Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Fat Fritz
- - By cghori (**) Date 2019-11-12 02:45
I just purchased Fritz 17 with Fat Fritz and when I went to use Fat Fritz from the engine menu, I see 2 versions-Fat Fritz cpu and Fat Fritz opencl.  Anyone have any idea what are the differences in these versions.  Which one should be used?

Thanks in advance.
Parent - - By Graham Banks (*****) Date 2019-11-12 11:48 Upvotes 1
Fat Fritz opencl should be used with a GPU.
Parent - - By cghori (**) Date 2019-11-13 15:13
What is the difference between the 2 versions?
Parent - - By dorszcz (**) Date 2019-11-13 16:01
I think opencl version is much faster.
Parent - - By user923005 (****) Date 2019-11-14 02:52
That is almost always true.
If you had a whomping CPU like 2 x 7742 AMD and a wimpy GPU card (or no GPU card), the CPU version could be faster.
But in general, the CPU version is intended for people who do not have a GPU that can run OpenCL.
Parent - - By user923005 (****) Date 2019-11-14 03:01
You can use GPU-Z to find out about your video card (GPU).

If you intend to use LC0 / Fat Fritz for analysis, you should get one if you don't have one.
This will show you the best "bang for the buck":
https://www.videocardbenchmark.net/gpu_value.html
Parent - - By cghori (**) Date 2019-11-14 18:58
My friend who also bought Fat Fritz has a 3rd option (which don't have): Fat Fritz cuda.

We both have Nvidia graphic cards.  Why doesn't mine have the cuda option?

Thanks in advance.
Parent - By user923005 (****) Date 2019-11-14 19:27
Cuda is better for Nvidia cards (as long as they are fairly recent).
Some of the special new cards have stupid-fast matrix multiplications (something absurd like a 4x4 matrix multiply in a single cycle).
If you have a cuda card, Fat Fritz Cuda would be better.
I don't know how to get it because I only have the standard LC0.
If you have an old Nvidia GPU, it probably won't matter much because it is the new instructions that really make it shine.
Parent - By witchesbutt (**) Date 2019-11-14 21:39
You're probably missing the cuda dev kit - go here - install then re-install the FatFritz setup and FF17 cuda/Lc0 cuda should display.
Then check that the backend shows cudnn - if it does you should be OK - if not you may have issues.
(I'm a noob - the above via a good friend - anything more I cannot help you - regrets .. :cry:)

.https://developer.nvidia.com/cuda-downloads
Parent - - By cma6 (****) Date 2019-11-16 04:14
Graham,
My impression from reading various articles about FF (an lc0 derivative) is that one needs a very powerful GPU setup, e.g., 2 X RTX 2080 Ti, to get even a fairly decent output from lc0 or FF that would compete with SF paired with a powerful CPU, e.g., Threadripper 2 or 3.
Is that a correct impression?
Parent - By user923005 (****) Date 2019-11-16 04:31
Everything is relative.
If you have a decent card like a 1660 super or a single 2080 it won't be as powerfui as SF on a top end threadripper, but the card is a lot cheaper than a top end threadripper

Also, you can simply analyze longer.  If your card is 1/8 as powerful, then analyze 8 times longer and your result is just as strong.
Parent - - By jpqy (**) Date 2019-11-16 12:15 Upvotes 1
When you check Ipmanchess website he has a fast system using a i9 7980XE 18cores/36threads + RTX 2080 !
And he can let Lc0 win from Stockfish using 18cores @4.2Ghz!
Even with these fast time controls : 1m+0s ,1m+1s ,4m+0s and 5m+1s Lc0 wins.
You hear always Lc0 needs more time but Ipman shows with tuned settings that you find back in his spreadsheet that it's possible
to win from Stockfish.
People with equal or lower cpu as his system (overclocked) and a RTX2080 should see Lc0 winning with right setup & tuning ofcourse & nets he used..
Parent - - By cma6 (****) Date 2019-11-16 19:55
If one is in effect using very long time controls, e.g., for correspondence analysis, do you think the conclusion would be equally valid that a well tuned LCO machine can edge out a powerful SF cpu machine?
Parent - - By MrKris (***) Date 2019-11-17 11:01
If you have just a very roughly equal CPU-GPU set up Stockfish and Lc0 will win and lose in completely different ways.
Lc0 is good at positional maneuvering;
Stockfish is good at tactics.

As white Lc0 out maneuvers Stockfish, 32...Rc8 would have given it some chance.
Same opening as black Lc0's 35...a5 shows it has a positional advantage.
[Event "G/3,1 Lc0 2060 - Sf 4770K"]
[Site "MB2600bin"]
[Date "2019.11.14"]
[Round "67"]
[White "Lc0-v0.22.0_49921"]
[Black "Stockfish_19111311c0"]
[Result "1-0"]
[ECO "E94"]
[Opening "King's Indian"]
[PlyCount "128"]
[Termination "adjudication"]
[TimeControl "180+1"]
[Variation "Orthodox Variation"]

