Help with Scorpio 3.0.8

Dann Corbit · Post by **Dann Corbit** » Sat Jun 06, 2020 1:27 am

0:
0:             Welcome to NTS (Network training server) for Scorpio Zero
0:                         http://scorpiozero.ddns.net
0:
0:
0:
0:
0:  Please enter a username and password and an account will be created.
0:  The password will be stored in plain text, so avoid using passwords used
0:  for sensitive applications.
0:
0:
0: login:
0: password:
0:
0: **** Starting NTS session as dann ****
0:
0: <version>
0: 2
0: 2
0: </version>
0: <parameters> 800 125 110 25 0 100
0: </parameters>
0: <checksum>
0: 1026903171
0: http://scorpiozero.ddns.net/scorpiozero/nets-9/ID-0-model-3.uff
0: </checksum>
0: Skipping net download.
0: Executing job : C:\Users\dannc\chess\nn-dist-master\scripts\job-client.bat
[1712] Games 404: + 209 - 160 = 35

Dann Corbit · Post by **Dann Corbit** » Sat Jun 06, 2020 1:28 am

I don't care if the output looks strange as long as it is working well.
I think it is worthwhile to get it configured right so you will get some nice data.

Dann Corbit · Post by **Dann Corbit** » Sat Jun 06, 2020 1:57 am

Another thing occurs to me (though it may be different because we are in the learning phase rather than the phase of playing chess using the information learned). It might be important to have the program run well on my machine because the big contests like TCEC are usually run on Windows machines.

Daniel Shawul · Post by **Daniel Shawul** » Sat Jun 06, 2020 2:18 am

Dann Corbit wrote: ↑Sat Jun 06, 2020 1:57 am Another thing occurs to me (though it may be different because we are in the learning phase rather than the phase of playing chess using the information learned). It might be important to have the program run well on my machine because the big contests like TCEC are usually run on Windows machines.

Yes, up until recently the windows installer was not even working properly. It finally seems to be working properly now.
I did optimizations for multi-GPU systems a week ago but there is more to do there, and a 50% GPU utllization is still not that great.
Btw TCEC has now switched to linux machine but I used to have a headache when they were using windows as well.

The error message you got witch CUDNN_STATUS_MAPPING_ERROR is a long-standing issue of TensorRT that I was
not able to solve fully. It is not specific to windows though.

Dann Corbit · Post by **Dann Corbit** » Sat Jun 06, 2020 11:16 pm

It just hangs now.
It does not seem to help if I restart the batch either.
I can't tell what is going on with this "less chatty" version.

MMarco · Post by **MMarco** » Sun Jun 07, 2020 5:23 am

Thanks for Scorpio Mr. Shawul.

I have a GTX 1660 ti, and here my bench with sm=96 (the 1660 ti has 24 sm cores) for INT8:

Code: Select all

PS C:\Users\redoi\OneDrive\Documents\ENGINES\SHAWUL\Scorpio\bin\Windows> ./scorpio.bat go quit
feature done=0
Number of cores 8 of 8
treeht 83886080 X 320 = 25600.0 MB
processors [96]
ht 67108864 X 16 = 1024.0 MB
eht 262144 X 8 X 96 = 384.0 MB
pht 32768 X 24 X 96 = 72.0 MB
EgbbProbe 4.3 by Daniel Shawul
egbb_cache 4084 X 8216 = 32.0 MB
180 egbbs loaded !
nn_cache 131072 X 1568 = 196.0 MB
Loading neural network : C:\Users\redoi\OneDrive\Documents\ENGINES\SHAWUL\Scorpio\nets-scorpio/ens-net-20x256.uff
With "INT8" precision
Loading graph on /gpu:0
0. main_input 2048 = 32 8 8
1. policy/Reshape 1024 = 1024 1 1
2. value/BiasAdd 3 = 3 1 1
Neural network loaded !
nn_cache 131072 X 1568 = 196.0 MB
Loading neural network : C:\Users\redoi\OneDrive\Documents\ENGINES\SHAWUL\Scorpio\nets-maddex/ME.uff
With "INT8" precision
Loading graph on /gpu:0
0. main_input 7168 = 112 8 8
1. value_head 3 = 3 1 1
2. policy_head 1858 = 1858 1 1
Neural network loaded !
loading_time = 3s
# rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
# [st = 13336ms, mt = 29250ms , hply = 0 , moves_left 10]
32 29 92 6713 32 7257 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 Ke1-g1 Nf6xe4 Rf1-e1 Ne4-d6 Nf3xe5 Bf8-e7 Bb5-f1 Nc6xe5 Re1xe5 Ke8-g8 d2-d4 Be7-f6 Re5-e1 Rf8-e8 c2-c3 Re8xe1 Qd1xe1 Nd6-e8 Bc1-f4 d7-d5 Nb1-d2 c7-c6 Bf1-d3
33 29 187 14641 33 7808 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Bf8-e7 Rf1-e1 b7-b5 Ba4-b3 d7-d6 c2-c3 Ke8-g8 h2-h3 Bc8-b7 d2-d4 Rf8-e8 Nb1-d2 Be7-f8 a2-a4 h7-h6 Bb3-c2 Ra8-b8
34 29 281 22787 34 8083 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Bf8-e7 Rf1-e1 b7-b5 Ba4-b3 d7-d6 c2-c3 Ke8-g8 h2-h3 Bc8-b7 d2-d4 Rf8-e8 Nb1-d2 Be7-f8 a2-a4 h7-h6 Bb3-c2 Ra8-b8 Bc2-b1
35 29 377 31032 35 8218 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Nf6xe4 d2-d4 b7-b5 Ba4-b3 d7-d5 d4xe5 Bc8-e6 c2-c3 Bf8-e7 Nb1-d2 Ne4-c5 Bb3-c2 d5-d4 Nd2-b3 d4xc3 Nb3xc5 Be7xc5 Bc2-e4 Qd8-d7 b2xc3 Ra8-d8 Qd1xd7 Be6xd7
36 28 472 39346 36 8320 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Nf6xe4 d2-d4 b7-b5 Ba4-b3 d7-d5 d4xe5 Bc8-e6 c2-c3 Bf8-e7 Nb1-d2 Ne4-c5 Bb3-c2 d5-d4 Nd2-b3 d4-d3 Nb3xc5 d3xc2 Qd1xd8 Ra8xd8 Nc5xe6 f7xe6 Bc1-e3 Rd8-d5
37 28 566 47392 37 8368 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Nf6xe4 d2-d4 b7-b5 Ba4-b3 d7-d5 d4xe5 Bc8-e6 c2-c3 Bf8-e7 Nb1-d2 Ne4-c5 Bb3-c2 d5-d4 Nd2-b3 d4-d3 Nb3xc5 d3xc2 Qd1xd8 Ra8xd8 Nc5xe6 f7xe6 Bc1-e3 Rd8-d5
38 28 662 56207 38 8489 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Nf6xe4 d2-d4 b7-b5 Ba4-b3 d7-d5 d4xe5 Bc8-e6 c2-c3 Bf8-e7 Nb1-d2 Ne4-c5 Bb3-c2 d5-d4 Nd2-b3 d4-d3 Nb3xc5 d3xc2 Qd1xd8 Ra8xd8 Nc5xe6 f7xe6 Bc1-e3 Rd8-d5
39 30 761 63355 39 8317 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Nf6xe4 d2-d4 b7-b5 Ba4-b3 d7-d5 d4xe5 Bc8-e6 c2-c3 Bf8-e7 Nb1-d2 Ne4-c5 Bb3-c2 d5-d4 Nd2-b3 d4xc3 Nb3xc5 Be7xc5 Bc2-e4 Qd8-d7 b2xc3 Ra8-d8 Qd1xd7 Be6xd7
40 31 858 70278 40 8189 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Bf8-e7 Rf1-e1 b7-b5 Ba4-b3 d7-d6 c2-c3 Ke8-g8 h2-h3 Bc8-b7 d2-d4 Rf8-e8 Nb1-d2 Be7-f8 a2-a4 h7-h6 Bb3-c2 Ra8-b8 Bc2-b1 b5-b4 a4-a5 b4xc3
41 31 928 75788 41 8163 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Bf8-e7 Rf1-e1 b7-b5 Ba4-b3 d7-d6 c2-c3 Ke8-g8 h2-h3 Bc8-b7 d2-d4 Rf8-e8 Nb1-d2 Be7-f8 a2-a4 h7-h6 Bb3-c2 Ra8-b8 Bc2-b1 b5-b4 a4-a5 b4xc3
# nodes 908972 <0% qnodes> time 9356ms nps 97153 eps 0 nneps 8157 tbhits 0
# Tree: nodes 97918 depth 31 pps 8154 visits 75826 qcalls 0 scalls 0
move e2e4
Bye Bye

