Komodo MCTS scales worse with TC than Komodo A/B?

[tpoppins]

We already found and fixed a bug that was probably related to the crash, although I'm not sure that the problem is fully solved yet. I think that it is likely that the huge number of threads rather than the long time control may be related to the problem.

I've seen at least four crashes (out of about 180 games) in the ongoing MCTS 4CPU test. Default settings with 5-men Syzygy and MCTS Hash set to what was the default in the previous version (320 MB). Got WER reports on three of them:

Code: Select all

Faulting application name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Faulting module name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Exception code: 0xc00000fd
Fault offset: 0x000000000003e366
Faulting process id: 0x13d8
Faulting application start time: 0x01d47c2637142d62
Faulting application path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Faulting module path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Report Id: 0278af43-e826-11e8-b60a-d89d67f3f098

Code: Select all

Faulting application name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Faulting module name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Exception code: 0xc0000005
Fault offset: 0x0000000000005aac
Faulting process id: 0xa54
Faulting application start time: 0x01d47c14286425ef
Faulting application path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Faulting module path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Report Id: ae225d69-e812-11e8-b60a-d89d67f3f098

Code: Select all

Faulting application name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Faulting module name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Exception code: 0xc0000005
Fault offset: 0x000000000000f45f
Faulting process id: 0xc28
Faulting application start time: 0x01d47d3207b314a9
Faulting application path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Faulting module path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Report Id: b1ff3b3c-e92e-11e8-982b-d89d67f3f098

Are these errors what you referred to above or is it something that needs to be looked at more closely?

[lkaufman]

We already found and fixed a bug that was probably related to the crash, although I'm not sure that the problem is fully solved yet. I think that it is likely that the huge number of threads rather than the long time control may be related to the problem.

I've seen at least four crashes (out of about 180 games) in the ongoing MCTS 4CPU test. Default settings with 5-men Syzygy and MCTS Hash set to what was the default in the previous version (320 MB). Got WER reports on three of them:

Code: Select all

Faulting application name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Faulting module name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Exception code: 0xc00000fd
Fault offset: 0x000000000003e366
Faulting process id: 0x13d8
Faulting application start time: 0x01d47c2637142d62
Faulting application path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Faulting module path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Report Id: 0278af43-e826-11e8-b60a-d89d67f3f098

Code: Select all

Faulting application name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Faulting module name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Exception code: 0xc0000005
Fault offset: 0x0000000000005aac
Faulting process id: 0xa54
Faulting application start time: 0x01d47c14286425ef
Faulting application path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Faulting module path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Report Id: ae225d69-e812-11e8-b60a-d89d67f3f098

Code: Select all

Faulting application name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Faulting module name: komodo-12.2.2-64bit.exe, version: 0.0.0.0, time stamp: 0x5be7762d
Exception code: 0xc0000005
Fault offset: 0x000000000000f45f
Faulting process id: 0xc28
Faulting application start time: 0x01d47d3207b314a9
Faulting application path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Faulting module path: C:\Program Files\ChessBase\Engines.x64\komodo-12.2.2-64bit.exe
Report Id: b1ff3b3c-e92e-11e8-982b-d89d67f3f098

Are these errors what you referred to above or is it something that needs to be looked at more closely?

The bug we found and fixed was in the position repetition code, which probably caused the crash in TCEC. If the crashes happened in positions where a repetition was happening or appeared likely to happen they were probably related to it. If you know whether this was the case or not please advise.

[tpoppins]

I had all the games replayed and now have the score for only one of them:

[pgn][Event "Komodo 12.2.2 MCTS 64-bit 4CPU 40/40"] [Site "Dual E5-2690v2"] [Date "2018.11.15"] [Round "2"] [White "Komodo 12.2.2 MCTS 64-bit 4CPU"] [Black "Xiphos 0.4 64-bit 4CPU"] [Result "0-1"] [ECO "D84"] [GameDuration "01:14:15"] [GameEndTime "2018-11-15T18:39:08.890 Eastern Standard Time"] [GameStartTime "2018-11-15T17:24:53.737 Eastern Standard Time"] [Opening "Gruenfeld"] [PlyCount "142"] [Termination "stalled connection"] [TimeControl "40/1200"] [Variation "Gruenfeld gambit accepted"] 1. d4 {book} Nf6 {book} 2. c4 {book} g6 {book} 3. Nc3 {book} d5 {book} 4. Bf4 {book} Bg7 {book} 5. e3 {book} O-O {book} 6. cxd5 {book} Nxd5 {book} 7. Nxd5 {book} Qxd5 {book} 8. Bxc7 {book} Na6 {book} 9. Bxa6 {book} Qxg2 {book} 10. Qf3 {book} Qxf3 {book} 11. Nxf3 {book} bxa6 {book} 12. Rg1 {book} Bb7 {book} 13. Ke2 {+0.19/21 38s} f6 {0.00/30 39s} 14. Rgd1 {+0.31/21 38s} g5 {0.00/29 37s} 15. Rac1 {+0.24/21 37s} Rf7 {0.00/30 68s} 16. d5 {+0.33/21 36s} g4 {0.00/31 51s} 17. Nh4 {+0.28/21 36s} f5 {0.00/32 24s} 18. Bg3 {+0.21/21 35s} a5 {0.00/33 51s} 19. b3 {+0.23/21 35s} Bb2 {-0.14/33 94s} 20. Rc2 {+0.36/21 35s} Ba6+ {-0.05/32 34s} 21. Ke1 {+0.34/21 34s} Ba3 {-0.16/33 54s} 22. f3 {+0.15/21 35s} Rc8 {-0.09/32 52s} 23. Rxc8+ {+0.12/21 35s} Bxc8 {-0.10/30 20s} 24. Kf2 {+0.15/21 35s} Rf8 {-0.09/28 21s} 25. Be5 {+0.29/21 35s} Kf7 {-0.04/30 39s} 26. fxg4 {+0.23/21 35s} fxg4 {-0.10/32 65s} 27. e4 {+0.18/21 36s} Ke8+ {-0.14/31 30s} 28. Kg3 {+0.07/21 36s} Bb4 {-0.13/32 48s} 29. Bb8 {+0.21/21 36s} h5 {-0.08/31 34s} 30. Ng6 {+0.09/21 37s} Rf3+ {-0.17/33 35s} 31. Kh4 {+0.07/21 37s} Re3 {-0.14/32 33s} 32. Kxh5 {+0.06/21 38s} Rxe4 {-0.28/31 27s} 33. Rf1 {+0.07/21 39s} Bd6 {-0.15/33 23s} 34. Bxd6 {+0.07/21 39s} exd6 {-0.09/36 65s} 35. Nf4 {+0.04/21 40s} a4 {-0.06/36 57s} 36. bxa4 {+0.03/21 42s} Rxa4 {-0.06/37 36s} 37. Rf2 {+0.04/21 43s} a5 {-0.05/35 26s} 38. Kg6 {+0.02/21 45s} Rc4 {0.00/39 25s} 39. Kg7 {+0.01/21 47s} Ke7 {0.00/41 39s} 40. Nh5 {+0.01/21 50s} Re4 {0.00/40 35s} 41. Kg6 {+0.01/21 51s} Re5 {0.00/42 42s} 42. Nf4 {0.00/21 47s} Bd7 {+0.10/36 68s} 43. Rg2 {0.00/21 44s} a4 {0.00/37 46s} 44. Kh6 {0.00/21 41s} Kf6 {+0.07/36 46s} 45. Nh5+ {0.00/21 39s} Kf7 {0.00/41 30s} 46. Rf2+ {0.00/20 37s} Ke7 {0.00/39 20s} 47. Nf6 {0.00/20 35s} a3 {0.00/40 36s} 48. Nxd7 {0.00/20 33s} Kxd7 {0.00/42 20s} 49. Rd2 {0.00/20 32s} Kc7 {0.00/39 20s} 50. Rd3 {0.00/20 31s} Kb6 {0.00/41 25s} 51. Rxa3 {0.00/20 30s} Rxd5 {0.00/38 22s} 52. Ra4 {0.00/20 29s} Rd2 {0.00/39 20s} 53. Ra8 {0.00/20 28s} Rxh2+ {+1.00/41 20s} 54. Kg5 {0.00/19 23s} g3 {+0.94/40 17s} 55. Kf4 {0.00/19 23s} g2 {+0.86/42 16s} 56. Rg8 {0.00/20 27s} Kc5 {+0.85/41 16s} 57. Ke3 {0.00/20 27s} Rh3+ {+0.98/36 22s} 58. Kf2 {0.00/20 27s} Ra3 {+1.22/41 40s} 59. Kxg2 {0.00/22 26s} Rxa2+ {+0.80/45 28s} 60. Kf3 {-0.58/21 26s} Kd4 {+0.77/44 16s} 61. Rg4+ {-0.44/21 26s} Kc3 {+0.75/45 17s} 62. Rg5 {-0.44/21 26s} Kc4 {+0.75/47 18s} 63. Rg4+ {-0.44/21 26s} Kd3 {+0.73/48 24s} 64. Rg5 {-0.44/21 26s} Ra4 {+0.72/48 25s} 65. Rd5+ {-0.41/21 26s} Rd4 {+0.72/47 24s} 66. Ra5 {-0.44/21 26s} Kc4 {+0.71/44 32s} 67. Ra6 {-0.28/21 27s} d5 {+0.76/42 43s} 68. Ke2 {-0.01/21 27s} Re4+ {+0.71/41 25s} 69. Kd2 {-0.01/21 27s} Kd4 {+0.78/41 32s} 70. Ra4+ {0.00/23 27s} Ke5 {+0.24/40 36s} 71. Rxe4+ {0.00/24 28s} Kxe4 {0.00/48 23s, White's connection stalls} 0-1 [/pgn]

I believe it corresponds to the last error report in my previous post. The other games were almost all TB draws as well, with perhaps one late middlegame involving a repetition.