1. d4 {book} Nf6 {book} 2. c4 {book} g6 {book} 3. Nc3 {book} Bg7 {book} 4.
e4 {book} d6 {book} 5. Nf3 {book} O-O {book} 6. Be2 {book} e5 {book} 7. O-O
{book} exd4 {book} 8. Nxd4 {+0.63/11 2.3s} Re8 {-0.52/26 6.6s} 9. f3
{+0.63/12 3.0s} Nc6 {-0.48/26 5.9s} 10. Be3 {+0.62/11 3.7s} Nh5 {-0.57/27
31s} 11. Qd2 {+0.59/12 7.0s} Bxd4 {-0.57/27 9.2s} 12. Bxd4 {+0.63/12 4.2s}
Nxd4 {-0.58/27 2.7s} 13. Qxd4 {+0.62/11 2.3s} Qf6 {-0.61/26 3.9s} 14. Qd2
{+0.65/10 10s} Qf4 {-0.80/25 4.8s} 15. Rfd1 {+0.63/10 4.8s} Qxd2 {-0.69/24
3.0s} 16. Rxd2 {+0.69/9 3.8s} Be6 {-0.68/25 2.2s} 17. Rc1 {+0.66/9 10s}
Rec8 {-0.32/24 3.9s} 18. Kf2 {+0.74/8 4.7s} Kg7 {-0.60/30 15s} 19. b3
{+0.76/8 3.6s} h6 {-0.50/24 5.3s} 20. Nb5 {+0.67/8 6.8s} b6 {-0.38/23 1.9s}
21. Nc3 {+0.62/8 5.1s} a5 {-0.08/23 1.9s} 22. h4 {+0.55/7 5.9s} Nf6
{-0.13/24 3.5s} 23. Nd5 {+0.48/7 5.2s} h5 {0.00/28 1.8s} 24. Rc3 {+0.49/7
4.1s} Nd7 {-0.17/30 9.4s} 25. Re3 {+0.46/7 4.1s} Rab8 {-0.19/26 6.1s} 26.
Nc3 {+0.41/8 6.8s} Re8 {-0.18/25 5.2s} 27. Nb5 {+0.36/8 5.5s} Rbc8
{-0.18/24 1.5s} 28. Na7 {+0.34/8 4.5s} Rcd8 {-0.23/29 9.5s} 29. Nc6
{+0.31/8 5.9s} Ra8 {-0.27/23 1.5s} 30. Nd4 {+0.34/8 2.7s} f6 {-0.30/27
7.7s} 31. f4 {+0.45/9 5.2s} Re7 {-0.19/24 3.4s} 32. Bf3 {+0.64/10 4.0s} Rf8
{0.00/22 1.9s ??} (32. .. Rc8 {! ---- Can keep the white king from invading.}
33. Nf5+ Bxf5 34. exf5 Rxe3 35. Kxe3 gxf5 36. Kd4 Nc5 37. Kd5 c6+ 38. Kxd6
{??} Rd8+) 33. Nf5+ {+1.49/15 3.9s} Bxf5 {-0.75/32 6.1s} 34. exf5 {+1.40/18
2.4s} Rxe3 {-0.93/30 4.6s} 35. Kxe3 {+1.43/18 1.6s} Nc5 {-0.81/29 2.0s}
(35. .. gxf5 36. Kd4 Nc5 37. Kd5 {---- the king invades} Rf7 38. Bxh5 Re7
39. Re2) 36. Kd4 {+1.49/18 6.7s} gxf5 {-0.92/30 3.4s} 37. Kd5 {+1.48/23
0.23s} Kg6 {-1.94/34 23s} 38. Re2 {+1.55/17 5.0s} Nd3 {-1.59/32 1.8s} 39.
Re7 {+1.58/22 4.0s} Nb4+ {-1.08/24 1.2s} 40. Ke6 {+1.51/25 2.4s} Nxa2
{-0.94/28 1.3s} 41. Rxc7 {+1.48/24 6.5s} Nc1 {-1.75/28 4.5s} 42. Bd1
{+1.47/24 0.84s} Nd3 {-1.64/25 1.1s} 43. Kxd6 {+1.46/23 2.3s} Rd8+
{-1.64/26 4.9s} 44. Rd7 {+1.39/17 11s} Rc8 {-1.62/27 1.1s} 45. Rc7
{+1.21/20 18s} Rd8+ {-1.67/26 1.5s} 46. Rd7 {+1.20/22 0.22s} Rc8 {-1.63/27
1.2s} 47. Rb7 {+1.21/20 0.36s} a4 {-1.67/29 1.8s} 48. bxa4 {+1.19/19 0.49s}
Rxc4 {-1.56/29 1.1s} 49. Rxb6 {+1.18/17 0.60s} Nxf4 {-1.55/29 2.5s} 50. Bb3
{+2.02/15 7.5s} Rc8 {-1.47/21 0.89s} 51. a5 {+1.91/18 3.7s} Nxg2 {-1.88/25
3.1s} 52. a6 {+1.91/19 0.24s} Ne3 {-2.61/27 4.8s} 53. Kd7 {+1.89/18 3.7s}
Ra8 {-2.33/23 2.2s} 54. Ba2 {+1.92/16 0.48s} Ng2 {-2.59/26 1.6s} 55. Kc7
{+1.64/18 15s} Nxh4 {-2.84/27 1.9s} 56. Bc4 {+1.64/17 0.23s} f4 {-1.96/22
0.77s} 57. Rb8 {+1.64/18 0.24s} Ra7+ {-3.77/29 6.4s} 58. Kb6 {+1.64/15
0.23s} Rd7 {-4.41/28 1.5s} 59. Bb5 {+1.50/14 7.2s} Re7 {-6.07/29 3.3s} 60.
Rb7 {+2.22/11 1.2s} Re6+ {-4.69/23 1.1s} 61. Ka5 {+2.28/11 1.3s} Rxa6+
{-3.41/19 0.50s} 62. Bxa6 {+2.33/11 0.81s} Nf3 {-4.41/24 1.8s} 63. Kb4
{+2.62/10 1.7s} Ne5 {-6.64/25 1.5s} 64. Kc3 {+3.33/8 1.9s} Kf5 {-8.49/25
1.0s, White wins by adjudication} 1-0

[Event "G/3,1 Lc0-Sf"]
[Site "MB2600bin"]
[Date "2019.11.14"]
[Round "67"]
[White "Stockfish_19111311c0"]
[Black "Lc0-v0.22.0_49921"]
[Result "0-1"]
[ECO "E94"]
[Opening "King's Indian"]
[PlyCount "157"]
[Termination "adjudication"]
[TimeControl "180+1"]
[Variation "Orthodox Variation"]