I ran a 200 match blitz match (time control bench to CCRL blitz TC), against Komodo 14 4CPU (rated 3525 in CCRL Blitz list), Scorpio using ens-net-20x256.uff and ME.uff has suggested in the ini file right after the installation (which went without problems in windows 10 by the way). Cute-Chess 1.0.0 GUI (the script installation went easy too):
Games: http://s000.tinyupload.com/index.php?fi ... 3684651401

Code: Select all

Rank Name                Elo    +    - games score oppo. draws
   1 Komodo 14 4CPU     3525   18   18   200   57%  3489   77%
   2 Scorpio 3.08 INT8  3489   18   18   200   43%  3525   77%

   1 Komodo 14 4CPU     3525 200.0 (114.5 :  85.5)
                             200.0 (114.5 :  85.5) Scorpio 3.08 INT8  3489
   2 Scorpio 3.08 INT8  3489 200.0 ( 85.5 : 114.5)

                   Ko Sc
Komodo 14 4CPU        97
Scorpio 3.08 INT8   2

Apart from the heavy numbers of draws, here are a few observations (I watched many games):

Game 189: Scorpio is a safe 2 pawns up but it failed to convert (move 43 and onward). It is reported that Lc0 fails to convert many winning endgames, but from what I observed the problem is even more acute with Scorpio and its current nets. Many promising positions were blown away (draw or even lost in some cases). [pgn][Event "My Tournament"]
[Site "?"]
[Date "2020.06.06"]
[Round "95"]
[White "Scorpio 3.08 INT8"]
[Black "Komodo 14 4CPU"]
[Result "1/2-1/2"]
[ECO "E00"]
[Opening "Catalan Opening"]
[PlyCount "191"]
[Termination "adjudication"]
[TimeControl "40/110"]

1. d4 {book} Nf6 {book} 2. c4 {book} e6 {book} 3. g3 {book} Bb4+ {book}
4. Bd2 {book} a5 {book} 5. Nf3 {book} d5 {-0.24/23 5.5s} 6. Qc2 {+0.35/43 6.0s}
c5 {-0.06/24 3.6s} 7. cxd5 {+0.31/39 2.7s} cxd4 {-0.31/22 2.3s}
8. Nxd4 {+0.33/43 2.4s} exd5 {-0.30/24 7.1s} 9. Bg2 {+0.41/38 2.3s}
Nc6 {-0.28/23 4.1s} 10. Nxc6 {+0.64/38 2.2s} bxc6 {-0.34/24 2.1s}
11. Qxc6+ {+0.58/42 2.1s} Bd7 {-0.52/24 2.3s} 12. Qc2 {+0.60/42 2.5s}
Rc8 {-0.57/25 4.9s} 13. Qd1 {+0.63/38 1.6s} Qb6 {-0.45/26 2.9s}
14. O-O {+0.63/37 1.2s} O-O {-0.48/27 1.8s} 15. Bxb4 {+0.60/38 1.9s}
Qxb4 {-0.18/23 2.0s} 16. b3 {+0.58/38 1.7s} h5 {-0.19/22 2.4s}
17. Nd2 {+0.81/38 2.5s} h4 {-0.30/23 2.6s} 18. Re1 {+0.87/37 1.4s}
h3 {-0.80/23 7.2s} 19. Bf1 {+0.97/38 2.3s} Rfe8 {-0.97/26 8.3s}
20. Nf3 {+0.93/37 2.2s} a4 {-1.00/23 3.7s} 21. Ng5 {+1.34/42 4.4s}
Rc3 {-0.96/25 4.2s} 22. Bxh3 {+1.39/39 2.5s} Bxh3 {-1.02/26 2.2s}
23. Nxh3 {+1.34/35 0.89s} axb3 {-1.09/27 2.1s} 24. axb3 {+1.38/36 1.8s}
Rxb3 {-0.97/26 1.8s} 25. Nf4 {+1.35/39 2.2s} d4 {-1.07/23 2.1s}
26. Qc2 {+1.36/40 3.0s} Rc3 {-1.01/25 4.1s} 27. Qa2 {+1.36/41 3.0s}
Qd6 {-0.99/22 3.0s} 28. Qa4 {+1.34/41 3.6s} Qe5 {-1.14/22 1.6s}
29. Rab1 {+1.42/41 3.5s} Qc5 {-1.10/24 3.9s} 30. h4 {+1.45/39 3.8s}
Ra3 {-1.48/25 1.5s} 31. Qb4 {+1.19/44 4.4s} Qxb4 {-1.14/26 1.6s}
32. Rxb4 {+1.40/35 1.3s} d3 {-1.23/26 1.5s} 33. e3 {+1.25/39 3.3s}
Nd7 {-1.23/28 1.7s} 34. Rd1 {+1.53/40 4.6s} Ne5 {-1.14/28 1.3s}
35. Rb5 {+1.62/39 2.5s} Ng4 {-1.16/27 2.7s} 36. Rd5 {+1.58/40 5.0s}
Ra2 {-1.44/24 1.7s} 37. Rf1 {+1.49/41 3.0s} g6 {-1.43/25 1.4s}
38. Rxd3 {+2.09/36 3.7s} Rc8 {-1.62/25 2.1s} 39. Nd5 {+2.16/39 5.1s}
Rcc2 {-1.56/28 2.8s} 40. Nb4 {+2.63/38 6.9s} Rd2 {-1.70/27 1.9s}
41. Rd1 {+2.75/41 4.1s} Rxd3 {-1.76/29 3.7s} 42. Nxd3 {+2.90/37 2.9s}
f5 {-1.85/31 8.0s} 43. Rf1 {+3.36/41 6.6s} Kg7 {-1.95/30 7.5s}
44. Nf4 {+3.50/38 3.6s} Kh6 {-1.98/32 17s} 45. Ne6 {+3.39/63 8.1s}
Ne5 {-1.85/29 5.0s} 46. Ng5 {+3.64/37 3.6s} Nd3 {-1.95/29 3.0s}
47. Kg2 {+2.68/60 8.1s} Rb2 {-1.86/28 2.1s} 48. Nh3 {+3.77/39 3.8s}
Nc5 {-1.90/27 2.7s} 49. Rc1 {+4.21/66 5.8s} Ne4 {-1.72/27 1.8s}
50. Rc6 {+3.99/72 4.5s} Ra2 {-1.56/28 1.8s} 51. Re6 {+3.75/67 4.6s}
Kg7 {-1.71/30 3.6s} 52. Rb6 {+3.71/68 4.6s} Kh6 {-1.80/28 2.0s}
53. Rb7 {+3.62/70 4.7s} Rc2 {-1.76/32 7.9s} 54. Re7 {+3.70/71 4.0s}
Ra2 {-1.74/28 1.2s} 55. Rf7 {+3.64/65 2.0s} Ra6 {-1.77/30 2.8s}
56. Rc7 {+3.56/66 3.7s} Ra2 {-1.78/28 1.1s} 57. Re7 {+3.42/71 2.3s}
Rc2 {-1.80/29 4.7s} 58. Ra7 {+3.52/41 2.5s} Rd2 {-1.70/28 3.0s}
59. Kf3 {+3.47/71 4.3s} Rb2 {-1.77/27 1.1s} 60. Ra4 {+3.59/47 2.9s}
Nd2+ {-1.73/28 3.4s} 61. Kg2 {+3.61/64 2.5s} Ne4 {-1.84/27 1.2s}
62. Ra6 {+3.11/68 1.6s} Rc2 {-1.56/29 2.3s} 63. Kf3 {+3.26/43 2.0s}
Kg7 {-1.54/25 0.88s} 64. Ra4 {+3.61/41 1.6s} Kh6 {-1.51/26 1.1s}
65. Rd4 {+3.34/61 1.4s} Kg7 {-1.44/27 2.6s} 66. Rxe4 {+4.28/41 1.3s}
fxe4+ {-1.04/23 1.1s} 67. Kf4 {+3.98/59 1.3s} Rc4 {-0.88/29 2.8s}
68. Ke5 {+2.44/59 1.7s} Kh6 {-0.42/28 1.5s} 69. g4 {+1.20/61 1.3s}
Rc1 {0.00/33 0.63s} 70. Ng5 {+0.49/62 1.4s} Rh1 {0.00/33 0.70s}
71. Kf4 {+0.19/61 1.1s} Rxh4 {0.00/30 0.62s} 72. Nf7+ {+0.04/55 1.1s}
Kg7 {0.00/34 0.80s} 73. Nd6 {+0.02/53 0.86s} Rh1 {0.00/35 0.85s}
74. Nxe4 {+0.06/54 0.90s} Re1 {0.00/37 0.84s} 75. Ng5 {+0.05/53 1.2s}
Ra1 {0.00/40 0.81s} 76. Ne6+ {+0.03/56 1.2s} Kf7 {0.00/37 0.81s}
77. Nd4 {+0.02/48 0.98s} Ra2 {0.00/37 0.85s} 78. f3 {+0.05/48 0.88s}
Ra3 {0.00/41 0.99s} 79. g5 {+0.03/44 1.2s} Ra2 {0.00/46 1.1s}
80. Ke4 {+0.04/56 1.6s} Ra4 {0.00/50 1.3s} 81. f4 {+0.07/62 5.7s}
Ra7 {0.00/45 3.0s} 82. Nc6 {+0.07/59 6.4s} Ra4+ {0.00/47 3.4s}
83. Kd5 {+0.09/59 4.9s} Kf8 {0.00/48 3.2s} 84. Ne5 {+0.10/69 6.6s}
Kg7 {0.00/48 2.7s} 85. e4 {+0.09/63 3.4s} Ra5+ {0.00/48 2.4s}
86. Ke6 {+0.13/64 5.0s} Ra6+ {0.00/49 2.8s} 87. Kd7 {+0.11/66 5.3s}
Ra7+ {0.00/50 2.2s} 88. Kd6 {+0.12/67 3.4s} Ra6+ {0.00/55 2.2s}
89. Nc6 {+0.44/64 4.2s} Rb6 {0.00/46 2.3s} 90. e5 {+0.43/62 4.5s}
Kf8 {0.00/45 2.1s} 91. Kd7 {+0.25/63 3.4s} Rb7+ {0.00/53 2.4s}
92. Ke6 {+0.05/66 3.8s} Rf7 {0.00/49 2.2s} 93. f5 {-0.05/70 3.5s}
Rxf5 {0.00/53 2.0s} 94. Ne7 {-0.13/66 2.1s} Rxg5 {0.00/65 2.7s}
95. Kf6 {-0.09/65 1.5s} Rg1 {0.00/77 1.8s}
96. Nxg6+ {-0.08/67 2.1s, Draw by adjudication: SyzygyTB} 1/2-1/2[/pgn]

