check the rough NPS of lazySMP compared to master in some 3 minute searches on 20 threads?

Code: Select all

Stockfish-master
Stockfish-77b4

r1bqk2r/1p2nppp/p1npp3/8/2P1P3/1NNBb3/PP2QPPP/R3K2R w KQkq - 0 1
info depth 38 seldepth 47 multipv 1 score cp 43 upperbound nodes 4351334705 nps 25099702 hashfull 974 tbhits 0 time 173362 pv e2e3 e8g8
info depth 38 currmove e2e3 currmovenumber 1
info nodes 4527074525 time 180001
bestmove e2e3 ponder e8g8

3qr1k1/1b1n2pp/2n2p2/p3p3/Pr2P3/1P1P2P1/1B1N1QBP/R4RK1 w - - 0 1
info depth 37 seldepth 53 multipv 1 score cp 156 nodes 4367250618 nps 26216192 hashfull 977 tbhits 0 time 166586 pv b2c3 d7b8 c3b4 c6b4 d2f3 b8c6 a1d1 b7a6 d1d2 c6d$
info depth 38 currmove b2c3 currmovenumber 1
info nodes 4744783656 time 180002
bestmove b2c3 ponder d7b8

8/8/1p1p4/4k3/1r2p3/2N1K3/1P4R1/8 w - - 0 1
info depth 41 seldepth 72 multipv 1 score cp 470 lowerbound nodes 3262421221 nps 18683296 hashfull 190 tbhits 0 time 174617 pv g2g5
info depth 41 currmove g2g5 currmovenumber 1
info nodes 3358822719 time 180002
bestmove g2g5


Stockfish-lazySMP
Stockfish-9f5c

r1bqk2r/1p2nppp/p1npp3/8/2P1P3/1NNBb3/PP2QPPP/R3K2R w KQkq - 0 1
info depth 33 seldepth 0 multipv 1 score cp 43 nodes 3152664431 nps 29282438 hashfull 937 tbhits 0 time 107664 pv e2e3 e8g8 a1d1 d8c7 d3e2 f8d8 e1g1 c8d7 d1d2 c6e5 $
info depth 34 currmove e2e3 currmovenumber 1
info depth 34 currmove f2e3 currmovenumber 2
info depth 34 currmove a1d1 currmovenumber 3
info nodes 5306929269 time 180003
bestmove e2e3 ponder d8c7

3qr1k1/1b1n2pp/2n2p2/p3p3/Pr2P3/1P1P2P1/1B1N1QBP/R4RK1 w - - 0 1
info depth 31 seldepth 0 multipv 1 score cp 177 nodes 5583210073 nps 34081578 hashfull 986 tbhits 0 time 163819 pv b2c3 b7a6 c3b4 c6b4 f1d1 b4d3 f2e3 d7c5 d2c4 a6c4$
info depth 32 currmove b2c3 currmovenumber 1
info nodes 6153567907 time 180000
bestmove b2c3 ponder b7a6

8/8/1p1p4/4k3/1r2p3/2N1K3/1P4R1/8 w - - 0 1
info depth 38 seldepth 0 multipv 1 score cp 608 lowerbound nodes 11125819123 nps 62988694 hashfull 329 tbhits 0 time 176632 pv g2g5
info depth 38 currmove g2g5 currmovenumber 1
info nodes 11345769125 time 180004
bestmove g2g5

Thanks for your time, Louis!

Well, it seems that the endgame NPS are pretty crazily diverging, and it's hard to speculate on nodes per move basis. But let's be speculative and separate the NPS issue to long TC from the issue about scaling with threads. I will assume that 70% of the game weight is on opening-middlegame with NPS factor of 1.3 between Master and Lazy_SMP. The rest 30% of the game weight is on endgame with NPS factor of 3.0 between Master and Lazy_SMP. The combined factor for NPS is (1.3**0.7) * (3**0.3) ~1.67. That is the weighted average NPS advantage of Lazy_SMP on 20 threads, very long time control. I went even further, setting 2.00 NPS advantage of Lazy_SMP, and then used Fixed Nodes approach.

10 million nodes/move SF Master versus 20 million nodes/move SF Lazy_SMP:

Code: Select all

Score of SF master vs SF lazy_smp: 54 - 33 - 138  [0.547] 225
ELO difference: 33
Finished match

Lazy_SMP seems to lose pretty badly even with twice nodes (or NPS) of Master. The fixed nodes result I think holds to very long time control, where the NPS issue stabilizes to that at most a factor of 2.00 advantage of Lazy_SMP.

This is all a bit of improvisation, one needs to check at chosen time control on at least 20 core machine. The problem with fixed time is that YBW kicks on very slowly compared to Lazy_SMP, so at short time controls Lazy_SMP overperforms. To check to really long time controls, one must go at least to (an educated guess) 900 seconds + 9 seconds games or so, where 20 threaded YBW stabilizes. I think someone with a 20 core box can decide if Lazy_SMP performs reasonably in 2-3 days of testing, no more than 100 games at 900''+9'' would be necessary, as ELO differences are pretty large.

[zullil]

Lazy_SMP seems to lose pretty badly even with twice nodes (or NPS) of Master. The fixed nodes result I think holds to very long time control, where the NPS issue stabilizes to that at most a factor of 2.00 advantage of Lazy_SMP.

This is all a bit of improvisation, one needs to check at chosen time control on at least 20 core machine. The problem with fixed time is that YBW kicks on very slowly compared to Lazy_SMP, so at short time controls Lazy_SMP overperforms. To check to really long time controls, one must go at least to (an educated guess) 900 seconds + 9 seconds games or so, where 20 threaded YBW stabilizes. I think someone with a 20 core box can decide if Lazy_SMP performs reasonably in 2-3 days of testing, no more than 100 games at 900''+9'' would be necessary, as ELO differences are pretty large.

Kai,

I'm considering a 900" + 3'' test. 20 cores. 2-move book. 16 GB hash per engine. Is it still your belief that 100 games will reveal something meaningful?

Louis