1. d4 {book} Nf6 {book} 2. c4 {book} g6 {book} 3. Nc3 {book} Bg7 {book} 4.
e4 {book} d6 {book} 5. Nf3 {book} O-O {book} 6. Be2 {book} e5 {book} 7. O-O
{book} exd4 {book} 8. Nxd4 {+0.82/28 8.0s} Nc6 {-0.58/11 5.0s} 9. Be3
{+0.82/27 2.5s} Nxd4 {-0.59/11 3.7s} 10. Bxd4 {+0.77/27 5.8s} Re8 {-0.57/10
2.0s} 11. f3 {+0.96/24 3.1s} b6 {-0.58/9 6.2s} 12. Qd2 {+0.79/25 6.4s} Bb7
{-0.58/9 3.9s} 13. Rae1 {+0.89/24 11s} a6 {-0.55/8 9.0s} 14. a3 {+0.92/22
2.5s} Rb8 {-0.53/8 6.7s} 15. Rc1 {+1.00/25 8.1s} h5 {-0.50/7 4.3s} 16. Bd1
{+1.01/21 2.3s} h4 {-0.33/8 3.2s} 17. Be2 {+0.85/26 15s} Nh5 {-0.46/8 7.8s}
18. Be3 {+1.09/20 1.5s} h3 {-0.50/9 5.2s} 19. g4 {+0.96/27 16s} Nf6
{-0.44/10 2.3s} 20. Bg5 {+0.84/23 2.6s} Qc8 {-0.50/10 6.9s} 21. Nd5
{+1.07/23 5.2s} Nh7 {-0.39/14 3.0s} 22. Bh6 {+1.21/25 6.1s} Be5 {-0.22/9
5.1s} 23. Be3 {+0.90/26 26s} c5 {-0.32/11 8.2s} 24. g5 {+0.99/24 5.4s} b5
{-0.30/11 3.5s} 25. b4 {+0.90/26 4.8s} Bxd5 {-0.42/14 2.6s} 26. cxd5
{+1.19/24 1.1s} c4 {-0.19/12 3.9s} 27. Kh1 {+1.09/24 2.7s} Bg7 {-0.12/10
5.4s} 28. Rg1 {+0.98/26 4.6s} Ra8 {-0.05/12 4.6s} 29. a4 {+1.15/23 1.5s} c3
{-0.12/14 7.0s} 30. Qc2 {+1.04/24 3.6s} bxa4 {-0.09/13 1.4s} 31. Qxa4
{+0.95/28 14s} Qb7 {-0.08/13 6.8s} 32. Bd3 {+0.79/27 9.7s} Nf8 {+0.07/12
4.4s} 33. Rg3 {+0.69/26 1.9s} Reb8 {+0.22/14 5.5s} 34. Rb1 {+0.16/26 7.8s}
Qd7 {+0.26/15 2.1s} 35. Qa3 {+1.09/24 6.4s} a5 {+0.66/16 7.1s} 36. b5
{-0.30/25 4.3s} (36. bxa5 {??} Rxb1+ 37. Bxb1 Qb5 38. Bc2 Rxa5 39. Qc1 Qe2
40. Bd1 Qa6 41. Rg1 Ra1 42. Qc2 Ra2 43. Qc1 c2 44. Bxc2 Qe2 {+-}) 36. .. c2
{+0.77/19 2.8s} 37. Rc1 {-0.46/28 9.1s} Rc8 {+0.78/17 0.24s} 38. Bd2
{-0.45/25 1.6s} Qa7 {+0.95/17 6.9s} 39. f4 {-0.09/24 2.6s} Nd7 {+0.84/18
4.6s} 40. Be3 {-0.61/24 4.0s} Qb7 {+0.82/17 0.22s} 41. Bd2 {-0.55/25 3.1s}
Nb6 {+0.86/16 18s} 42. Qa2 {0.00/20 0.91s} a4 {+0.85/17 0.22s} 43. Rxc2
{-1.28/25 6.5s} Rxc2 {+1.44/15 5.9s} 44. Qxc2 {-1.00/25 1.1s} a3 {+1.61/14
4.6s} 45. Qa2 {-1.90/26 1.8s} Qc8 {+1.87/13 6.4s} 46. f5 {-1.90/22 1.1s}
Bb2 {+1.92/13 1.3s} 47. Rxh3 {-1.90/21 0.63s} Nd7 {+1.93/13 3.0s} 48. Kg2
{-2.62/23 3.3s} Qc5 {+1.96/14 4.5s} 49. Be3 {-2.69/23 0.63s} Qb4 {+2.10/13
4.5s} 50. Qc4 {-1.96/22 0.69s} Qxc4 {+1.94/17 3.9s} 51. Bxc4 {-1.76/21
0.49s} a2 {+1.91/18 2.1s} 52. Bxa2 {-1.81/23 0.83s} Rxa2 {+1.86/18 1.8s}
53. Rf3 {-1.62/22 1.0s} Be5+ {+1.83/17 2.4s} 54. Rf2 {-1.47/24 1.2s} Ra3
{+1.81/16 3.5s} 55. Re2 {-1.33/28 2.6s} Rb3 {+1.78/16 2.8s} 56. h4
{-1.22/25 0.33s} Rxb5 {+1.76/15 2.2s} 57. Ra2 {-1.77/32 1.7s} Rb2+
{+1.76/14 2.0s} 58. Rxb2 {-1.95/35 0.84s} Bxb2 {+1.75/13 0.93s} 59. Kf3
{-1.94/33 0.41s} Ne5+ {+1.77/10 7.6s} 60. Kf2 {-1.95/34 1.1s} Nc4 {+1.76/9
3.5s} 61. Ke2 {-1.95/34 0.54s} Be5 {+1.92/8 1.6s} 62. Bf2 {-1.95/36 0.49s}
Bh2 {+1.84/9 2.4s} 63. Kf3 {-1.95/36 0.56s} Ne5+ {+1.81/8 1.7s} 64. Kg2
{-1.95/37 0.58s} Bf4 {+1.87/9 0.84s} 65. Bd4 {-1.96/38 0.69s} Nc4 {+1.81/9
1.1s} 66. Kf3 {-1.95/37 0.53s} Be5 {+1.74/9 1.1s} 67. Bg1 {-1.95/38 0.67s}
Nd2+ {+1.74/8 2.5s} 68. Ke3 {-1.95/40 2.1s} Nb1 {+1.72/9 0.97s} 69. Kd3
{-1.89/31 0.54s} Na3 {+1.67/10 1.4s} 70. Bd4 {-2.01/32 1.7s} Bg3 {+2.08/9
0.59s} 71. Bf6 {-2.75/31 2.9s} Bxh4 {+2.60/8 1.0s} 72. Bb2 {-3.31/27 1.8s}
Nb5 {+3.26/8 0.91s} 73. Kc4 {-3.55/29 0.80s} Nc7 {+3.37/9 0.91s} 74. Bc1
{-3.67/32 1.2s} Bg3 {+4.74/7 1.3s} 75. Bd2 {-3.93/32 0.72s} Kf8 {+5.85/7
1.4s} 76. fxg6 {-4.35/33 1.3s} fxg6 {+7.68/7 0.81s} 77. Kd3 {-5.08/30 1.0s}
Na6 {+12.79/6 2.1s} 78. Ke2 {-5.69/30 1.0s} Be5 {+21.07/6 1.0s} 79. Ba5
{-6.51/29 1.0s, Black wins by adjudication} 0-1