Game 161: The endgame net seems to kick in around move 74, where the piececount = 10, although it is indicated that it would come in at piececount=14 in the .ini file. It is the same in other games too.

Game 123: On move 31, the eval depth spike suddently to 59 from around 40 from earlier moves. There are other depth spike in this game, and in many other games. I'm not sure if this is a normal behavior. Note that it lost this game, in spite of being an healthy pawn up at this point. [pgn][Event "My Tournament"]
[Site "?"]
[Date "2020.06.06"]
[Round "62"]
[White "Scorpio 3.08 INT8"]
[Black "Komodo 14 4CPU"]
[Result "0-1"]
[ECO "A53"]
[Opening "Old Indian defense"]
[PlyCount "242"]
[Termination "adjudication"]
[TimeControl "40/110"]

1. d4 {book} Nf6 {book} 2. c4 {book} d6 {book} 3. Nc3 {book} Nbd7 {book}
4. e4 {book} e5 {book} 5. Nf3 {book} Be7 {book} 6. Be2 {+0.97/43 6.1s}
O-O {-0.66/21 2.3s} 7. O-O {+0.94/41 1.7s} a5 {-0.78/26 8.3s}
8. Qc2 {+0.98/46 5.0s} Re8 {-0.65/25 3.5s} 9. Be3 {+0.90/45 2.7s}
Ng4 {-0.75/23 2.4s} 10. Bd2 {+0.81/37 1.6s} exd4 {-0.57/24 2.1s}
11. Nxd4 {+0.87/35 0.89s} Nc5 {-0.71/25 3.0s} 12. Rad1 {+0.89/39 2.8s}
Bf6 {-0.71/23 2.1s} 13. Bc1 {+0.84/37 1.4s} Bxd4 {-0.76/26 4.1s}
14. Rxd4 {+0.77/35 1.1s} Qh4 {-0.74/27 1.9s} 15. Bxg4 {+0.71/38 1.8s}
Qxg4 {-0.75/27 3.9s} 16. f3 {+0.69/45 3.2s} Qd7 {-0.78/26 7.3s}
17. Nd5 {+0.62/41 2.7s} Na6 {-0.71/26 8.2s} 18. Qc3 {+0.94/42 4.4s}
b6 {-0.66/23 2.5s} 19. e5 {+1.02/41 3.0s} Kh8 {-0.84/24 4.0s}
20. Bf4 {+2.80/40 3.6s} Bb7 {-1.60/24 4.3s} 21. exd6 {+3.08/35 0.87s}
Bxd5 {-1.43/24 1.7s} 22. Rxd5 {+3.21/37 1.6s} Nb4 {-2.10/21 1.9s}
23. Rdd1 {+3.21/39 2.2s} cxd6 {-1.74/22 1.8s} 24. a3 {+3.55/38 2.7s}
Nc6 {-1.98/22 1.8s} 25. Rxd6 {+3.58/54 2.8s} Qf5 {-1.39/24 1.7s}
26. Bg3 {+3.37/55 2.8s} Qc5+ {-1.61/25 3.3s} 27. Kh1 {+3.37/37 1.7s}
b5 {-1.48/26 2.6s} 28. cxb5 {+3.34/40 3.7s} Qxb5 {-1.49/25 3.3s}
29. Ra1 {+3.46/41 3.0s} Ne7 {-1.18/25 3.7s} 30. a4 {+3.66/37 2.4s}
Qg5 {-1.55/26 1.9s} 31. Rd7 {+2.27/59 5.5s} Nf5 {-0.98/21 1.9s}
32. Bf2 {+2.12/42 2.6s} Re2 {-0.74/23 2.1s} 33. Qc5 {+2.28/39 2.8s}
Rae8 {-0.96/22 2.0s} 34. Rxf7 {+1.40/38 3.2s} Re1+ {-0.54/29 1.9s}
35. Rxe1 {+1.26/39 2.0s} Rxe1+ {-0.40/30 4.1s} 36. Bxe1 {+1.22/38 2.6s}
Ng3+ {-0.36/29 1.7s} 37. hxg3 {+1.20/38 2.8s} Qxc5 {-0.46/29 1.8s}
38. Rf4 {+1.13/39 3.7s} Kg8 {-0.48/31 2.5s} 39. Bc3 {+1.06/40 5.4s}
g6 {-0.43/32 4.3s} 40. Rd4 {+0.93/38 8.2s} Qh5+ {-0.43/31 1.3s}
41. Rh4 {+0.84/41 6.4s} Qd5 {-0.36/35 13s} 42. Kg1 {+0.81/43 8.6s}
Qc5+ {-0.32/32 7.9s} 43. Kh2 {+0.79/42 6.0s} Qd5 {-0.29/33 3.2s}
44. Rg4 {+0.77/43 7.5s} Qh5+ {-0.27/30 2.2s} 45. Kg1 {+0.73/38 3.1s}
Qc5+ {-0.35/33 2.8s} 46. Rd4 {+0.73/42 4.0s} Qg5 {-0.27/35 3.2s}
47. g4 {+0.69/46 5.5s} h5 {-0.25/30 1.9s} 48. gxh5 {+0.68/41 4.0s}
Qxh5 {-0.24/30 1.8s} 49. Re4 {+0.71/47 7.4s} Qg5 {-0.23/30 2.2s}
50. Kf1 {+0.67/47 6.6s} Kf7 {-0.23/29 2.1s} 51. Rc4 {+0.63/44 4.5s}
Qe3 {-0.22/29 1.9s} 52. Rg4 {+0.64/41 3.6s} Qd3+ {-0.23/31 3.1s}
53. Kf2 {+0.66/41 3.0s} Qf5 {-0.21/30 1.8s} 54. Re4 {+0.60/45 3.2s}
Qc5+ {-0.22/26 1.6s} 55. Kg3 {+0.54/45 3.5s} Qg5+ {-0.22/29 1.4s}
56. Rg4 {+0.51/43 2.2s} Qc5 {-0.22/30 2.0s} 57. Rf4+ {+0.53/40 1.8s}
Kg8 {-0.22/31 1.9s} 58. Re4 {+0.57/40 1.7s} Qg5+ {-0.21/28 1.6s}
59. Kf2 {+0.54/40 1.6s} Kf7 {-0.21/28 1.4s} 60. Rd4 {+0.51/44 2.5s}
Qc5 {-0.21/31 1.7s} 61. Kf1 {+0.53/44 2.5s} Qf5 {-0.19/29 1.5s}
62. Rh4 {+0.45/40 1.5s} Qg5 {-0.19/32 3.4s} 63. Rg4 {+0.51/42 1.8s}
Qf5 {-0.19/30 1.5s} 64. Kf2 {+0.51/39 1.2s} Qc2+ {-0.19/30 1.7s}
65. Kg3 {+0.56/40 1.3s} Qc1 {-0.19/29 1.5s} 66. Rf4+ {+0.51/38 0.61s}
Kg8 {-0.19/29 1.6s} 67. Rd4 {+0.51/42 1.3s} Qg5+ {-0.18/28 1.7s}
68. Rg4 {+0.47/42 1.3s} Qf5 {-0.14/28 1.8s} 69. Kh2 {+0.37/40 1.00s}
Kf7 {-0.13/30 2.3s} 70. Re4 {+0.40/40 1.2s} Qh5+ {-0.11/30 1.8s}
71. Kg1 {+0.37/38 0.82s} Qc5+ {-0.11/29 1.8s} 72. Rd4 {+0.36/38 0.69s}
Qg5 {-0.12/31 2.3s} 73. g4 {+0.39/40 1.4s} Qe3+ {0.00/32 1.7s}
74. Kg2 {+0.43/35 0.54s} Qe2+ {0.00/36 2.4s} 75. Kg3 {+0.39/36 0.69s}
g5 {0.00/36 1.8s} 76. Re4 {+0.31/40 1.7s} Qd1 {0.00/37 2.8s}
77. Kf2 {+0.28/38 1.3s} Qc2+ {0.00/36 2.1s} 78. Ke3 {+0.28/40 1.8s}
Qc1+ {0.00/39 2.7s} 79. Kd4 {+0.20/40 1.9s} Qd1+ {0.00/41 2.8s}
80. Kc5 {+0.11/39 1.6s} Qxf3 {0.00/38 2.7s} 81. Rc4 {+0.06/63 7.7s}
Qb7 {0.00/39 3.1s} 82. Bxa5 {+0.02/60 3.8s} Qe7+ {0.00/36 3.0s}
83. Kb5 {+0.02/63 6.6s} Qe5+ {0.00/35 3.0s} 84. Ka6 {-0.03/64 4.0s}
Qxb2 {0.00/38 2.8s} 85. Bb6 {-0.05/63 4.0s} Qe2 {0.00/39 2.7s}
86. Kb5 {-0.08/62 2.3s} Ke6 {0.00/40 2.5s} 87. Kc5 {-0.12/68 5.7s}
Qg2 {0.00/43 2.3s} 88. Rd4 {+0.04/63 5.5s} Qc2+ {0.00/47 2.5s}
89. Kb4 {0.00/69 4.9s} Qc6 {0.00/42 2.4s} 90. Bc5 {-0.01/62 4.8s}
Qb7+ {0.00/39 2.2s} 91. Kc3 {-0.06/68 3.4s} Ke5 {0.00/41 2.4s}
92. Rc4 {-0.16/61 4.5s} Qb1 {0.00/28 4.6s} 93. Ba3 {-0.15/60 3.9s}
Qe1+ {+0.32/26 2.4s} 94. Kc2 {-0.15/58 3.8s} Kd5 {+0.20/25 2.1s}
95. Kb3 {-0.34/59 3.7s} Qg3+ {+0.07/28 6.9s} 96. Kb4 {-0.46/59 3.2s}
Qb8+ {+0.14/29 1.8s} 97. Kc3 {-0.32/54 2.0s} Qe5+ {+0.09/31 2.0s}
98. Kb4 {-0.32/61 2.6s} Qe7+ {+0.04/27 1.7s} 99. Kb3 {-0.37/57 2.1s}
Qe2 {+0.04/31 2.5s} 100. Rb4 {-0.15/55 2.1s} Qe3+ {+0.51/27 6.2s}
101. Kb2 {-0.20/57 3.1s} Ke6 {+1.14/26 3.7s} 102. Rc4 {-0.21/55 1.8s}
Qd3 {+1.37/25 1.7s} 103. Rb4 {-0.37/50 1.5s} Kd7 {+1.51/28 1.9s}
104. Ka2 {-0.29/57 1.6s} Qc3 {+1.66/27 2.2s} 105. a5 {-0.32/58 1.1s}
Kc7 {+1.68/27 2.5s} 106. a6 {-0.22/50 0.99s} Qd2+ {+2.04/27 3.3s}
107. Kb1 {-0.28/56 1.6s} Qd3+ {+2.14/28 2.1s} 108. Kb2 {-0.42/58 1.6s}
Qxa6 {+1.97/28 2.9s} 109. Ka2 {-0.54/63 1.5s} Qe6+ {+2.75/23 2.3s}
110. Kb2 {-0.78/57 2.1s} Qe2+ {+2.76/25 2.2s} 111. Ka1 {-1.29/57 2.1s}
Qe5+ {+3.03/23 3.1s} 112. Kb1 {-3.04/55 1.8s} Qc3 {+3.54/23 2.1s}
113. Ka2 {-3.55/52 0.88s} Kc6 {+3.78/24 3.6s} 114. Rb3 {-3.86/57 1.6s}
Qc4 {+3.81/22 4.0s} 115. Bb2 {-3.89/54 1.2s} Qxg4 {+4.67/20 4.2s}
116. Bc1 {-4.06/54 1.3s} Qe2+ {+6.63/20 2.1s} 117. Ka3 {-4.21/48 1.0s}
Qe1 {+10.40/18 2.7s} 118. Be3 {-4.56/51 1.8s} g4 {+250.00/17 0.62s}
119. Bf4 {-4.47/49 0.81s} Qa1+ {+250.00/42 0.32s} 120. Kb4 {-4.47/50 0s}
Qd4+ {+250.00/50 0.45s} 121. Ka5 {-7.75/58 11s}
Qxf4 {+250.00/53 0.47s, Black wins by adjudication: SyzygyTB} 0-1[/pgn]

I'll be following Scorpio progress. Thank you for your efforts.

Edit: The openings for the match against Komodo were "Noomen Short Lines 2015.pgn".