Lc0 as black gets tactically refuted beginning with 25...Rad8? (The reversed was a draw.)
Stockfish as black keeps its pawn from the book opening and goes on to win. (The reversed was a draw.)
[Event "G/3,1 Lc0-Sf"]
[Site "MB2600bin"]
[Date "2019.11.14"]
[Round "60"]
[White "Stockfish_19111311c0"]
[Black "Lc0-v0.22.0_49921"]
[Result "1-0"]
[ECO "E20"]
[Opening "Nimzo-Indian"]
[PlyCount "80"]
[Termination "adjudication"]
[TimeControl "180+1"]
[Variation "Kmoch Variation"]

1. d4 {book} Nf6 {book} 2. c4 {book} e6 {book} 3. Nc3 {book} Bb4 {book} 4.
f3 {book} O-O {book} 5. a3 {book} Bxc3+ {book} 6. bxc3 {book} d5 {book} 7.
cxd5 {book} exd5 {book} 8. e3 {+0.22/24 6.2s} c5 {-0.17/10 3.4s} 9. Bd3
{+0.12/25 3.2s} b6 {-0.22/13 3.3s} 10. Ne2 {+0.14/24 2.3s} Ba6 {-0.31/15
6.0s} 11. O-O {+0.35/27 12s} Bxd3 {-0.34/16 5.0s} 12. Qxd3 {+0.45/25 2.1s}
Re8 {-0.34/15 1.4s} 13. Ng3 {+0.38/28 18s} Nc6 {-0.37/14 1.3s} 14. Bb2
{+0.55/24 3.2s} h5 {-0.19/17 11s} 15. Rae1 {+0.69/23 2.6s} h4 {-0.18/17
0.26s} 16. Nf5 {+0.79/25 2.9s} h3 {-0.17/15 1.6s} 17. e4 {+0.85/26 3.8s}
Ne7 {-0.19/19 3.5s} 18. Ng3 {+0.69/26 4.4s} Ng6 {-0.19/20 0.21s} 19. Bc1
{+0.80/24 1.9s} hxg2 {-0.22/17 22s} 20. Kxg2 {+0.15/25 8.5s} dxe4 {-0.21/17
4.2s} 21. fxe4 {+0.17/24 3.5s} Qd7 {-0.20/17 6.0s} 22. Kh1 {+0.68/26 8.0s}
Qh3 {+0.03/11 4.8s} 23. Bg5 {+0.16/26 12s} Ng4 {+0.11/14 4.6s} 24. Qe2
{+0.26/24 1.4s} f6 {+0.08/15 7.8s} 25. Bc1 {+0.23/23 2.9s} Rad8 {+0.16/11
5.3s ?} (25. .. cxd4 {---- Sf 0.00} 26. cxd4 (26. Rf3 Nh4 27. Rf4 d3 28.
Qa2+ Kh8 29. Rd1 Rad8 30. Qd2 Rc8 31. Rg1 Rxc3 {+-}) 26. .. Rac8 27. Rf3
Nh4 28. Rd3 f5 29. e5 Rf8 30. e6 f4 31. e7 f3 32. exf8=Q+ Rxf8 33. Qa2+ Kh8
34. Rxf3 (34. Nf1 f2 35. Rxh3 fxe1=Q 36. Rxh4+ Qxh4 {+-}) 34. .. Nxf3 {+-})
26. Rf3 {+0.98/24 5.6s} cxd4 {+0.68/11 3.5s ?} (26. .. Nh4 {?} 27. Rf4)
(26. .. Rd5 {!} 27. Nf1 {! +/}) 27. Nf5 {+1.63/24 1.5s} Qh5 {-0.88/11 3.4s
??} (27. .. Qxh2+ 28. Qxh2 Nxh2 29. Kxh2 {+/}) 28. Rg3 {+2.86/27 2.0s} Rxe4
{-2.55/13 9.6s} 29. Qa2+ {+4.19/26 3.3s} Kf8 {-2.67/12 2.1s} 30. Rxe4
{+4.40/26 2.2s} Qxf5 {-2.79/11 2.4s} 31. Rexg4 {+4.46/27 2.8s} dxc3
{-2.91/10 4.2s} 32. Rg1 {+4.54/26 2.9s} c2 {-3.22/9 8.3s} 33. Qb3 {+5.01/22
2.7s} Rd1 {-4.19/9 9.7s} 34. Rxd1 {+5.04/22 2.9s} Qxg4 {-5.14/8 5.0s} 35.
Qxc2 {+5.24/24 4.8s} Qf3+ {-5.76/8 4.8s} 36. Kg1 {+5.58/25 2.6s} Qg4+
{-5.25/8 2.9s} 37. Kf2 {+5.61/25 2.1s} Ne5 {-5.55/8 8.1s} 38. Rd8+
{+5.81/24 2.3s} Ke7 {-4.19/9 4.3s} 39. Qc7+ {+6.28/22 3.7s} Nd7 {-5.52/9
5.2s} 40. Rc8 {+6.41/21 2.2s} g5 {-6.10/7 4.9s, White wins by adjudication}
1-0