Daniel Shawul · Post by **Daniel Shawul** » Sun Jun 07, 2020 2:56 pm

Dann Corbit wrote: ↑Sat Jun 06, 2020 11:16 pm It just hangs now.
It does not seem to help if I restart the batch either.
I can't tell what is going on with this "less chatty" version.

Yeah, it is not the most stable server/client out there. I have had many problems overnight with my vast.ai clients doing nothing
while burning money. I need to make this more stable, and also find out the right parameters for resuming training from an SL net without
rewriting what the net has learned. I have stopped the training run now, and maybe in a week everything will be fine at which point I will
let you know. Thanks for the contribution, much appreciated!

Daniel Shawul · Post by **Daniel Shawul** » Sun Jun 07, 2020 3:05 pm

MMarco wrote: ↑Sun Jun 07, 2020 5:23 am Thanks for Scorpio Mr. Shawul.

I have a GTX 1660 ti, and here my bench with sm=96 (the 1660 ti has 24 sm cores) for INT8:
Code: Select all
PS C:\Users\redoi\OneDrive\Documents\ENGINES\SHAWUL\Scorpio\bin\Windows> ./scorpio.bat go quit
feature done=0
Number of cores 8 of 8
treeht 83886080 X 320 = 25600.0 MB
processors [96]
ht 67108864 X 16 = 1024.0 MB
eht 262144 X 8 X 96 = 384.0 MB
pht 32768 X 24 X 96 = 72.0 MB
EgbbProbe 4.3 by Daniel Shawul
egbb_cache 4084 X 8216 = 32.0 MB
180 egbbs loaded !
nn_cache 131072 X 1568 = 196.0 MB
Loading neural network : C:\Users\redoi\OneDrive\Documents\ENGINES\SHAWUL\Scorpio\nets-scorpio/ens-net-20x256.uff
With "INT8" precision
Loading graph on /gpu:0
0. main_input 2048 = 32 8 8
1. policy/Reshape 1024 = 1024 1 1
2. value/BiasAdd 3 = 3 1 1
Neural network loaded !
nn_cache 131072 X 1568 = 196.0 MB
Loading neural network : C:\Users\redoi\OneDrive\Documents\ENGINES\SHAWUL\Scorpio\nets-maddex/ME.uff
With "INT8" precision
Loading graph on /gpu:0
0. main_input 7168 = 112 8 8
1. value_head 3 = 3 1 1
2. policy_head 1858 = 1858 1 1
Neural network loaded !
loading_time = 3s
# rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1
# [st = 13336ms, mt = 29250ms , hply = 0 , moves_left 10]
32 29 92 6713 32 7257 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 Ng8-f6 Ke1-g1 Nf6xe4 Rf1-e1 Ne4-d6 Nf3xe5 Bf8-e7 Bb5-f1 Nc6xe5 Re1xe5 Ke8-g8 d2-d4 Be7-f6 Re5-e1 Rf8-e8 c2-c3 Re8xe1 Qd1xe1 Nd6-e8 Bc1-f4 d7-d5 Nb1-d2 c7-c6 Bf1-d3
33 29 187 14641 33 7808 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Bf8-e7 Rf1-e1 b7-b5 Ba4-b3 d7-d6 c2-c3 Ke8-g8 h2-h3 Bc8-b7 d2-d4 Rf8-e8 Nb1-d2 Be7-f8 a2-a4 h7-h6 Bb3-c2 Ra8-b8
34 29 281 22787 34 8083 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Bf8-e7 Rf1-e1 b7-b5 Ba4-b3 d7-d6 c2-c3 Ke8-g8 h2-h3 Bc8-b7 d2-d4 Rf8-e8 Nb1-d2 Be7-f8 a2-a4 h7-h6 Bb3-c2 Ra8-b8 Bc2-b1
35 29 377 31032 35 8218 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Nf6xe4 d2-d4 b7-b5 Ba4-b3 d7-d5 d4xe5 Bc8-e6 c2-c3 Bf8-e7 Nb1-d2 Ne4-c5 Bb3-c2 d5-d4 Nd2-b3 d4xc3 Nb3xc5 Be7xc5 Bc2-e4 Qd8-d7 b2xc3 Ra8-d8 Qd1xd7 Be6xd7
36 28 472 39346 36 8320 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Nf6xe4 d2-d4 b7-b5 Ba4-b3 d7-d5 d4xe5 Bc8-e6 c2-c3 Bf8-e7 Nb1-d2 Ne4-c5 Bb3-c2 d5-d4 Nd2-b3 d4-d3 Nb3xc5 d3xc2 Qd1xd8 Ra8xd8 Nc5xe6 f7xe6 Bc1-e3 Rd8-d5
37 28 566 47392 37 8368 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Nf6xe4 d2-d4 b7-b5 Ba4-b3 d7-d5 d4xe5 Bc8-e6 c2-c3 Bf8-e7 Nb1-d2 Ne4-c5 Bb3-c2 d5-d4 Nd2-b3 d4-d3 Nb3xc5 d3xc2 Qd1xd8 Ra8xd8 Nc5xe6 f7xe6 Bc1-e3 Rd8-d5
38 28 662 56207 38 8489 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Nf6xe4 d2-d4 b7-b5 Ba4-b3 d7-d5 d4xe5 Bc8-e6 c2-c3 Bf8-e7 Nb1-d2 Ne4-c5 Bb3-c2 d5-d4 Nd2-b3 d4-d3 Nb3xc5 d3xc2 Qd1xd8 Ra8xd8 Nc5xe6 f7xe6 Bc1-e3 Rd8-d5
39 30 761 63355 39 8317 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Nf6xe4 d2-d4 b7-b5 Ba4-b3 d7-d5 d4xe5 Bc8-e6 c2-c3 Bf8-e7 Nb1-d2 Ne4-c5 Bb3-c2 d5-d4 Nd2-b3 d4xc3 Nb3xc5 Be7xc5 Bc2-e4 Qd8-d7 b2xc3 Ra8-d8 Qd1xd7 Be6xd7
40 31 858 70278 40 8189 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Bf8-e7 Rf1-e1 b7-b5 Ba4-b3 d7-d6 c2-c3 Ke8-g8 h2-h3 Bc8-b7 d2-d4 Rf8-e8 Nb1-d2 Be7-f8 a2-a4 h7-h6 Bb3-c2 Ra8-b8 Bc2-b1 b5-b4 a4-a5 b4xc3
41 31 928 75788 41 8163 0 e2-e4 e7-e5 Ng1-f3 Nb8-c6 Bf1-b5 a7-a6 Bb5-a4 Ng8-f6 Ke1-g1 Bf8-e7 Rf1-e1 b7-b5 Ba4-b3 d7-d6 c2-c3 Ke8-g8 h2-h3 Bc8-b7 d2-d4 Rf8-e8 Nb1-d2 Be7-f8 a2-a4 h7-h6 Bb3-c2 Ra8-b8 Bc2-b1 b5-b4 a4-a5 b4xc3
# nodes 908972 <0% qnodes> time 9356ms nps 97153 eps 0 nneps 8157 tbhits 0
# Tree: nodes 97918 depth 31 pps 8154 visits 75826 qcalls 0 scalls 0
move e2e4
Bye Bye
I ran a 200 match blitz match (time control bench to CCRL blitz TC), against Komodo 14 4CPU (rated 3525 in CCRL Blitz list), Scorpio using ens-net-20x256.uff and ME.uff has suggested in the ini file right after the installation (which went without problems in windows 10 by the way). Cute-Chess 1.0.0 GUI (the script installation went easy too):
Games: http://s000.tinyupload.com/index.php?fi ... 3684651401
Code: Select all
Rank Name                Elo    +    - games score oppo. draws
   1 Komodo 14 4CPU     3525   18   18   200   57%  3489   77%
   2 Scorpio 3.08 INT8  3489   18   18   200   43%  3525   77%

   1 Komodo 14 4CPU     3525 200.0 (114.5 :  85.5)
                             200.0 (114.5 :  85.5) Scorpio 3.08 INT8  3489
   2 Scorpio 3.08 INT8  3489 200.0 ( 85.5 : 114.5)

                   Ko Sc
Komodo 14 4CPU        97
Scorpio 3.08 INT8   2
Apart from the heavy numbers of draws, here are a few observations (I watched many games):

Game 189: Scorpio is a safe 2 pawns up but it failed to convert (move 43 and onward). It is reported that Lc0 fails to convert many winning endgames, but from what I observed the problem is even more acute with Scorpio and its current nets. Many promising positions were blown away (draw or even lost in some cases). [pgn][Event "My Tournament"]
[Site "?"]
[Date "2020.06.06"]
[Round "95"]
[White "Scorpio 3.08 INT8"]
[Black "Komodo 14 4CPU"]
[Result "1/2-1/2"]
[ECO "E00"]
[Opening "Catalan Opening"]
[PlyCount "191"]
[Termination "adjudication"]
[TimeControl "40/110"]

1. d4 {book} Nf6 {book} 2. c4 {book} e6 {book} 3. g3 {book} Bb4+ {book}
4. Bd2 {book} a5 {book} 5. Nf3 {book} d5 {-0.24/23 5.5s} 6. Qc2 {+0.35/43 6.0s}
c5 {-0.06/24 3.6s} 7. cxd5 {+0.31/39 2.7s} cxd4 {-0.31/22 2.3s}
8. Nxd4 {+0.33/43 2.4s} exd5 {-0.30/24 7.1s} 9. Bg2 {+0.41/38 2.3s}
Nc6 {-0.28/23 4.1s} 10. Nxc6 {+0.64/38 2.2s} bxc6 {-0.34/24 2.1s}
11. Qxc6+ {+0.58/42 2.1s} Bd7 {-0.52/24 2.3s} 12. Qc2 {+0.60/42 2.5s}
Rc8 {-0.57/25 4.9s} 13. Qd1 {+0.63/38 1.6s} Qb6 {-0.45/26 2.9s}
14. O-O {+0.63/37 1.2s} O-O {-0.48/27 1.8s} 15. Bxb4 {+0.60/38 1.9s}
Qxb4 {-0.18/23 2.0s} 16. b3 {+0.58/38 1.7s} h5 {-0.19/22 2.4s}
17. Nd2 {+0.81/38 2.5s} h4 {-0.30/23 2.6s} 18. Re1 {+0.87/37 1.4s}
h3 {-0.80/23 7.2s} 19. Bf1 {+0.97/38 2.3s} Rfe8 {-0.97/26 8.3s}
20. Nf3 {+0.93/37 2.2s} a4 {-1.00/23 3.7s} 21. Ng5 {+1.34/42 4.4s}
Rc3 {-0.96/25 4.2s} 22. Bxh3 {+1.39/39 2.5s} Bxh3 {-1.02/26 2.2s}
23. Nxh3 {+1.34/35 0.89s} axb3 {-1.09/27 2.1s} 24. axb3 {+1.38/36 1.8s}
Rxb3 {-0.97/26 1.8s} 25. Nf4 {+1.35/39 2.2s} d4 {-1.07/23 2.1s}
26. Qc2 {+1.36/40 3.0s} Rc3 {-1.01/25 4.1s} 27. Qa2 {+1.36/41 3.0s}
Qd6 {-0.99/22 3.0s} 28. Qa4 {+1.34/41 3.6s} Qe5 {-1.14/22 1.6s}
29. Rab1 {+1.42/41 3.5s} Qc5 {-1.10/24 3.9s} 30. h4 {+1.45/39 3.8s}
Ra3 {-1.48/25 1.5s} 31. Qb4 {+1.19/44 4.4s} Qxb4 {-1.14/26 1.6s}
32. Rxb4 {+1.40/35 1.3s} d3 {-1.23/26 1.5s} 33. e3 {+1.25/39 3.3s}
Nd7 {-1.23/28 1.7s} 34. Rd1 {+1.53/40 4.6s} Ne5 {-1.14/28 1.3s}
35. Rb5 {+1.62/39 2.5s} Ng4 {-1.16/27 2.7s} 36. Rd5 {+1.58/40 5.0s}
Ra2 {-1.44/24 1.7s} 37. Rf1 {+1.49/41 3.0s} g6 {-1.43/25 1.4s}
38. Rxd3 {+2.09/36 3.7s} Rc8 {-1.62/25 2.1s} 39. Nd5 {+2.16/39 5.1s}
Rcc2 {-1.56/28 2.8s} 40. Nb4 {+2.63/38 6.9s} Rd2 {-1.70/27 1.9s}
41. Rd1 {+2.75/41 4.1s} Rxd3 {-1.76/29 3.7s} 42. Nxd3 {+2.90/37 2.9s}
f5 {-1.85/31 8.0s} 43. Rf1 {+3.36/41 6.6s} Kg7 {-1.95/30 7.5s}
44. Nf4 {+3.50/38 3.6s} Kh6 {-1.98/32 17s} 45. Ne6 {+3.39/63 8.1s}
Ne5 {-1.85/29 5.0s} 46. Ng5 {+3.64/37 3.6s} Nd3 {-1.95/29 3.0s}
47. Kg2 {+2.68/60 8.1s} Rb2 {-1.86/28 2.1s} 48. Nh3 {+3.77/39 3.8s}
Nc5 {-1.90/27 2.7s} 49. Rc1 {+4.21/66 5.8s} Ne4 {-1.72/27 1.8s}
50. Rc6 {+3.99/72 4.5s} Ra2 {-1.56/28 1.8s} 51. Re6 {+3.75/67 4.6s}
Kg7 {-1.71/30 3.6s} 52. Rb6 {+3.71/68 4.6s} Kh6 {-1.80/28 2.0s}
53. Rb7 {+3.62/70 4.7s} Rc2 {-1.76/32 7.9s} 54. Re7 {+3.70/71 4.0s}
Ra2 {-1.74/28 1.2s} 55. Rf7 {+3.64/65 2.0s} Ra6 {-1.77/30 2.8s}
56. Rc7 {+3.56/66 3.7s} Ra2 {-1.78/28 1.1s} 57. Re7 {+3.42/71 2.3s}
Rc2 {-1.80/29 4.7s} 58. Ra7 {+3.52/41 2.5s} Rd2 {-1.70/28 3.0s}
59. Kf3 {+3.47/71 4.3s} Rb2 {-1.77/27 1.1s} 60. Ra4 {+3.59/47 2.9s}
Nd2+ {-1.73/28 3.4s} 61. Kg2 {+3.61/64 2.5s} Ne4 {-1.84/27 1.2s}
62. Ra6 {+3.11/68 1.6s} Rc2 {-1.56/29 2.3s} 63. Kf3 {+3.26/43 2.0s}
Kg7 {-1.54/25 0.88s} 64. Ra4 {+3.61/41 1.6s} Kh6 {-1.51/26 1.1s}
65. Rd4 {+3.34/61 1.4s} Kg7 {-1.44/27 2.6s} 66. Rxe4 {+4.28/41 1.3s}
fxe4+ {-1.04/23 1.1s} 67. Kf4 {+3.98/59 1.3s} Rc4 {-0.88/29 2.8s}
68. Ke5 {+2.44/59 1.7s} Kh6 {-0.42/28 1.5s} 69. g4 {+1.20/61 1.3s}
Rc1 {0.00/33 0.63s} 70. Ng5 {+0.49/62 1.4s} Rh1 {0.00/33 0.70s}
71. Kf4 {+0.19/61 1.1s} Rxh4 {0.00/30 0.62s} 72. Nf7+ {+0.04/55 1.1s}
Kg7 {0.00/34 0.80s} 73. Nd6 {+0.02/53 0.86s} Rh1 {0.00/35 0.85s}
74. Nxe4 {+0.06/54 0.90s} Re1 {0.00/37 0.84s} 75. Ng5 {+0.05/53 1.2s}
Ra1 {0.00/40 0.81s} 76. Ne6+ {+0.03/56 1.2s} Kf7 {0.00/37 0.81s}
77. Nd4 {+0.02/48 0.98s} Ra2 {0.00/37 0.85s} 78. f3 {+0.05/48 0.88s}
Ra3 {0.00/41 0.99s} 79. g5 {+0.03/44 1.2s} Ra2 {0.00/46 1.1s}
80. Ke4 {+0.04/56 1.6s} Ra4 {0.00/50 1.3s} 81. f4 {+0.07/62 5.7s}
Ra7 {0.00/45 3.0s} 82. Nc6 {+0.07/59 6.4s} Ra4+ {0.00/47 3.4s}
83. Kd5 {+0.09/59 4.9s} Kf8 {0.00/48 3.2s} 84. Ne5 {+0.10/69 6.6s}
Kg7 {0.00/48 2.7s} 85. e4 {+0.09/63 3.4s} Ra5+ {0.00/48 2.4s}
86. Ke6 {+0.13/64 5.0s} Ra6+ {0.00/49 2.8s} 87. Kd7 {+0.11/66 5.3s}
Ra7+ {0.00/50 2.2s} 88. Kd6 {+0.12/67 3.4s} Ra6+ {0.00/55 2.2s}
89. Nc6 {+0.44/64 4.2s} Rb6 {0.00/46 2.3s} 90. e5 {+0.43/62 4.5s}
Kf8 {0.00/45 2.1s} 91. Kd7 {+0.25/63 3.4s} Rb7+ {0.00/53 2.4s}
92. Ke6 {+0.05/66 3.8s} Rf7 {0.00/49 2.2s} 93. f5 {-0.05/70 3.5s}
Rxf5 {0.00/53 2.0s} 94. Ne7 {-0.13/66 2.1s} Rxg5 {0.00/65 2.7s}
95. Kf6 {-0.09/65 1.5s} Rg1 {0.00/77 1.8s}
96. Nxg6+ {-0.08/67 2.1s, Draw by adjudication: SyzygyTB} 1/2-1/2[/pgn]