[Event "G/3,1 Lc0-Sf"]
[Site "MB2600bin"]
[Date "2019.11.14"]
[Round "75"]
[White "Lc0-v0.22.0_49921"]
[Black "Stockfish_19111311c0"]
[Result "0-1"]
[ECO "D44"]
[Opening "Semi-Slav"]
[Variation "Botvinnik, 6.a4"]
[TimeControl "180+1"]
[Termination "adjudication"]
[PlyCount "163"]
[WhiteType "human"]
[BlackType "human"]

1. d4 {book} d5 {book} 2. c4 {book} c6 {book} 3. Nc3 {book} Nf6 {book} 4.
Nf3 {book} dxc4 {book} 5. a4 {book} e6 {book} 6. Bg5 {book} Bb4 {book} 7.
e3 {book} b5 {book} 8. Be2 {-0.47/11 3.0s} Nbd7 {+0.66/22 5.2s} 9. O-O
{-0.44/14 2.3s} Qb6 {+0.64/22 3.0s} 10. Qc2 {-0.46/14 3.3s} O-O {+0.69/21
2.6s} 11. e4 {-0.32/14 5.1s} a6 {+0.94/25 9.0s} 12. Be3 {-0.29/17 5.4s} Bb7
{+0.69/27 4.4s} 13. e5 {-0.32/18 2.5s} Nd5 {+0.76/28 2.9s} 14. Ng5
{-0.36/15 2.8s} g6 {+0.68/28 5.1s} 15. Nxd5 {-0.39/15 3.2s} cxd5 {+1.13/22
2.1s} 16. h4 {-0.42/15 3.7s} bxa4 {+1.09/26 5.8s} 17. Qxa4 {-0.43/17 7.1s}
Bc6 {+1.09/24 2.4s} 18. Qd1 {-0.44/17 3.6s} Be7 {+1.12/24 4.1s} 19. Qd2
{-0.41/12 12s} a5 {+1.47/26 6.8s} 20. Rfb1 {-0.49/12 5.9s} Rfb8 {+1.23/30
16s} 21. Bd1 {-0.52/13 5.6s} Qc7 {+1.17/30 7.9s} 22. h5 {-0.34/11 11s} a4
{+1.34/24 1.4s} 23. Bc2 {-0.37/11 2.1s} a3 {+1.82/25 2.2s} 24. bxa3
{-0.32/13 4.3s} Rxb1+ {+1.68/28 6.5s} 25. Rxb1 {-0.32/12 2.4s} Rxa3
{+1.54/28 4.4s} 26. g3 {-0.45/12 6.5s} Qa7 {+1.78/23 2.3s} 27. Qc1
{-0.92/13 8.2s} Ba4 {+1.93/25 2.5s} 28. Bxa4 {-0.91/15 2.4s} Qxa4 {+1.86/28
4.2s} 29. Nf3 {-0.91/15 4.2s} c3 {+1.85/28 2.8s} 30. Ne1 {-0.97/12 3.5s}
Kf8 {+1.80/28 4.1s} 31. Kg2 {-1.01/12 7.2s} Ke8 {+1.76/27 4.7s} 32. h6
{-0.98/13 2.7s} Qc4 {+1.59/29 5.4s} 33. Nf3 {-0.96/13 7.0s} Ra8 {+0.91/34
33s} 34. Ra1 {-0.48/12 5.7s} Rc8 {+0.91/30 2.4s} 35. Qc2 {-0.31/11 5.3s}
Nb6 {+0.97/28 2.7s} 36. Bc1 {-0.28/14 5.1s} Bd8 {+0.92/29 2.7s} 37. Ra5
{-0.13/10 6.5s} Qb4 {+0.79/30 4.7s} 38. Ra3 {-0.13/10 6.0s} Qc4 {+0.27/31
14s} 39. Ra5 {-0.11/10 2.9s} Qb4 {+0.33/30 0.93s} 40. Ra3 {-0.08/10 4.6s}
Rc4 {+0.33/30 1.4s} 41. Ra7 {-0.11/9 9.3s} Nc8 {+0.75/25 1.6s} 42. Ra3
{-0.07/10 3.0s} Be7 {+0.75/29 8.9s} 43. Ra2 {-0.09/10 2.7s} Kf8 {+0.95/26
3.3s} 44. Ra1 {-0.12/10 5.8s} Kg8 {+1.34/20 0.95s} 45. Rb1 {-0.10/14 3.1s}
Qa5 {+1.15/24 3.3s} 46. Rb7 {-0.22/11 6.5s} Qa6 {+1.35/26 4.2s} 47. Qb3
{-0.10/13 2.8s} Bf8 {+1.25/28 2.9s} 48. Rb8 {-0.22/12 3.3s} c2 {+1.25/25
1.4s} 49. Qb2 {-0.11/12 3.0s} Rc7 {+1.34/27 1.8s} 50. Nh2 {-0.17/13 2.7s}
Qd3 {0.00/30 10s} 51. Qa1 {-0.21/14 3.5s} Rc4 {+0.46/24 1.3s} 52. Ra8
{-0.26/12 5.4s} Qd1 {+0.80/28 2.8s} 53. Qb2 {-0.17/11 2.7s} Rc7 {+0.93/30
1.1s} 54. Qb8 {-0.10/12 2.1s} Rc3 {+1.53/25 0.83s} 55. Qb2 {-0.11/14 1.7s}
Qd3 {+0.89/33 5.4s} 56. Ng4 {+0.07/14 2.5s} Qf3+ {+1.20/29 0.46s} 57. Kh2
{+0.03/10 0.29s} g5 {+0.79/32 1.9s} 58. Ra1 {-0.87/12 5.3s} f5 {+1.90/23
0.74s} 59. Nf6+ {-1.01/16 2.9s} Kf7 {+2.64/25 0.55s} 60. Be3 {-0.97/18
0.23s} Rxe3 {+2.83/25 0.80s} 61. fxe3 {-1.01/16 0.22s} Qe2+ {+3.04/28 3.3s}
62. Kh1 {-0.98/15 1.1s} Qf3+ {+3.09/27 1.3s} 63. Kh2 {-0.97/16 2.4s} Qe2+
{+3.05/29 1.6s} 64. Kh1 {-1.01/18 0.21s} Qd1+ {+3.07/26 0.68s} 65. Kg2
{-1.62/19 5.2s} g4 {+3.10/27 1.3s} 66. Qb5 {-1.63/19 0.64s} Qd2+ {+7.08/26
0.33s} 67. Kg1 {-1.88/20 1.8s} Qxe3+ {+8.68/29 2.7s} 68. Kg2 {-2.04/16
1.1s} Qd2+ {+8.71/28 0.32s} 69. Kh1 {-2.07/16 1.2s} c1=R+ {+9.08/28 1.4s}
70. Rxc1 {-2.16/15 0.95s} Qxh6+ {+9.46/29 1.9s} 71. Kg2 {-2.29/16 1.1s}
Qd2+ {+9.66/24 0.38s} 72. Kf1 {-2.58/15 2.4s} Qxc1+ {+9.75/25 0.52s} 73.
Kg2 {-2.71/14 0.74s} Nd6 {+10.05/24 0.51s} 74. Qd7+ {-2.50/13 0.74s} Be7
{+12.10/25 0.72s} 75. exd6 {-2.58/13 0.58s} Qd2+ {+16.39/24 0.63s} 76. Kf1
{-2.98/13 1.9s} Qd1+ {+25.58/25 2.0s} 77. Kg2 {-3.27/14 1.00s} Qf3+
{+29.74/24 0.35s} 78. Kh2 {-3.60/14 0.59s} Qf2+ {+59.14/30 0.47s} 79. Kh1
{-4.07/14 0.030s} Qf1+ {+66.25/33 1.4s} 80. Kh2 {-4.84/12 0.037s} Qh3+
{+77.16/35 2.6s} 81. Kg1 {-5.76/10 0.036s} Qxg3+ {+147.92/35 1.2s} 82. Kh1
{-8.01/9 5.1s, Black wins by adjudication} 0-1
Parent - - By cma6 (****) Date 2019-11-17 17:28
Mr. Kris:
Thanks for the very interesting examples with annotations. Which engine is doing the evals. in each example?
It seems that all these tests are at very fast time controls. I wonder if anything in the balance between the two programs would change at 10 minutes/move, a typical deep analysis parameter.
Parent - By MrKris (***) Date 2019-11-19 22:08
Thanks cma6:

I used for annotations Lc0 2 threads, Stockfish 6 threads simultaneously in Arena (4 core/8 thread i7).

I should have detailed that more, however I just wanted to give examples of tactical-Stockfish vs. Lc0-positional.

Yes, the TC was fast - my impression is that the general tactical-positional situation 'scales' to higher.
Perhaps I should read more TCEC and CCC games on this point however.

I will try to write clearer in the future.
I left out details in this case (how long, which engine etc.) because I think my words to the games are correct: the errors major enough to be verifiable.

Here: http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=33235
on my equipment i7-4770K vs. RTX 2060 Lco should, very roughly 1.7-1.8 Leela Ratio, have an advantage over Stockfish.

G/30"+.5", Stockfish was +28 Elo (low # of games);
G/3'+1", Lc0 was +6 Elo, seems very roughly (low # of games) the minimum for Lc0, on my equipment

Longer TC's I believe would increase for Lc0 to whatever the 1.8 L.R. allows (but somewhat slowly as with all computer chess the exponential curve flattening problem).

My point in answering on the Fat Fritz thread is it seems to me (though I've had my first NN, Lc0, running only a few months now) that:
the NN vs. AB differences seem to me greater than those in an AB vs. another AB engine.

I believe, sorry I do not have extensive data:
if Stockfish loses to an NN it will often (unknown how often) not be able to correct its error in say for example a blitz game loss attempting to correct at 10 minutes/move analysis;
then the reverse: if an NN loses to Stockfish etc.

- - - -

This is better than my original, which was unclear and had errors.
Lc0 seemed 'over its head' in its queen & 2 knights attack tactically.
25...Rad8 ? definitely an advantage to white,
27...Qh5 ?? lost,
25...cxd4, from Stockfish after the game, had a good chance to draw.

(At the bottom here Lc0_49921 still had 25...Rad8? at 3 minutes, Stockfish found 25...cxd4 at 17 seconds,
at the very bottom Lc0 with a Leelenstein net quickly found 25...cxd4).