Game 161: The endgame net seems to kick in around move 74, where the piececount = 10, although it is indicated that it would come in at piececount=14 in the .ini file. It is the same in other games too.

Game 123: On move 31, the eval depth spike suddently to 59 from around 40 from earlier moves. There are other depth spike in this game, and in many other games. I'm not sure if this is a normal behavior. Note that it lost this game, in spite of being an healthy pawn up at this point. [pgn][Event "My Tournament"]
[Site "?"]
[Date "2020.06.06"]
[Round "62"]
[White "Scorpio 3.08 INT8"]
[Black "Komodo 14 4CPU"]
[Result "0-1"]
[ECO "A53"]
[Opening "Old Indian defense"]
[PlyCount "242"]
[Termination "adjudication"]
[TimeControl "40/110"]

1. d4 {book} Nf6 {book} 2. c4 {book} d6 {book} 3. Nc3 {book} Nbd7 {book}
4. e4 {book} e5 {book} 5. Nf3 {book} Be7 {book} 6. Be2 {+0.97/43 6.1s}
O-O {-0.66/21 2.3s} 7. O-O {+0.94/41 1.7s} a5 {-0.78/26 8.3s}
8. Qc2 {+0.98/46 5.0s} Re8 {-0.65/25 3.5s} 9. Be3 {+0.90/45 2.7s}
Ng4 {-0.75/23 2.4s} 10. Bd2 {+0.81/37 1.6s} exd4 {-0.57/24 2.1s}
11. Nxd4 {+0.87/35 0.89s} Nc5 {-0.71/25 3.0s} 12. Rad1 {+0.89/39 2.8s}
Bf6 {-0.71/23 2.1s} 13. Bc1 {+0.84/37 1.4s} Bxd4 {-0.76/26 4.1s}
14. Rxd4 {+0.77/35 1.1s} Qh4 {-0.74/27 1.9s} 15. Bxg4 {+0.71/38 1.8s}
Qxg4 {-0.75/27 3.9s} 16. f3 {+0.69/45 3.2s} Qd7 {-0.78/26 7.3s}
17. Nd5 {+0.62/41 2.7s} Na6 {-0.71/26 8.2s} 18. Qc3 {+0.94/42 4.4s}
b6 {-0.66/23 2.5s} 19. e5 {+1.02/41 3.0s} Kh8 {-0.84/24 4.0s}
20. Bf4 {+2.80/40 3.6s} Bb7 {-1.60/24 4.3s} 21. exd6 {+3.08/35 0.87s}
Bxd5 {-1.43/24 1.7s} 22. Rxd5 {+3.21/37 1.6s} Nb4 {-2.10/21 1.9s}
23. Rdd1 {+3.21/39 2.2s} cxd6 {-1.74/22 1.8s} 24. a3 {+3.55/38 2.7s}
Nc6 {-1.98/22 1.8s} 25. Rxd6 {+3.58/54 2.8s} Qf5 {-1.39/24 1.7s}
26. Bg3 {+3.37/55 2.8s} Qc5+ {-1.61/25 3.3s} 27. Kh1 {+3.37/37 1.7s}
b5 {-1.48/26 2.6s} 28. cxb5 {+3.34/40 3.7s} Qxb5 {-1.49/25 3.3s}
29. Ra1 {+3.46/41 3.0s} Ne7 {-1.18/25 3.7s} 30. a4 {+3.66/37 2.4s}
Qg5 {-1.55/26 1.9s} 31. Rd7 {+2.27/59 5.5s} Nf5 {-0.98/21 1.9s}
32. Bf2 {+2.12/42 2.6s} Re2 {-0.74/23 2.1s} 33. Qc5 {+2.28/39 2.8s}
Rae8 {-0.96/22 2.0s} 34. Rxf7 {+1.40/38 3.2s} Re1+ {-0.54/29 1.9s}
35. Rxe1 {+1.26/39 2.0s} Rxe1+ {-0.40/30 4.1s} 36. Bxe1 {+1.22/38 2.6s}
Ng3+ {-0.36/29 1.7s} 37. hxg3 {+1.20/38 2.8s} Qxc5 {-0.46/29 1.8s}
38. Rf4 {+1.13/39 3.7s} Kg8 {-0.48/31 2.5s} 39. Bc3 {+1.06/40 5.4s}
g6 {-0.43/32 4.3s} 40. Rd4 {+0.93/38 8.2s} Qh5+ {-0.43/31 1.3s}
41. Rh4 {+0.84/41 6.4s} Qd5 {-0.36/35 13s} 42. Kg1 {+0.81/43 8.6s}
Qc5+ {-0.32/32 7.9s} 43. Kh2 {+0.79/42 6.0s} Qd5 {-0.29/33 3.2s}
44. Rg4 {+0.77/43 7.5s} Qh5+ {-0.27/30 2.2s} 45. Kg1 {+0.73/38 3.1s}
Qc5+ {-0.35/33 2.8s} 46. Rd4 {+0.73/42 4.0s} Qg5 {-0.27/35 3.2s}
47. g4 {+0.69/46 5.5s} h5 {-0.25/30 1.9s} 48. gxh5 {+0.68/41 4.0s}
Qxh5 {-0.24/30 1.8s} 49. Re4 {+0.71/47 7.4s} Qg5 {-0.23/30 2.2s}
50. Kf1 {+0.67/47 6.6s} Kf7 {-0.23/29 2.1s} 51. Rc4 {+0.63/44 4.5s}
Qe3 {-0.22/29 1.9s} 52. Rg4 {+0.64/41 3.6s} Qd3+ {-0.23/31 3.1s}
53. Kf2 {+0.66/41 3.0s} Qf5 {-0.21/30 1.8s} 54. Re4 {+0.60/45 3.2s}
Qc5+ {-0.22/26 1.6s} 55. Kg3 {+0.54/45 3.5s} Qg5+ {-0.22/29 1.4s}
56. Rg4 {+0.51/43 2.2s} Qc5 {-0.22/30 2.0s} 57. Rf4+ {+0.53/40 1.8s}
Kg8 {-0.22/31 1.9s} 58. Re4 {+0.57/40 1.7s} Qg5+ {-0.21/28 1.6s}
59. Kf2 {+0.54/40 1.6s} Kf7 {-0.21/28 1.4s} 60. Rd4 {+0.51/44 2.5s}
Qc5 {-0.21/31 1.7s} 61. Kf1 {+0.53/44 2.5s} Qf5 {-0.19/29 1.5s}
62. Rh4 {+0.45/40 1.5s} Qg5 {-0.19/32 3.4s} 63. Rg4 {+0.51/42 1.8s}
Qf5 {-0.19/30 1.5s} 64. Kf2 {+0.51/39 1.2s} Qc2+ {-0.19/30 1.7s}
65. Kg3 {+0.56/40 1.3s} Qc1 {-0.19/29 1.5s} 66. Rf4+ {+0.51/38 0.61s}
Kg8 {-0.19/29 1.6s} 67. Rd4 {+0.51/42 1.3s} Qg5+ {-0.18/28 1.7s}
68. Rg4 {+0.47/42 1.3s} Qf5 {-0.14/28 1.8s} 69. Kh2 {+0.37/40 1.00s}
Kf7 {-0.13/30 2.3s} 70. Re4 {+0.40/40 1.2s} Qh5+ {-0.11/30 1.8s}
71. Kg1 {+0.37/38 0.82s} Qc5+ {-0.11/29 1.8s} 72. Rd4 {+0.36/38 0.69s}
Qg5 {-0.12/31 2.3s} 73. g4 {+0.39/40 1.4s} Qe3+ {0.00/32 1.7s}
74. Kg2 {+0.43/35 0.54s} Qe2+ {0.00/36 2.4s} 75. Kg3 {+0.39/36 0.69s}
g5 {0.00/36 1.8s} 76. Re4 {+0.31/40 1.7s} Qd1 {0.00/37 2.8s}
77. Kf2 {+0.28/38 1.3s} Qc2+ {0.00/36 2.1s} 78. Ke3 {+0.28/40 1.8s}
Qc1+ {0.00/39 2.7s} 79. Kd4 {+0.20/40 1.9s} Qd1+ {0.00/41 2.8s}
80. Kc5 {+0.11/39 1.6s} Qxf3 {0.00/38 2.7s} 81. Rc4 {+0.06/63 7.7s}
Qb7 {0.00/39 3.1s} 82. Bxa5 {+0.02/60 3.8s} Qe7+ {0.00/36 3.0s}
83. Kb5 {+0.02/63 6.6s} Qe5+ {0.00/35 3.0s} 84. Ka6 {-0.03/64 4.0s}
Qxb2 {0.00/38 2.8s} 85. Bb6 {-0.05/63 4.0s} Qe2 {0.00/39 2.7s}
86. Kb5 {-0.08/62 2.3s} Ke6 {0.00/40 2.5s} 87. Kc5 {-0.12/68 5.7s}
Qg2 {0.00/43 2.3s} 88. Rd4 {+0.04/63 5.5s} Qc2+ {0.00/47 2.5s}
89. Kb4 {0.00/69 4.9s} Qc6 {0.00/42 2.4s} 90. Bc5 {-0.01/62 4.8s}
Qb7+ {0.00/39 2.2s} 91. Kc3 {-0.06/68 3.4s} Ke5 {0.00/41 2.4s}
92. Rc4 {-0.16/61 4.5s} Qb1 {0.00/28 4.6s} 93. Ba3 {-0.15/60 3.9s}
Qe1+ {+0.32/26 2.4s} 94. Kc2 {-0.15/58 3.8s} Kd5 {+0.20/25 2.1s}
95. Kb3 {-0.34/59 3.7s} Qg3+ {+0.07/28 6.9s} 96. Kb4 {-0.46/59 3.2s}
Qb8+ {+0.14/29 1.8s} 97. Kc3 {-0.32/54 2.0s} Qe5+ {+0.09/31 2.0s}
98. Kb4 {-0.32/61 2.6s} Qe7+ {+0.04/27 1.7s} 99. Kb3 {-0.37/57 2.1s}
Qe2 {+0.04/31 2.5s} 100. Rb4 {-0.15/55 2.1s} Qe3+ {+0.51/27 6.2s}
101. Kb2 {-0.20/57 3.1s} Ke6 {+1.14/26 3.7s} 102. Rc4 {-0.21/55 1.8s}
Qd3 {+1.37/25 1.7s} 103. Rb4 {-0.37/50 1.5s} Kd7 {+1.51/28 1.9s}
104. Ka2 {-0.29/57 1.6s} Qc3 {+1.66/27 2.2s} 105. a5 {-0.32/58 1.1s}
Kc7 {+1.68/27 2.5s} 106. a6 {-0.22/50 0.99s} Qd2+ {+2.04/27 3.3s}
107. Kb1 {-0.28/56 1.6s} Qd3+ {+2.14/28 2.1s} 108. Kb2 {-0.42/58 1.6s}
Qxa6 {+1.97/28 2.9s} 109. Ka2 {-0.54/63 1.5s} Qe6+ {+2.75/23 2.3s}
110. Kb2 {-0.78/57 2.1s} Qe2+ {+2.76/25 2.2s} 111. Ka1 {-1.29/57 2.1s}
Qe5+ {+3.03/23 3.1s} 112. Kb1 {-3.04/55 1.8s} Qc3 {+3.54/23 2.1s}
113. Ka2 {-3.55/52 0.88s} Kc6 {+3.78/24 3.6s} 114. Rb3 {-3.86/57 1.6s}
Qc4 {+3.81/22 4.0s} 115. Bb2 {-3.89/54 1.2s} Qxg4 {+4.67/20 4.2s}
116. Bc1 {-4.06/54 1.3s} Qe2+ {+6.63/20 2.1s} 117. Ka3 {-4.21/48 1.0s}
Qe1 {+10.40/18 2.7s} 118. Be3 {-4.56/51 1.8s} g4 {+250.00/17 0.62s}
119. Bf4 {-4.47/49 0.81s} Qa1+ {+250.00/42 0.32s} 120. Kb4 {-4.47/50 0s}
Qd4+ {+250.00/50 0.45s} 121. Ka5 {-7.75/58 11s}
Qxf4 {+250.00/53 0.47s, Black wins by adjudication: SyzygyTB} 0-1[/pgn]

I'll be following Scorpio progress. Thank you for your efforts.

Edit: The openings for the match against Komodo were "Noomen Short Lines 2015.pgn".

Hi MMarco,

Thanks for running tests and giving feedback on the games, much appreciated!
I am glad it did so well with the GPU that you have using INT8. I think INT8 is valuable for those that
don't have RTX GPUs.
The benchmark of 8 knps is really good! I believe with a similar GPU in the past I was only able to get upto 5 knps.
I will try to modify the default formula to best reflect this.

Yes, it gobbles up endgames sometimes. For less than 10 pieces or less alpha-beta kicks in so it maybe able to save games.
The endgame net kicks in at <= 14 pieces for better endgame play, and it is also smaller so it is faster.
There are still things to work out with NN and endgames, so hopefully in the future it will be nicer.

thanks again,
Daniel

Dann Corbit · Post by **Dann Corbit** » Sun Jun 07, 2020 8:21 pm

Once you get your client/server stuff sorted out we will try it again

carldaman · Post by **carldaman** » Sun Jun 07, 2020 10:25 pm

It is reported that Lc0 fails to convert many winning endgames, but from what I observed the problem is even more acute with Scorpio and its current nets. Many promising positions were blown away (draw or even lost in some cases).

Same observation here. It was getting very frustrating to watch, so I gave up testing it for now. I mainly test engines so I can watch the games, so they'd better be worthwhile.

Help with Scorpio 3.0.8

Re: Help with Scorpio 3.0.8

Re: Help with Scorpio 3.0.8

Re: Help with Scorpio 3.0.8

Re: Help with Scorpio 3.0.8

Re: Help with Scorpio 3.0.8

Re: Help with Scorpio 3.0.8

Re: Help with Scorpio 3.0.8

Re: Help with Scorpio 3.0.8

Re: Help with Scorpio 3.0.8

Re: Help with Scorpio 3.0.8