[Event "G/3,1 Lc0-Sf"]
[Site "MB2600bin"]
[Date "2019.11.14"]
[Round "60"]
[White "Stockfish_19111311c0"]
[Black "Lc0-v0.22.0_49921"]
[Result "1-0"]
[BlackElo "2400"]
[ECO "E27"]
[Opening "Nimzo-Indian"]
[Variation "Sämisch, 5...O-O 6.f3 d5 7.cxd5"]
[WhiteElo "2400"]
[TimeControl "180+1"]
[Termination "normal"]
[PlyCount "80"]
[WhiteType "human"]
[BlackType "human"]

1. d4 {book} Nf6 {book} 2. c4 {book} e6 {book} 3. Nc3 {book} Bb4 {book} 4.
f3 {book} O-O {book} 5. a3 {book} Bxc3+ {book} 6. bxc3 {book} d5 {book} 7.
cxd5 {book} exd5 {book} 8. e3 {+0.22/24 6.2s} c5 {-0.17/10 3.4s} 9. Bd3
{+0.12/25 3.2s} b6 {-0.22/13 3.3s} 10. Ne2 {+0.14/24 2.3s} Ba6 {-0.31/15
6.0s} 11. O-O {+0.35/27 12s} Bxd3 {-0.34/16 5.0s} 12. Qxd3 {+0.45/25 2.1s}
Re8 {-0.34/15 1.4s} 13. Ng3 {+0.38/28 18s} Nc6 {-0.37/14 1.3s} 14. Bb2
{+0.55/24 3.2s} h5 {-0.19/17 11s} 15. Rae1 {+0.69/23 2.6s} h4 {-0.18/17
0.26s} 16. Nf5 {+0.79/25 2.9s} h3 {-0.17/15 1.6s} 17. e4 {+0.85/26 3.8s}
Ne7 {-0.19/19 3.5s} 18. Ng3 {+0.69/26 4.4s} Ng6 {-0.19/20 0.21s} 19. Bc1
{+0.80/24 1.9s} hxg2 {-0.22/17 22s} 20. Kxg2 {+0.15/25 8.5s} dxe4 {-0.21/17
4.2s} 21. fxe4 {+0.17/24 3.5s} Qd7 {-0.20/17 6.0s} 22. Kh1 {+0.68/26 8.0s}
Qh3 {+0.03/11 4.8s} 23. Bg5 {+0.16/26 12s} Ng4 {+0.11/14 4.6s} 24. Qe2
{+0.26/24 1.4s} f6 {+0.08/15 7.8s} 25. Bc1 {+0.23/23 2.9s} Rad8 {+0.16/11
5.3s ? +/} (25. .. cxd4 {---- Sf 0.00 =} 26. cxd4 (26. Rf3 {??} Nh4 27. Rf4
d3 28. Qa2+ Kh8 29. Rd1 Rad8 30. Qd2 Rc8 31. Rg1 Rxc3 {-+}) 26. .. Rac8 27.
Rf3 Nh4 28. Rd3 f5 29. e5 Rf8 30. Rf1 (30. e6 {??} f4 31. e7 f3 32. exf8=Q+
Rxf8 33. Qa2+ Kh8 34. Rxf3 (34. Nf1 f2 35. Rxh3 fxe1=Q 36. Rxh4+ Qxh4 {-+})
34. .. Nxf3 {-+}) 30. .. f4 31. Bxf4 Ng6 32. Nf5 Qh5 33. Rg3 Qxf5 34. Rxg4
{11 minutes = : Sf 0.00 34...Qe6 ; Lc0 0.00 34...Rc3 ----}) 26. Rf3
{+0.98/24 5.6s} cxd4 {+0.68/11 3.5s ?!} (26. .. Nh4 {?} 27. Rf4 {+-}) (26.
.. Rd5 {!} 27. Nf1 {! +/}) 27. Nf5 {+1.63/24 1.5s} Qh5 {-0.88/11 3.4s ??}
(27. .. Qxh2+ 28. Qxh2 Nxh2 29. Kxh2 {+/}) 28. Rg3 {+2.86/27 2.0s} Rxe4
{-2.55/13 9.6s} 29. Qa2+ {+4.19/26 3.3s} Kf8 {-2.67/12 2.1s} 30. Rxe4
{+4.40/26 2.2s} Qxf5 {-2.79/11 2.4s} 31. Rexg4 {+4.46/27 2.8s} dxc3
{-2.91/10 4.2s} 32. Rg1 {+4.54/26 2.9s} c2 {-3.22/9 8.3s} 33. Qb3 {+5.01/22
2.7s} Rd1 {-4.19/9 9.7s} 34. Rxd1 {+5.04/22 2.9s} Qxg4 {-5.14/8 5.0s} 35.
Qxc2 {+5.24/24 4.8s} Qf3+ {-5.76/8 4.8s} 36. Kg1 {+5.58/25 2.6s} Qg4+
{-5.25/8 2.9s} 37. Kf2 {+5.61/25 2.1s} Ne5 {-5.55/8 8.1s} 38. Rd8+
{+5.81/24 2.3s} Ke7 {-4.19/9 4.3s} 39. Qc7+ {+6.28/22 3.7s} Nd7 {-5.52/9
5.2s} 40. Rc8 {+6.41/21 2.2s} g5 {-6.10/7 4.9s, White wins by adjudication}
1-0


- - - -

After the game, 3 minutes the base time of the game, although not the 10 minutes as cma6 mentioned:
Lc0 could not find 25...cxd4, Stockfish found it in 17 seconds:
Lc0-0.22_49921: * has the correct 25...cxd4 for only a fraction of a second
- the rest of the 3 minutes with 25...Rad8 ??
 6/16  00:00   998  4k  +0.53  25. ... Rad8 26.Rd1 Kf7 27.Qa2+ Re6 28.e5 Rh8 29.Ne4 Ke7
 7/16  00:00   1k  4k  +0.53  25. ... Rad8 26.Rd1 Kf7 27.Qa2+ Re6 28.e5 Rh8 29.Ne4 Ke7
 7/16  00:00   1k  4k  +0.83 *25. ... cxd4 26.cxd4 Rac8 27.Rf3 Nh4 28.Rf4 f5 29.Nxf5 Nxf5 30.Rxf5 Rxe4 31.Qxe4 Qxh2+
 7/17  00:00   2k  4k  +0.66 *25. ... cxd4 26.cxd4 Rac8 27.Qa2+ Kh7 28.Qg2 Qxg2+ 29.Kxg2 Rc2+ 30.Re2 Rxe2+ 31.Nxe2 Rxe4 32.Kf3 f5 33.Rh1 Nh4+
 7/17  00:00   3k  5k  +0.52  25. ... Rad8 26.Be3 Rc8 27.Bg1 cxd4 28.cxd4 Rc3 29.Rf3 Rec8 30.Qf1 Rxf3 31.Qxh3 Nf2+
...
 19/45  03:07   2,015k  11k  +0.02  25. ... Rad8 26.Rf5 Re6 27.Rd1 Rde8 28.Rdf1 cxd4 29.cxd4 Rxe4 30.Nxe4 Rxe4 31.Qa2+ Kh7 32.R5f3 Qh5 33.Rf5 Qh3 34.R5f3 Qh5 35.Rf5 Qh3

Stockfish_19111201c0:
- finds the correct 25...cxd4 at 17 seconds and stays with it for the 3 minutes
 26/45  00:14   25,344k  1,760k  -1.11  25. ... Rad8 26.Rf3 Rd5 27.Nf1 Qh4 28.Kg1 Rd7 29.Ng3 cxd4 30.Nf5 Qxh2+ 31.Qxh2 Nxh2 32.Kxh2 d3 33.Rd1 Rxe4 34.Rfxd3 Re2+ 35.Kg3 Rxd3+ 36.Rxd3 Ne5 37.Rd8+ Kf7 38.Nd4 Ra2 39.Rc8 Kg6 40.Bf4 Rxa3 41.Rc7 Nd3 42.Ne6 Ra2
 27/44+  00:17   30,301k  1,768k  -1.00  25. ... cxd4
 ...
 41/47  03:07   335,168k  1,791k   0.00  25. ... cxd4 26.cxd4 Rac8 27.Rf3 Nh4 28.Rf4 f5 29.Rg1 g5 30.Rff1 fxe4 31.Bxg5 Nxh2 32.Qa2+ Re6 33.Qxh2 Qxh2+ 34.Kxh2 Nf3+ 35.Rxf3 exf3 36.Rf1 Rc2+ 37.Kh3 f2 38.Kg2 Re1 39.Rxf2 Rxf2+ 40.Kxf2 Ra1 41.Nf5 Rxa3 42.d5 Kf7 43.Ke2 Ke8 44.Nd6+ Kd7 45.Ne4 Ra5 46.Ke3 Rxd5 47.Nf6+ Kc6 48.Nxd5

- - - -

Leelenstein net from:  https://www.patreon.com/jjosh/posts {found 25...cxd4 quickly}
free posts, Sept 16, 2019, net 20x256SE-jj-9-swa-27500000.pb.gz
In my post: http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=33237

Lc0-v0.22.0_jj2750:
 1/2  00:00   3  120  -0.41  25. ... Rad8 26.Nf5
...
 5/10  00:00   168  1k  +0.65  25. ... cxd4 26.cxd4 Rac8 27.Nf5 Rxe4 28.Qg2 Rxe1 29.Rxe1
...
 9/23  00:01   7k  6k  +0.38  25. ... cxd4 26.cxd4 Rac8 27.Rf3 Nh4 28.Rf4 f5 29.Nxf5 Nxf5 30.Rxg4 Nxd4 31.Qa2+ Ne6 32.Rg3 Qh4 33.Bg5 Qh5 34.Reg1
...
 11/30  00:04   31k  7k  +0.20  25. ... Rad8 26.Rf5 Re6 27.Rg1 cxd4 28.Rh5 Nh4 29.Nf5 d3 30.Qxg4 Qxg4 31.Nh6+ gxh6 32.Rxg4+ Kf7 33.Rhxh4 d2 34.Bxd2 Rxd2 35.Rxh6 Rc6 36.e5 Rxc3 37.Rxf6+ Ke8 38.Rg8+ Ke7
 11/30  00:04   35k  7k  +0.20  25. ... cxd4 26.cxd4 Rac8 27.Rf3 Nh4 28.Rf4 f5 29.Nxf5 Nxf5 30.Qxg4 Qxg4 31.Rxg4 Nxd4 32.Bb2 Ne6 33.Reg1 Re7 34.h4 Kh7 35.R1g2 Rc4 36.Kh2 b5 37.h5
... same move,  except * very brief one-time 'glitch'
 15/35  00:13   128k  10k  +0.16  25. ... cxd4 26.cxd4 Rac8 27.Rf3 Nh4 28.Rf4 f5 29.Nxf5 Nxf5 30.Qxg4 Qxg4 31.Rxg4 Nxd4 32.Bb2 Ne6 33.Rd1 Re7 34.Rd6 Rc2 35.Bf6 Re8 36.Rd7 Rf8 37.Bxg7 Rf1+ 38.Rg1 Rxg1+ 39.Kxg1
 15/35  00:17   167k  10k  +0.15 *25. ... Rad8 26.Rd1 Kf7 27.Kg1 cxd4 28.cxd4 Rh8 29.Rd2 Rc8 30.e5 Nf4 31.Rxf4 Rxc1+ 32.Rf1 Rhc8 33.e6+ Kf8 34.Nf5 Rxf1+ 35.Qxf1 Qxf1+ 36.Kxf1 Nh6 37.Nxh6 gxh6 38.d5 Ke7 39.Kf2 Kd6 40.Kf3
 15/35  00:19   187k  10k  +0.10  25. ... cxd4 26.cxd4 Rac8 27.Rf3 Nh4 28.Rf4 f5 29.Nxf5 Nxf5 30.Qxg4 Qxg4 31.Rxg4 Nxd4 32.Bb2 Ne6 33.Rd1 Re7 34.Rd6 Rc2 35.Bf6 Re8 36.Rd7 Rf8 37.Bxg7 Rf1+ 38.Rg1 Rxg1+ 39.Kxg1
... same move ...
 24/56  05:33   8,032k  24k  +0.06  25. ... cxd4 26.cxd4 Rac8 27.Rf3 Nh4 28.Rd3 f5 29.e5 Rf8 30.Rf1 Ng6 31.Rdf3 Nh4 32.Rd3 Ng6 33.Rdf3 Nh4 34.Rd3

Up Topic The Rybka Lounge / Computer Chess / Fat Fritz

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill