Fat Fritz outmatches Stockfish (part 2)

by Elshan Moradiabadi
11/7/2019 – Yesterday we announced the results of an internal 100-game match pitting our new Fat Fritz engine against two versions of Stockfish. Today, further annotated highlights from Grandmaster ELSHAN MORADIABADI along with some of your feedback and questions. Stay tuned for more news next week on Fat Fritz that you won't want to miss!

ChessBase 15 - Mega package ChessBase 15 - Mega package

Find the right combination! ChessBase 15 program + new Mega Database 2020 with 8 million games and more than 80,000 master analyses. Plus ChessBase Magazine (DVD + magazine) and CB Premium membership for 1 year!

More...

Next generation engine duels

Recapping part one: ChessBase recently staged two matches between Fat Fritz and Stockfish. The computer we used for both Fat Fritz and Stockfish was a Ryzen 7 3700X with an NVIDIA RTX 2080, 32 GB RAM. Ponder was switched off, the Silver Openings Suite was used to ensure maximum opening diversity, and the time control was 20m+10s.

Stockfish ran on 16 threads at 16 million nodes per second, while the pre-release version of Fat Fritz ran at a little over 11 thousand NPS (for the Leela specialists this was done with the slower CUDNN backend). The speed conditions are actually a good deal worse than AlphaZero, which had around a 900 to 1 NPS speed difference, while here Stockfish was running at 1450 times faster in NPS, so if anything, it favored Stockfish. 

We conducted matches against Stockfish 8 and Stockfish 10. Here are the results:

match results table

You'll find all games below.

Now, picking up on the commentary from GM Elshan Moradiabadi...


Where human intuition trains AI (part 2)

Having spoken about FF’s ability to play dynamic chess, let us see this pure demonstration of positional ideas, very similar to Leela in this regard, which distinguishes neural networks from other types of engines.

[Event "Rapid match 20m+10s"] [Site ""] [Date "2019.10.23"] [Round "5"] [White "Fat Fritz"] [Black "Stockfish 10 64 POPCNT"] [Result "1-0"] [ECO "A28"] [Annotator "Elshan"] [PlyCount "139"] [EventDate "2019.??.??"] [TimeControl "1200+10"] 1. c4 e5 2. Nc3 Nf6 3. Nf3 Nc6 4. e4 {0.17/13 25 The provided theory was already over a move ago, but FF amazes me by once again choosing the most fashionable line these days.} Bb4 {-0.30/33 62} 5. d3 {0.17/12 15} d6 {-0.20/ 30 18} 6. Be2 {0.16/12 35 This is also Magnus Carlsen's choice, though the general belief is that White should not let Black play a5.} a5 {-0.31/31 28 (0-0)} 7. O-O {0.15/12 56 (Le3)} O-O {-0.18/32 51} 8. h3 {0.13/12 11 (a3)} Bc5 {-0.44/32 53} 9. Be3 {0.11/14 40} Re8 {-0.39/29 19 (Sd4)} 10. Qd2 {0.13/11 31 (Sd5)} Bd7 {-0.43/31 13} 11. Bd1 {0.11/12 38 (Sd5)} h6 {-0.12/33 86} 12. Kh1 { 0.14/15 50 (Tc1)} Nh7 {-0.09/25 25 (Sd4)} 13. Nh2 {0.20/12 23} Rf8 {-0.11/30 64 } 14. Bg4 {0.21/13 41 (Tc1)} Nf6 {-0.15/28 14} 15. Bd1 {0.19/13 53 (Lxd7)} b6 { -0.15/33 151 (Lxe3)} 16. f4 {0.35/12 21 (Tc1)} Bxe3 {0.00/33 66} 17. Qxe3 { 0.33/14 16} Rb8 {0.00/34 34} 18. b3 {0.28/14 61 (fxe5)} Qe7 {-0.14/30 9} 19. f5 {0.27/13 7 (fxe5)} Nd4 {-0.37/30 15} 20. a4 {0.28/14 35 (Sd5)} c6 {-0.38/30 31 (Kh7)} 21. Nf3 {0.36/13 33} c5 {-0.27/32 17 (Sxf3)} 22. Ra2 {0.56/13 42 (Sh2)} Bc6 {0.00/39 56} 23. g4 {0.53/14 19} Nh7 {0.09/39 117} 24. Rh2 {0.50/13 26} Nxf3 {0.09/42 18 (g5)} 25. Bxf3 {0.60/16 52} Qh4 {0.09/44 14 (g5)} 26. Nd1 { 0.79/12 58 (Tg1)} Rbc8 {0.73/32 188 (Dg5)} 27. Be2 {0.94/11 29 (De2)} Rc7 { 0.97/32 60} 28. Rg1 {1.00/11 26 (Dg1)} g6 {0.88/28 11} 29. Qd2 {1.01/11 30} Kh8 {0.91/30 24 (Kg7)} 30. Ne3 {1.04/12 33} Qf6 {1.22/34 66 (Te8)} 31. h4 {1.13/16 32} g5 {1.12/39 43} 32. Nf1 {1.19/15 21} Qd8 {1.12/40 14 (gxh4)} 33. hxg5 { 1.25/14 39} Nxg5 {1.12/42 20 (hxg5)} 34. Rxh6+ {1.33/25 41} Kg7 {1.12/45 14 [#] This is a typical example of the advantage of AI (neural networks) over other brute force chess engines: intuition. I honestly don't know how they train themselves on it, but its resemblance to human intuition is astonishing. Here I had my computer run on this position and it only showed White's pseudo exchange-sac on depth 35. Needless to say, my engine couldn't detect the idea 3 or 4 moves in advance.} 35. Rxd6 $1 {1.41/28 22 Black is busted, the position is closed and white's extra space is the crucial factor here. It is the closest you can find human's intuition to AI.} Rh8+ {1.09/33 23} 36. Kg2 { 1.42/25 11} Qxd6 {1.75/34 22} 37. Qxg5+ {1.50/25 23} Kf8 {1.67/34 7} 38. f6 { 1.54/24 14} Ke8 {1.48/36 17} 39. Qg7 {1.55/22 0 (Th1)} Qf8 {1.72/36 12} 40. Qxf8+ {1.59/20 25 (Dg5)} Kxf8 {0.78/31 9} 41. Ne3 {1.60/21 0 (g5)} Rc8 { 2.21/41 73 (Ld7)} 42. g5 {1.58/17 88} Rg8 {2.23/36 39} 43. Kf3 {1.66/20 46 (Kf2)} Bd7 {2.24/31 7} 44. Nf5 {1.72/20 19} Rc6 {2.12/35 10} 45. Rg3 {1.79/20 42 (Kf2)} Bxf5 {2.12/33 4 (Th8)} 46. exf5 {2.46/18 62} Rd6 {2.87/37 13} 47. Ke4 {2.48/19 13} Rxf6 {2.96/39 15} 48. Kxe5 {2.50/17 17} Rc6 {2.92/40 7 (Th6)} 49. g6 {2.47/19 45} Ke7 {3.76/41 26 (Tc7)} 50. Bf3 {2.73/18 44} Rf6 {6.82/31 14} 51. Rh3 {2.74/19 10 (g7)} Rg7 {2.95/36 2} 52. gxf7 {2.76/17 17} Rgxf7 {2.97/40 6} 53. Rh5 {2.80/17 9} Rd6 {3.73/39 21 (Tf8)} 54. Bd5 {2.99/17 40} Rff6 { 4.21/32 11} 55. Rh7+ {3.02/16 12} Kd8 {4.58/32 10} 56. Rb7 {3.11/15 18 (d4)} Rh6 {2.83/36 3} 57. Be6 {3.12/15 13} Rxd3 {2.82/37 5} 58. Rxb6 {3.05/19 11} Kc7 {2.87/43 20} 59. Rb5 {2.90/17 30} Re3+ {3.07/41 9 (Kc6)} 60. Kd5 {3.38/12 40} Rexe6 {4.26/41 14 (Td3+)} 61. fxe6 {3.55/16 29} Rh5+ {5.37/40 10} 62. Ke4 { 3.42/16 0} Kd6 {5.86/39 10} 63. Rb6+ {3.68/14 16} Ke7 {6.35/35 11} 64. Ra6 { 3.71/13 9} Kf6 {6.37/36 10 (Tg5)} 65. Kd3 {3.97/12 20} Rg5 {7.38/32 10 (Ke7)} 66. Kc2 {3.91/11 35 (Kd2)} Rg2+ {5.28/26 3 (Th5)} 67. Kb1 {3.83/11 34} Rg5 { 8.77/35 17} 68. Kb2 {3.76/11 8 (Ka2)} Ke7 {9.39/32 10 (Tf5)} 69. Ka3 {3.83/11 11} Rh5 {10.49/35 10} 70. Rxa5 {3.43/11 10} 1-0

The following game is another breathtaking draw, which reminds a classical chess geek such as myself of the famous Kotov-Gligoric game from Zurich 1953. In this game FF offers not one but two positional exchange sacrifices, locking down White’s pieces and forcing Stockfish to return the sacrificed material seeking equality. This game is your assurance of working with a very creative engine whose intuition is trained at the highest level. If one ever suffered from skepticism over the accuracy of an engine then a game like this is like a life-time warranty, which assures one of the soundness of the engine’s judgement.

[Event "Rapid match 20m+10s"] [Site ""] [Date "2019.10.24"] [Round "12"] [White "Stockfish 10 64 POPCNT"] [Black "Fat Fritz"] [Result "1/2-1/2"] [ECO "A67"] [Annotator "Elshan"] [PlyCount "409"] [EventDate "2019.??.??"] [TimeControl "1200+10"] 1. d4 Nf6 2. c4 c5 3. d5 e6 4. Nc3 exd5 5. cxd5 d6 6. e4 g6 {Last book move.} 7. f4 {1.09/32 175} Bg7 {0.80/18 17} 8. Bb5+ {0.95/29 19} Nfd7 {0.81/20 18} 9. Bd3 {1.01/29 17 (a4)} Qh4+ {0.66/15 47 (0-0)} 10. g3 {0.98/29 17} Qd8 {0.66/18 13} 11. Nf3 {0.99/28 14} Na6 {0.66/17 60 (0-0)} 12. Nb5 {0.86/29 97 (0-0)} Nb6 {0.72/19 29} 13. f5 {0.80/29 15} O-O {0.83/21 69 (c4)} 14. O-O {1.00/28 13} c4 {0.84/20 0} 15. Bb1 {1.12/30 29 (Lc2)} Nd7 {0.24/17 40} 16. Kh1 {0.81/30 37 (Le3)} Qb6 {0.13/19 35} 17. a4 {0.98/30 24} Nac5 {0.11/19 17} 18. Bc2 {0.70/33 123 (Lf4)} a6 {0.06/20 42} 19. Nc3 {0.80/28 11} Qc7 {0.05/24 21 (Db4)} 20. g4 { 0.57/31 99 (Lf4)} Rb8 {-0.13/18 47} 21. a5 {0.12/31 72 [#] Unlike Leela, Fat Fritz has a sweet tooth for complexity, and while Leela deals with positional sacrifices as they happen (as a matter of necessity or not), Fat Fritz actively seeks complex lines, making it a great companion for those who take a liking to the games of Kasparov and Tal!} b6 {-0.16/20 28 weakening c6,} 22. Nd4 {0.43/25 17 (De1)} b5 {-0.26/22 54 Couldn't you do it earlier?!} 23. Nc6 { 0.69/27 16 [#]} b4 $3 {-0.25/31 31 The two exclamation points is for the aesthetic essence of the idea and not its soundness, though the idea is one of the three sound ones in this position.} 24. Nxb8 {0.00/35 79 (Sa4)} Qxb8 { -0.37/22 39} 25. Na4 {0.00/36 17} b3 {-0.38/23 20} 26. Bb1 {0.00/35 14} Nxa4 { -0.35/24 23} 27. Rxa4 {0.00/35 18} Qb5 {-0.35/26 0} 28. Ra3 {0.00/37 19 (Ta1)} Be5 {-0.47/17 63 White's attack and position is completely blocked.} 29. Qe2 { 0.15/28 23} Nc5 $5 {-0.47/18 0 Direct and aggressive.} 30. Bh6 {0.00/32 92 (Tf2)} Bd7 $5 {-0.29/32 86 [#] The second exchange sacrifice! I heard Fat Fritz had gone through all of the games on Mega Database, I am sure it took a close look at Kotov-Gligoric Zurich 1953.} 31. Bxf8 {0.00/36 15 (Tc1)} Kxf8 { -0.31/27 63} 32. Rc1 {0.00/39 17} Qb7 {-0.26/32 25} 33. Rxc4 $1 {0.00/40 22 Stockfish realizes the dire situation and gives back material for activity.} Bb5 {-0.23/31 18} 34. Bd3 {0.00/42 17} Bxc4 {-0.23/27 0} 35. Bxc4 {0.00/43 39} Qb4 {-0.23/26 0} 36. Ra1 {0.00/45 14} g5 $1 {-0.23/25 44 JIT: Just-in-time!} 37. Kg2 {0.00/36 41 (h4)} h6 {-0.46/22 120 [#] (Sa4) The position is dynamically balanced and the two giant engines concede to this fact after a hard-fought battle.} 38. h3 {0.00/38 52 (Tf1)} Kg8 {-0.90/19 61 (Ke7)} 39. Rf1 {0.00/37 16 (f6)} Bxb2 {-0.84/21 37 (Dxa5)} 40. Rb1 {0.00/40 13} Be5 {-0.81/19 23 (Ld4)} 41. Bxb3 {0.00/45 17} Qxa5 {-0.78/19 27 (Dc3)} 42. Qe3 {0.00/43 12} Qa3 {-0.76/21 108} 43. Qf3 {0.00/45 13} Qb4 {-0.77/17 0} 44. Bc2 {0.00/41 22} Qc4 {-0.76/23 10 (Dd2+)} 45. Rb8+ {0.00/45 15 (Lb3)} Kg7 {-0.89/17 45} 46. Bb1 {0.00/45 12} Nd7 {-0.85/18 62 (Kf6)} 47. Rb7 {0.00/42 16} Qc8 {-0.85/16 0 (Sc5) } 48. Rb3 {0.00/40 12} a5 {-0.84/17 21 (Sc5)} 49. Qe3 {0.00/41 14 (Dd1)} Nc5 { -0.90/16 73 (Kf6)} 50. Rb5 {0.00/39 15} Qa6 {-0.82/20 30} 51. Qe2 {0.00/44 12 (Txc5)} a4 {-0.77/15 26} 52. Qc4 {0.00/45 18} Qa7 {-0.75/21 7} 53. Ba2 { 0.00/41 11 (Db4)} Nd7 {-1.27/12 22} 54. Qd3 {0.00/42 29} Nf6 {-1.26/13 22} 55. Rb4 {0.00/45 11} Nd7 {-1.24/13 8} 56. Bc4 {0.00/42 12 (Tb5)} a3 {-1.38/13 23} 57. Ba2 {0.00/46 13} Qa5 {-1.35/13 31 (Dc5)} 58. Qb5 {0.00/45 15} Qa7 {-1.34/ 13 3 (Dc7)} 59. Qd3 {0.00/45 13 (De2)} Bb2 {-1.33/12 26 (Dc5)} 60. Qd2 { 0.00/41 11} Qa6 {-1.29/13 29} 61. Bb1 {0.00/40 11 (Lc4)} Be5 {-1.29/10 9 (Sc5)} 62. Qc2 {0.00/42 12 (La2)} Qa7 {-1.41/12 11} 63. Ba2 {0.00/46 25 (Db3)} Qa5 { -1.34/12 22 (De3)} 64. Qd2 {0.00/44 8 (Dc4)} Ba1 {-1.30/10 12 (Sc5)} 65. Rb2 { 0.00/42 12} Qc5 {-1.24/12 13 (Da6)} 66. Rb3 {0.00/42 13 (Tc2)} Be5 {-1.12/13 14 (Lb2)} 67. Bb1 {0.00/41 21} Bb2 {-0.81/17 10 (Sf6)} 68. h4 {0.00/40 15 (Tb7) } Bf6 {-0.99/19 6 (Kf6)} 69. hxg5 {0.00/42 12} hxg5 {-0.91/21 6} 70. Rb7 { 0.00/40 12 (De3)} Ne5 {-1.06/15 13} 71. Rb3 {0.00/43 11} Nd7 {-0.99/21 5 (Sxg4) } 72. Ba2 {0.00/46 17 (De3)} Ne5 {-1.11/14 13 (Dc4)} 73. Qc3 {0.00/37 8} Nc4 { -1.05/15 6} 74. Qd3 {0.00/38 10} Be5 {-0.74/18 31 (Lb2)} 75. Bb1 {0.00/41 10} Bf6 {-0.65/21 1} 76. Ba2 {0.00/41 11} Bd4 {-0.57/19 11 (Se5)} 77. Bb1 {0.00/42 10} Ne5 {-0.54/22 8 (Lf6)} 78. Qd2 {0.00/42 9} Kf6 {-0.53/21 5} 79. Qb4 { 0.00/44 11} Qxb4 {-0.52/20 4} 80. Rxb4 {0.00/47 15} Bc5 {-0.52/17 7} 81. Rb8 { 0.00/46 12 (Ta4)} Nxg4 {-0.57/12 16} 82. Re8 {0.00/44 10} Ne5 {-0.53/12 22} 83. Ba2 {0.00/47 21 (Kg3)} Nd7 {-0.52/11 19 (Le3)} 84. Kg3 {0.00/47 23} Ne5 { -0.49/11 10} 85. Bb1 {0.00/44 24 (Ta8)} Bb4 {-0.49/10 10 (Sc4)} 86. Ba2 { 0.00/42 6 (Ta8)} Be1+ {-0.51/10 7} 87. Kg2 {0.00/48 8} Bd2 {-0.50/11 14 (Lb4)} 88. Rd8 {0.00/48 8 (Ta8)} Bb4 {-0.50/11 5} 89. Ra8 {0.00/49 11 (Tb8)} Nd3 { -0.47/10 15 (Lc5)} 90. Bb1 {0.00/45 15 (Kg3)} Ne5 {-0.47/11 10} 91. Ba2 { 0.00/48 9 (Kg3)} Bc5 {-0.45/10 8} 92. Kg3 {0.00/48 53} Kg7 {-0.42/10 10 (Lb4)} 93. Bb3 {0.00/47 5 (Lb1)} Kh6 {-0.42/10 12 (Kf6)} 94. Bd1 {0.00/44 7 (Tg8)} Kh7 {-0.45/9 9 (Sd7)} 95. Bh5 {0.00/46 7 (Lc2)} Kg7 {-0.42/11 9} 96. Be2 {0.00/50 10} Kh7 {-0.39/11 11 (Sd7)} 97. Kh3 {0.00/50 9 (Ld1)} Kh6 {-0.41/10 8 (Kg7)} 98. Kg3 {0.00/51 8} Kg7 {-0.38/11 11} 99. Ra6 {0.00/52 12 (Lb5)} Kf6 {-0.41/10 12} 100. Kh3 {0.00/48 11 (Ta4)} Ke7 {-0.43/11 9} 101. Kg2 {0.00/52 9 (Lb5)} Bb4 {-0.47/11 12 (Kf6)} 102. Ra8 {0.00/50 18 (Lb5)} Bc5 {-0.42/12 10 (Kf6)} 103. Kg3 {0.00/49 7} Bb4 {-0.33/13 9 (Kf6)} 104. Bh5 {0.00/42 8 (Ta4)} Nd7 {-0.41/ 12 10 (Kf6)} 105. Ra4 {0.00/45 10} Bc5 {-0.38/14 3} 106. Be2 {0.00/54 28} Nb6 { -0.35/13 9 (Se5)} 107. Ra6 {0.00/51 6} Nd7 {-0.33/13 6} 108. Kg4 {0.00/51 11 (Ta4)} f6 {-0.39/14 11 (Kf6)} 109. Bb5 {0.31/48 7} Ne5+ {-0.34/15 6} 110. Kg3 { 0.31/52 7 (Kh5)} Kd8 {-0.40/11 11} 111. Bf1 {0.31/50 8 (Ta4)} Kc7 {-0.37/11 24} 112. Be2 {0.31/51 7 (Ta4)} Kb7 {-0.37/10 10 (Kd8)} 113. Kh2 {0.31/43 7 (Ta4)} Kc8 {-0.42/10 9 (Kc7)} 114. Kg3 {0.30/53 47} Kc7 {-0.37/9 11 (Kd8)} 115. Bf1 { 0.30/50 10 (Ta4)} Nd7 {-0.36/10 10 (Kd7)} 116. Bc4 {0.29/46 6 (Lb5)} Kb7 { -0.37/10 10 (Se5)} 117. Bb5 {0.29/51 8} Ne5 {-0.37/10 4} 118. Bf1 {0.28/48 16 (Le2)} Kb8 {-0.36/10 16 (Kc7)} 119. Ra5 {0.21/51 6 (Lb5)} Kc8 {-0.37/9 9 (Kb7)} 120. Be2 {0.20/53 6 (Ta4)} Nd7 {-0.37/9 11 (Kc7)} 121. Bb5 {0.20/48 13 (Ta4)} Ne5 {-0.34/10 4} 122. Bf1 {0.20/54 8 (Le2)} Nd7 {-0.38/9 16 (Kc7)} 123. Ra8+ { 0.11/45 9 (Lb5)} Kc7 {-0.37/11 8 (Kb7)} 124. Bb5 {0.01/49 24 (Ta4)} Ne5 { -0.36/11 5} 125. Bf1 {0.01/53 7 (Le2)} Kb6 {-0.31/10 16 (Lb4)} 126. Ra6+ { 0.00/52 9 (Ta4)} Kb7 {-0.29/10 7 (Kc7)} 127. Be2 {0.00/46 31 (Ta4)} Kc8 { -0.30/10 13 (Kc7)} 128. Bd1 {0.00/50 5 (Ta4)} Nd7 {-0.37/11 12 (Kc7)} 129. Bb3 {0.00/50 11 (Le2)} Ne5 {-0.41/10 10 (Kb7)} 130. Ra8+ {0.00/43 5 (Ld1)} Kc7 { -0.39/13 5} 131. Ba2 {0.00/47 9 (Tf8)} Nd7 {-0.36/10 9 (Sd3)} 132. Rh8 { 0.00/41 10 (Te8)} Be3 {-0.33/10 16 (Ld4)} 133. Kf3 {0.00/43 6 (Ta8)} Bc1 { -0.40/10 7} 134. Ra8 {0.00/44 7 (Th7)} Nc5 {-0.45/12 7 (Kb7)} 135. Bc4 { 0.00/42 13} Kb7 {-0.44/14 6 (Lb2)} 136. Ra5 {0.00/42 5} Nd7 {-0.43/13 5} 137. Rb5+ {0.00/41 10} Ka7 {-0.42/12 16 (Kc7)} 138. Ke2 {0.00/42 6 (Ta5+)} Ne5 { -0.48/12 14 (Lb2)} 139. Ba2 {0.00/44 7} Bb2 {-0.43/13 12 (Sd7)} 140. Kd2 { 0.00/42 8 (Ke3)} Nd7 {-0.52/11 9} 141. Ke3 {0.00/44 8 (Ke2)} Bc3 {-0.50/11 13 (Lc1+)} 142. Rb3 {0.00/45 6 (Lb1)} Bb2 {-0.61/12 5} 143. Rb5 {0.00/50 12 (Tb4)} Ne5 {-0.39/11 15 (Lc1+)} 144. Rb4 {0.00/44 7 (Ke2)} Nf7 {-0.36/11 11 (Sd7)} 145. Bc4 {0.00/42 14 (Tb5)} Ka8 {-0.30/12 10 (Se5)} 146. Kd3 {0.00/43 13 (Le2)} Ne5+ {-0.16/14 9} 147. Kc2 {0.00/46 8} g4 {-0.11/15 6} 148. Kb1 {0.00/44 20 (Lf1)} g3 {-0.23/18 7} 149. Bf1 {0.00/46 8} Nf3 {-0.15/20 5} 150. Bh3 {0.00/43 14} Ka7 {-0.10/21 11} 151. Ka2 {0.00/46 6} Ng1 {-0.07/21 20} 152. Bf1 {0.00/45 13} Nf3 {-0.08/22 0} 153. Bh3 {0.00/46 15} Ne1 {-0.05/19 6 (Sg1)} 154. Rb3 { 0.80/35 5} Be5 {-0.03/21 3 (g2)} 155. Rxa3+ {0.84/36 6} Kb6 {0.03/20 7} 156. Re3 {0.84/43 7} Nc2 {0.06/19 5} 157. Rb3+ {0.83/45 13} Kc5 {0.08/18 6} 158. Kb1 {0.83/49 9} Nd4 {0.09/17 6} 159. Rc3+ {0.82/48 12 (Td3)} Kb4 {0.06/14 24} 160. Rd3 {0.82/49 6} Kc4 {0.10/14 13 (Sb5)} 161. Re3 {1.09/46 5} Nb5 {0.16/15 29 (Kc5)} 162. Kc2 {1.07/45 7} Nd4+ {0.15/13 0} 163. Kd1 {1.00/50 9} Kb5 {0.16/13 15 (Lf4)} 164. Ra3 {1.00/45 6 (Lf1+)} Kb4 {0.19/12 9} 165. Ra2 {1.00/47 7 (Ta8) } Kc3 {0.16/10 15} 166. Bf1 {1.00/51 8} Nb3 {0.19/11 6} 167. Ra8 {1.00/52 12 (Ta3)} Nc5 {0.17/10 11 (Kd4)} 168. Bg2 {1.00/49 7 (Ke2)} Nb3 {0.15/10 9} 169. Bf1 {1.00/55 8 (Ta3)} Nc5 {0.16/10 11 (Kd4)} 170. Bg2 {1.00/52 8 (Ke2)} Nb3 { 0.13/9 9} 171. Ra4 {1.00/54 7 (Ta3)} Kd3 {0.14/10 16} 172. Ke1 {1.00/52 12 (Ta3)} Nd4 {0.05/10 7 (Lc3+)} 173. Ra3+ {1.00/45 7} Kc2 {0.07/11 8} 174. Bf1 { 1.00/50 7} Bf4 {0.14/11 11 (Sb3)} 175. Ra2+ {1.00/47 9} Kc3 {0.18/11 13 (Kc1)} 176. Kd1 {1.00/47 8 (Ta8)} Be5 {0.16/10 8} 177. Ra4 {1.00/55 9 (Ta8)} Nb3 { 0.15/10 12} 178. Rc4+ {1.00/56 12} Kb2 {0.19/10 0} 179. Rc2+ {1.00/60 21 (Tc8)} Ka3 {0.22/10 13} 180. Rc6 {1.00/50 15 (Tc8)} Nc5 {0.16/9 16 (Kb2)} 181. Kc2 { 1.27/43 4 (Tb6)} Kb4 {0.19/10 15 (Sxe4)} 182. Rc8 {1.58/43 6} Ka5 {0.20/9 10} 183. Bg2 {1.58/46 8 (Tb8)} Kb4 {0.17/8 10} 184. Bf1 {1.58/44 7 (Kd2)} Ka5 { 0.18/9 10} 185. Ra8+ {1.29/45 58 (Tb8)} Kb6 {0.18/9 10} 186. Bd3 {1.14/45 13 (Tb8+)} Kb7 {0.12/9 8} 187. Rg8 {1.14/45 8} Kb6 {0.15/9 9 (Kc7)} 188. Rg7 { 1.14/41 5 (Lf1)} Ka5 {0.12/9 10 (Lf4)} 189. Rc7 {1.14/45 6 (Lf1)} Kb6 {0.10/9 9 } 190. Rg7 {1.14/46 6} Ka5 {0.11/9 11} 191. Rg4 {1.14/46 8 (Lf1)} Kb4 {0.09/9 13 (Kb6)} 192. Rg8 {1.14/44 10 (Lf1)} Ka4 {0.11/9 10 (Sd7)} 193. Rc8 {1.13/47 36 (Lf1)} Kb4 {0.11/9 10} 194. Rf8 {1.05/49 13 (Ta8)} Ka5 {0.11/9 11 (Ka3)} 195. Rb8 {1.05/42 4} Nd7 {0.11/10 8 (Lf4)} 196. Rb1 {1.05/40 5} Nc5 {0.12/10 5 (Lf4)} 197. Kd2 {1.05/43 9} Ka4 {0.12/10 9 (Lf4+)} 198. Bb5+ {1.04/45 11 (Ke3)} Ka3 {0.11/10 14 (Ka5)} 199. Kc2 {1.04/48 4} Ka2 {0.10/11 6 (Ld4)} 200. Be2 { 1.04/49 7 (Lf1)} Ka3 {0.08/9 8} 201. Rb8 {1.00/48 14} Ka4 {0.05/9 9 (Sxe4)} 202. Bf1 {1.00/45 4} Ka5 {0.04/7 19 (Lf4)} 203. Bc4 {0.00/58 29 (Tb1)} Nd7 { 0.00/5 11} 204. Ra8+ {0.00/78 7 (Tb1)} Kb4 {0.00/2 12} 205. Bf1 {0.00/127 0 Draw agreed.} 1/2-1/2

Earlier in this article, I commented on FF’s technical play and its approach. Most of the time engines become very similar when the endgame is reached, as they all use tablebases. However, even in such cases, FF tries its best to solve difficult endings, remaining faithful to active ides and the concept of domination. Its preference for prophylactic play is a feature I personally cherish most, though one should be careful learning these paths as they may not be the most feasible solutions for over the board games. Still, let us look at the following game, where FF handles a slightly worse position in a very unconventional but principled way.

[Event "Rapid match 20m+10s"] [Site ""] [Date "2019.10.24"] [Round "13"] [White "Fat Fritz"] [Black "Stockfish 10 64 POPCNT"] [Result "1/2-1/2"] [ECO "A87"] [Annotator "Elshan"] [PlyCount "145"] [EventDate "2019.??.??"] [TimeControl "1200+10"] 1. d4 f5 2. g3 Nf6 3. Bg2 g6 4. Nf3 Bg7 5. O-O O-O 6. c4 d6 7. Nc3 Qe8 8. d5 { Last book move.} Na6 {0.25/31 95} 9. Be3 {0.69/13 31} Bd7 {0.17/31 51} 10. Rb1 {0.73/13 20 (Tc1)} c5 {0.37/30 47 (c6)} 11. Qd2 {0.75/18 26} Nc7 {0.44/31 19} 12. b4 {0.72/18 34} cxb4 {0.32/32 17} 13. Rxb4 {0.73/19 13} b6 {0.59/36 107} 14. h3 {0.72/18 58 (Tb3)} Na6 {0.27/31 20 (Dc8)} 15. Rb3 {0.71/19 30} Qc8 { 0.44/32 22} 16. Nd4 {0.67/22 87} Nc5 {0.22/32 45 (Dxc4)} 17. Ra3 {0.86/18 22} Nfe4 {0.30/36 57 (Sce4)} 18. Nxe4 {0.90/17 32} Nxe4 {0.40/35 16} 19. Qd3 { 0.87/20 58 (Dc2)} Nc5 {0.33/36 20} 20. Qd1 {0.94/13 38 (Dd2)} Ne4 {0.33/36 26 (Lf6)} 21. Qd3 {0.89/18 35 (Dc2)} Nc5 {0.35/37 19} 22. Qd2 {0.88/23 24} Ne4 { 0.46/36 34} 23. Qc2 {0.83/24 15} b5 {0.43/36 22} 24. Bxe4 {0.81/24 23} Bxd4 { 0.42/35 17} 25. Bxd4 {0.81/24 20} fxe4 {0.48/37 65} 26. g4 {0.78/27 69 (Te3)} Qxc4 {0.14/33 20} 27. Qxe4 {0.77/31 0} b4 {0.48/35 45} 28. Rd3 {0.76/29 18} Rae8 {0.25/33 14} 29. Rd2 {0.75/27 23} e5 {0.26/38 98} 30. dxe6 {0.73/26 22} Rxe6 {0.26/36 13} 31. Qb7 {0.73/26 30} Rf7 {0.62/37 115 (Te7)} 32. Qb8+ { 1.07/19 92} Bc8 {0.33/34 13 (Te8)} 33. Bxa7 {0.67/21 125 (Le3)} Rb7 {0.00/38 13 (h5)} 34. Qa8 {0.24/19 32} d5 {0.00/43 19} 35. Rdd1 {0.15/21 27} Ra6 { 0.00/43 33} 36. Rc1 {0.14/20 16} Rbxa7 {0.00/41 28} 37. Qxc8+ {0.13/18 15} Qxc8 {0.00/45 28} 38. Rxc8+ {0.14/17 15} Kg7 {0.00/45 22} 39. Rb8 {0.13/16 141 (Tb1) } Rxa2 {0.00/42 16} 40. e3 {0.13/14 0} Rb2 {0.00/43 43 (d4)} 41. Kg2 {0.18/11 54 (Tb5)} Kf6 {0.00/40 24 (Td7)} 42. Kg3 {0.17/13 69} Rf7 {0.00/41 18} 43. Rd1 {0.17/14 8 (Tb6+)} Kg7 {0.00/41 15} 44. Rf1 {0.16/13 18} Rb3 {0.00/42 35} ( 44... Rd7 45. Rb5 d4 46. exd4 Rxd4 47. Rb7+ Kf6) 45. Kg2 {0.16/13 38 [#] Here is another difference between FF to any other engine or human way of thinking: It has a great interest in fortresses. Like the previous game, it creates a situation making it impossible for White to improve. By contrast, GMs would try to activate the king and trade the d-pawn for the e-pawn and hold a 2-3 endgame.} h5 {0.00/43 29} (45... Rd7 46. Rb5 d4 47. exd4 Rxd4) 46. Rb5 { 0.16/13 42 (Td1)} h4 $1 {0.00/40 23 (d4)} 47. Rxd5 {0.25/12 26} Rb2 {0.00/43 20 Now white is tied down.} 48. Rb5 {0.23/12 36 (Td4)} Kh6 {0.00/39 14 (b3)} 49. Rb8 {0.23/11 52 (Tb6)} Re7 {0.00/46 39 (b3)} 50. Re1 {0.26/11 24 (Kf3)} Rf7 {0.00/45 17} 51. Rf1 {0.19/13 9} b3 {0.00/46 10 (Te7)} 52. Rb5 {0.23/11 37} Kg7 {0.00/46 11} 53. Rb6 {0.22/11 22 (g5)} Kf8 {0.00/42 11 (Kh6)} 54. e4 {0.24/11 19 (g5)} Rf4 {0.00/37 12 (Kg7)} 55. e5 {0.19/12 18 (Kg1)} Ke7 {0.00/39 13 (g5)} 56. Kg1 {0.24/13 14} Kd7 {0.00/41 17 (Te4)} 57. e6+ {0.20/15 19 (Td1+)} Kc7 { 0.00/43 15} 58. Rb5 {0.14/16 19} Kd6 {0.00/41 16} 59. Rb6+ {0.12/15 13} Kc7 { 0.00/44 20} 60. Re1 {0.10/14 27 (e7)} Rfxf2 {0.00/46 9} 61. e7 {0.08/12 16} Rg2+ {0.00/45 24} 62. Kh1 {0.07/9 0} Rh2+ {0.00/46 19} 63. Kg1 {0.05/7 0} Rbg2+ {0.00/50 70 (Thg2+)} 64. Kf1 {0.12/13 0} Rf2+ {0.00/53 19 (Tb2)} 65. Kg1 { 0.13/10 0} Rfg2+ {0.00/47 17} 66. Kf1 {0.08/7 0} Rf2+ {0.00/50 20 (Td2)} 67. Kg1 {0.13/6 0} Rhg2+ {0.00/49 59} 68. Kh1 {0.08/6 0} Kxb6 {0.00/57 48 (Th2+)} 69. e8=Q {0.03/9 85} Rh2+ {0.00/58 5} 70. Kg1 {0.02/7 0} Rhg2+ {0.00/66 11} 71. Kh1 {0.01/5 0} Rh2+ {0.00/64 29} 72. Kg1 {0.00/3 0} Rhg2+ {0.00/67 8} 73. Kh1 { 0.00/2 0 Remis angenommen} 1/2-1/2

A fast learner!

The following game is another example that demonstrates how well Fat Fritz understands pawn structures. In this dodgy line of the Alekhine, Fat Fritz outplays Stockfish 10 in a very classy manner.

[Event "Rapid match 20m+10s"] [Site ""] [Date "2019.10.24"] [Round "19"] [White "Fat Fritz"] [Black "Stockfish 10 64 POPCNT"] [Result "1-0"] [ECO "B04"] [Annotator "Elshan"] [PlyCount "145"] [EventDate "2019.??.??"] [TimeControl "1200+10"] 1. e4 Nf6 2. e5 Nd5 3. d4 d6 4. Nf3 {Last book move.} g6 {0.31/33 114} 5. Bc4 { 0.77/17 23} Nb6 {0.26/28 9} 6. Bb3 {0.77/17 13} Bg7 {0.23/30 20 (a5)} 7. Ng5 { 0.77/24 26 (De2)} e6 {0.10/29 20} 8. Qf3 {0.75/24 19} O-O {0.45/34 87 (De7)} 9. Qh3 {0.61/20 43} h6 {0.24/30 8} 10. Ne4 {0.55/20 77 (Sf3)} d5 {0.24/31 19} 11. Nc5 {0.55/17 0} Kh7 {0.24/32 24} 12. Be3 {0.58/17 40 (c3)} N8d7 {0.23/32 75} 13. Nxd7 {0.59/18 15} Bxd7 {0.20/31 11} 14. Nd2 {0.61/16 19} Na4 {0.22/35 42 (f5)} (14... c5) 15. O-O-O {0.70/17 53 (Tb1)} b6 {0.19/29 58 (b5)} (15... c5 16. Bxa4 Bxa4 17. dxc5 Qa5 18. Kb1 b6) 16. g4 {1.11/16 49 (Kb1)} c5 {0.15/27 11 } 17. Rhg1 {1.13/19 20 (c3)} Rh8 {0.50/32 69} 18. f4 {1.15/18 21} Qc7 {0.49/29 14 (Tc8)} 19. c3 {1.10/22 47} b5 {0.16/30 20} 20. Kb1 {1.11/21 23 (dxc5)} c4 { -0.09/28 11 (Tab8)} 21. Bxa4 {1.17/17 51} bxa4 {0.21/31 41} 22. a3 {1.19/15 19} Rab8 {0.58/33 133 (Thb8)} 23. Ka1 {1.37/12 51} Rb7 {0.59/35 12 (Tb6)} 24. Nb1 { 1.50/11 48 (Tb1)} Kg8 {1.07/32 87 (Db6)} 25. Qg2 {1.51/13 44} Kf8 {1.26/38 106} 26. Rdf1 {1.52/14 29} Qc8 {1.17/40 77 (Ke8)} 27. h4 {1.75/19 43} h5 {2.01/34 44 } 28. f5 {1.78/20 18} exf5 {1.90/32 10} 29. gxf5 {1.81/19 14} Bxf5 {1.86/36 12} 30. Qxd5 {1.83/18 15} Rd7 {2.08/36 24 (Ld3)} 31. Qg2 {2.16/15 67} Rc7 {2.44/37 67} 32. Nd2 {2.16/15 41} Kg8 {2.27/36 48 (Ld3)} 33. Rf4 {2.16/14 58 (Tf2)} Kh7 {2.11/30 8} 34. Ne4 {2.13/15 65} Bxe4 {2.16/36 14} 35. Qxe4 {2.12/14 0} Re8 { 2.15/38 26 (Db7)} 36. Rgf1 {2.09/11 80 (Tg5)} Kg8 {1.61/35 13 (De6)} 37. R4f2 { 2.05/11 51 (Dc2)} Qb7 {1.59/35 37} 38. Qc2 {2.02/11 16} Qc6 {1.58/37 7} 39. Qc1 {2.02/11 35} Rd7 {1.73/41 65 (De4)} 40. Bh6 {2.08/12 27} Bh8 {1.73/41 9 (De4)} 41. Rg1 {2.13/10 46} Qd5 {1.73/42 10 (Tb7)} 42. Qc2 {2.14/10 28} Kh7 {1.73/43 10} 43. Bc1 {2.05/11 37} Bg7 {1.73/45 13} 44. Qxa4 {2.03/11 24} Re6 {1.73/45 12 } 45. Qc2 {2.02/14 16 (Lg5)} Rb7 {1.73/42 13} 46. Re2 {2.06/14 21 (Tgf1)} Kh8 { 1.73/40 21 (Lf8)} 47. Qe4 {2.12/10 44 (Teg2)} Qxe4 {2.02/36 29 (Db5)} 48. Rxe4 {2.40/12 22} Kh7 {2.19/35 15} 49. Bf4 {2.45/14 17 (Ka2)} Rd7 {1.69/28 7} 50. Ka2 {2.54/14 19 (Kb1)} f6 {1.87/33 19} 51. Rge1 {2.58/14 11} fxe5 {1.97/33 28} 52. dxe5 {2.64/13 27 (Lg3)} Rc7 {2.02/31 9} 53. Rd1 {2.65/12 11 (Lg3)} Bf8 { 2.73/32 58} 54. Rdd4 {2.70/15 29} Be7 {2.89/31 23} 55. Bh2 {2.76/16 19} g5 { 2.71/31 12} 56. hxg5 {2.78/15 12} Bxg5 {3.08/30 23} 57. Rxc4 {2.87/14 21} Rxc4 {3.20/27 10} 58. Rxc4 {2.84/13 10} Kg6 {3.40/28 13} 59. b4 {2.87/12 15} Kf5 { 3.50/27 10 (h4)} 60. Kb3 {3.17/11 31} Rg6 {3.68/26 10 (h4)} 61. Rd4 {3.45/10 30 (Tc7)} Be7 {3.90/24 10} 62. Rd2 {3.57/10 16} Bh4 {4.18/27 10 (h4)} 63. a4 { 3.91/9 28} Be1 {4.49/27 10 (Le7)} 64. Re2 {5.56/9 12 (Td1)} Bh4 {4.50/30 3} 65. a5 {4.67/9 23 (Td2)} Ke6 {4.71/28 17 (Le7)} 66. a6 {4.76/8 12 (Td2)} Rg7 { 5.23/26 10 (Le7)} 67. c4 {4.72/8 16 (b5)} Rd7 {5.78/24 10 (Tg4)} 68. c5 { 5.72/7 10 (Tg2)} Be7 {7.03/26 10 (Td4)} 69. Kc4 {6.96/7 9 (Tc2)} Rd1 {6.12/23 2 } 70. Kb5 {5.88/8 14 (Tc2)} h4 {7.19/27 18 (Tb1)} 71. Rb2 {5.41/8 8 (Tg2)} Bd8 {7.53/27 10 (Ta1)} 72. Kc6 {6.23/7 10} Rd7 {8.25/25 10 (Le7)} 73. Rg2 {7.20/8 12} 1-0

On the flipside, when facing the same system FF quickly realizes the threat of closing the position and Black’s problem with the lack of space. It opts for a slightly worse position, but holds the game in style. Here is the game, without notes, just to illustrate the difference in approach.

[Event "Rapid match 20m+10s"] [Site ""] [Date "2019.10.24"] [Round "20"] [White "Stockfish 10 64 POPCNT"] [Black "Fat Fritz"] [Result "1/2-1/2"] [ECO "B04"] [Annotator "Elshan"] [PlyCount "289"] [EventDate "2019.??.??"] [TimeControl "1200+10"] 1. e4 Nf6 2. e5 Nd5 3. d4 d6 4. Nf3 {Last book move.} dxe5 {0.54/15 16} 5. Nxe5 {1.00/31 64 (c4)} c6 {0.52/14 31} 6. Be2 {1.05/29 17 (Sf3)} Bf5 {0.55/15 25 (Sd7)} 7. O-O {0.97/28 27} Nd7 {0.55/16 19} 8. Nf3 {0.95/33 43} e6 {0.54/15 26 (h6)} 9. c4 {1.03/30 13} N5f6 {0.54/14 12} 10. Nh4 {0.84/35 136} Bg6 {0.53/14 37 (Lxb1)} 11. Nc3 {0.85/29 73} Bd6 {0.54/14 37} 12. f4 {0.89/29 43 (c5)} Ne4 { 0.29/14 54 (Lf5)} 13. Nxg6 {1.01/28 9 (g3)} Nxc3 {0.22/13 38} 14. bxc3 { 0.91/29 19} hxg6 {0.21/12 14} 15. g3 {0.91/31 28} O-O {0.21/11 47 (Da5)} 16. Be3 {0.98/30 29 (a4)} c5 {0.13/10 50 (b6)} 17. Qd3 {0.85/30 42} Qc7 {0.15/11 75 (Tc8)} 18. Bf2 {0.95/31 16 (a4)} Rfd8 {0.18/10 46 (Tac8)} 19. Qf3 {1.02/31 42} b6 {0.23/10 32 (Tac8)} 20. a4 {0.89/32 61} Rac8 {0.20/12 21} 21. Kg2 { 0.85/32 12 (Tfb1)} f5 {0.24/11 30 (Le7)} 22. Bd3 {0.67/33 88 (Tab1)} cxd4 { 0.23/14 39 (Sf6)} 23. cxd4 {0.69/30 16} Bb4 {0.22/19 9} 24. Rfc1 {0.79/34 103 (Tab1)} Nf6 {0.06/11 43} 25. Rc2 {0.69/32 31} Kh7 {0.05/17 62 (Dd7)} 26. Rb1 { 0.68/31 43} a5 {0.05/23 9} 27. Rbc1 {0.68/34 48 (c5)} Qd7 {0.05/14 40} 28. Re2 {0.68/35 10} Ng4 {0.10/15 28 (Te8)} 29. Bg1 {0.82/28 11} Nf6 {0.10/14 11} 30. Bf2 {0.68/35 62 (h3)} Ng4 {0.06/14 60 (Te8)} 31. Bg1 {0.55/31 19} Nf6 {0.08/15 14} 32. Kh1 {0.68/34 65} Qxa4 {0.00/18 41 (Tc7)} 33. Rxe6 {0.00/33 25} Rd6 { 0.00/23 0 (b5)} 34. Rxd6 {0.31/31 26 (Te5)} Bxd6 {-0.04/24 56} 35. Qf1 { 0.39/35 38} Ba3 {-0.04/30 27 (Dc6+)} 36. Ra1 {0.94/28 11 (Tb1)} Qb4 {-0.04/18 91 (Dc6+)} 37. Rb1 {0.13/33 62} Qc3 {-0.04/24 0 (Dd6)} 38. Rxb6 {0.00/35 27} Re8 {-0.04/21 0} 39. Be2 {0.00/38 9} Bc1 {-0.03/22 109} 40. c5 {0.00/42 21} a4 {-0.03/22 0 (Le3)} 41. Rb1 {0.09/34 13} Be3 {-0.03/21 6} 42. Bb5 {0.09/38 11} Rd8 {-0.02/20 20 (Dxd4)} 43. Qc4 {0.00/42 29 (Dd3)} Qd2 {-0.02/19 49} 44. Qd3 { 0.00/42 10} Qxd3 {-0.01/20 11} 45. Bxd3 {0.00/44 31} Bxd4 {-0.02/18 23} 46. Ba6 {0.00/45 10 (Lc4)} a3 {-0.06/17 64} 47. c6 {0.00/44 11} a2 {-0.03/18 56} 48. Rf1 {0.00/46 10} Ne8 {-0.03/18 7} 49. Bxd4 {0.00/47 12} Rxd4 {-0.02/17 4} 50. Ra1 {0.00/48 11} Ra4 {-0.01/16 22} 51. Be2 {0.00/49 11} Nc7 {-0.01/16 13 (Kh6)} 52. h4 {0.00/45 10 (Ld1)} Kh6 {-0.08/12 83} 53. Kg2 {0.00/43 11 (Lf3)} Nd5 { -0.10/14 19 (Kh7)} 54. Kh2 {0.00/44 17 (Kf2)} Kh7 {-0.13/13 18} 55. Kh3 { 0.00/48 12} Ra3 {-0.15/14 14 (Kh6)} 56. c7 {0.00/47 12} Rc3 {-0.16/13 11} 57. Ba6 {0.00/49 9 (Txa2)} Nxf4+ {-0.09/14 48} 58. Kh2 {0.00/1 0} Rc2+ {-0.11/20 0 (Txc7)} 59. Kh1 {0.00/49 20 (Kg1)} Rxc7 {-0.14/17 13} 60. gxf4 {0.00/50 7} Rc2 {-0.12/16 6} 61. Bd3 {0.00/54 29} Rb2 {-0.11/14 10} 62. Kg1 {0.00/55 13} Kh8 { -0.09/14 32 (Tb3)} 63. Kf1 {0.00/48 19 (Kh1)} Rh2 {-0.08/13 18} 64. Kg1 { 0.00/49 40 (Ke1)} Rb2 {-0.09/13 9} 65. Kf1 {0.00/52 7} Kh7 {-0.06/12 9} 66. Ke1 {0.00/51 50 (Kg1)} Kh6 {-0.09/12 15} 67. Kf1 {0.00/52 37} Rh2 {-0.06/12 13} 68. Kg1 {0.00/49 4} Rb2 {-0.05/10 9} 69. Kf1 {0.00/51 6} Rh2 {-0.04/9 13} 70. Ke1 { 0.00/52 12 (Kg1)} Rh1+ {-0.09/10 16} 71. Bf1 {0.00/47 4} Rxh4 {-0.09/11 9 (Th2) } 72. Rxa2 {0.09/35 6} Rxf4 {-0.09/9 4} 73. Ra6 {0.09/42 8 (Tf2)} Kh7 {-0.10/9 16} 74. Be2 {0.09/46 7} g5 {-0.09/10 8} 75. Bh5 {0.09/46 7} g4 {-0.08/11 10} 76. Bg6+ {0.09/47 9} Kh6 {-0.05/10 5} 77. Bxf5+ {0.09/49 14} Kg5 {-0.04/9 25 (Kh5)} 78. Be6 {0.09/43 7 (Lc8)} g3 {-0.05/8 12} 79. Bd5 {0.09/42 7} Rf5 { -0.03/7 12 (Tf2)} 80. Bg2 {0.09/43 8} Rf2 {-0.02/7 10 (g6)} 81. Ra5+ {0.09/43 9 (La8)} Kh4 {-0.03/8 11 (Tf5)} 82. Ra4+ {0.09/46 15} Kg5 {-0.01/8 9} 83. Bf1 { 0.09/45 6 (La8)} Rh2 {-0.05/7 10 (Tf4)} 84. Be2 {0.09/45 9 (Ta7)} Rh1+ { -0.09/8 12 (Th4)} 85. Kd2 {0.48/42 6} Rh4 {-0.09/8 8} 86. Ra7 {0.48/43 8 (Ta1)} g6 {-0.07/8 11} 87. Ke3 {0.58/44 8 (Lf3)} g2 {-0.05/8 10 (Tf4)} 88. Ra5+ { 0.90/42 7 (Kf2)} Kh6 {-0.04/9 11 (Kf6)} 89. Kf2 {0.91/44 6} Re4 {-0.02/8 11 (Tb4)} 90. Bf3 {0.91/47 7} Rb4 {-0.02/7 10 (Td4)} 91. Kxg2 {0.91/47 8} Kg7 { 0.00/7 10 (Tb2+)} 92. Kg3 {0.91/43 9 (Ta2)} Kf6 {0.00/6 10} 93. Bd1 {0.91/42 7 (Ta2)} Rd4 {0.00/7 10 (Tb2)} 94. Bc2 {0.96/38 6 (Ta1)} Ke7 {0.00/7 10 (Td6)} 95. Bxg6 {1.08/39 7 (Ta2)} Rd8 {0.00/6 10 (Ke6)} 96. Be4 {1.08/43 9 (Ta7+)} Rd1 {0.00/6 10 (Tf8)} 97. Kf3 {1.09/48 14 (Kf4)} Rd8 {0.00/6 10 (Ke6)} 98. Bd5 { 1.09/46 6 (Kf4)} Kf6 {0.00/6 10 (Tf8+)} 99. Kf4 {1.10/43 10} Rd6 {0.01/7 7} 100. Rc5 {1.10/46 14 (Tb5)} Rb6 {0.01/7 11} 101. Bc4 {1.10/47 14 (Tc4)} Rb4 { 0.01/7 11 (Td6)} 102. Rc6+ {1.11/40 7} Ke7 {0.05/8 5} 103. Ke5 {1.11/43 64} Kd7 {0.08/8 13} 104. Rc5 {1.11/40 35 (Ld5)} Rb6 {0.05/8 8} 105. Bb5+ {1.11/44 19 (Ld5)} Ke7 {0.04/8 13} 106. Rc7+ {1.11/42 11} Kd8 {0.03/8 5} 107. Rd7+ { 1.11/41 10} Kc8 {0.04/8 7} 108. Rd5 {1.11/31 10 (La4)} Kc7 {0.02/8 15 (Kb7)} 109. Rd7+ {1.11/40 6 (Tc5+)} Kc8 {0.03/8 10 (Kb8)} 110. Rd5 {1.11/37 6 (Lc6)} Kc7 {0.02/7 13 (Kb7)} 111. Rc5+ {1.11/48 6} Kb7 {0.03/8 9 (Kd8)} 112. Bc4 { 1.11/35 19 (Lc6+)} Rh6 {0.03/7 9} 113. Bd5+ {1.11/41 5 (Le6)} Kb6 {0.04/7 9} 114. Rc8 {1.11/43 4} Rg6 {0.04/7 12} 115. Ra8 {1.11/40 21 (Kd4)} Rh6 {0.03/7 12 } 116. Rb8+ {1.11/44 11} Kc5 {0.02/7 6} 117. Rc8+ {1.11/43 11} Kb6 {0.05/8 14} 118. Rb8+ {1.11/36 4 (Kd4)} Kc5 {0.01/7 6} 119. Rc8+ {1.11/43 16} Kb6 {0.04/7 13} 120. Be4 {1.12/45 4 (Kd4)} Rh4 {0.02/7 12 (Ka5)} 121. Kd4 {1.12/41 5} Rh6 { 0.04/7 10 (Tf4)} 122. Rb8+ {1.12/44 5} Kc7 {0.02/7 5} 123. Rb7+ {1.12/42 23} Kc8 {0.06/8 13} 124. Rf7 {1.12/40 13 (Kc5)} Kd8 {0.04/7 9 (Th4)} 125. Kd5 { 1.12/40 9 (Ke5)} Ke8 {0.03/7 10 (Th4)} 126. Ra7 {1.12/36 9 (Tb7)} Rb6 {0.04/8 13 (Kd8)} 127. Bf5 {1.12/37 11 (Ke5)} Kf8 {0.07/8 10 (Tb5+)} 128. Be6 {1.12/34 10 (Ke5)} Ke8 {0.06/8 6} 129. Bf7+ {1.11/30 11 (Ke5)} Kd8 {0.06/8 14} 130. Kc5 {1.11/30 10} Rf6 {0.07/8 8} 131. Kd5 {1.11/29 10 (La2)} Rb6 {0.05/8 6} 132. Be6 {1.10/26 10 (Kc5)} Ke8 {0.04/8 9} 133. Bg4 {1.09/25 10 (Ke5)} Kd8 {0.04/7 17 (Kf8)} 134. Rd7+ {1.02/22 10 (Ld7)} Ke8 {0.02/7 4} 135. Rh7 {1.02/21 10 (Le6)} Kd8 {0.04/7 16} 136. Be6 {0.77/18 10 (Ld7)} Ke8 {0.04/8 10 (Tb5+)} 137. Ke5 { 1.02/30 10} Kd8 {0.05/8 6} 138. Rd7+ {0.09/34 10 (Ld7)} Kc8 {0.03/8 13} 139. Ra7+ {0.09/33 3 (Kd5)} Kb8 {0.02/7 8 (Kd8)} 140. Ra5 {0.09/39 17 (Ta4)} Kc7 { 0.01/7 11} 141. Ba2 {0.00/26 5 (Tc5+)} Rb2 {0.00/7 12 (Kb8)} 142. Bc4 {0.00/41 15 (Lg8)} Rh2 {0.00/6 10 (Tc2)} 143. Rc5+ {0.00/52 4 (Ta8)} Kb6 {0.00/4 9} 144. Rb5+ {0.00/65 5 (Tc8)} Kc6 {0.00/2 10 (Kc7)} 145. Bf1 {0.00/127 0 Draw agreed.} 1/2-1/2

Accuracy and technical skills

Even though it is not unusual for Fat Fritz to opt for a risky line, it still manages to hold its ground with a high degree of accuracy. One problem for Leela is that sometimes, without a good hardware, it misses some tactical paths, which can cloud its evaluation. While this is a more general problem for NNs not running on good hardware, it seems that FF is on its way to overcoming this hurdle. In the following game, FF manages its king quite well, and doesn’t get into any danger despite choosing a sub-optimal line.

[Event "Rapid match 20m+10s"] [Site ""] [Date "2019.10.24"] [Round "23"] [White "Fat Fritz"] [Black "Stockfish 10 64 POPCNT"] [Result "1/2-1/2"] [ECO "B09"] [Annotator "Elshan"] [SetUp "1"] [FEN "N4rk1/pp3pbp/4b1p1/8/4P3/2qB1Q1P/P1P1K1P1/1R5R b - - 0 19"] [PlyCount "48"] [EventDate "2019.??.??"] [TimeControl "1200+10"] {[#]} 19... Qc6 {0.00/39 40 [#] From this point onward, Fat Fritz demonstrates great and accurate defensive skills.} 20. Nb6 {0.22/26 65} axb6 {0.09/40 31} 21. Qe3 {0.22/31 0} Qc3 {0.00/39 14} 22. Qxb6 {0.19/32 37} Bc4 {0.00/42 21} 23. Bxc4 {0.15/32 29} Qxc4+ {0.00/45 141} 24. Kf3 {0.14/31 0} f5 {0.00/39 19 (Dc3+) } 25. e5 {0.16/20 174} Qe4+ {0.00/42 44} 26. Kf2 {0.15/22 0} Qf4+ {0.00/44 29} 27. Kg1 {0.12/23 62} Bxe5 {0.00/41 18} 28. Rd1 {0.12/22 0} Re8 {0.00/40 37} 29. Qf2 {0.12/22 0} Qa4 {0.00/40 26 (De4)} 30. h4 {0.12/16 96 (Df3)} Bb8 {0.00/42 21} 31. Re1 {0.10/19 116} Rd8 {0.00/44 38 (Tc8)} 32. Qf3 {0.09/14 50} Ba7+ { 0.00/42 14 (Dd4+)} 33. Kf1 {0.07/11 18} Rd2 {0.00/47 53 (Dc4+)} 34. Re2 { 0.09/17 54 (Db3+)} Rd1+ {0.00/45 29 (Txe2)} 35. Re1 {0.23/17 0} Qc4+ {0.00/45 24 (Db5+)} 36. Qe2 {0.05/9 33} Qf4+ {0.00/50 20} 37. Qf3 {0.03/7 23} Rxe1+ { 0.00/46 19 (Dc4+)} 38. Kxe1 {0.13/11 0} Qc1+ {0.00/51 21} 39. Qd1 {0.03/9 54 (Ke2)} Qe3+ {0.00/50 20} 40. Qe2 {0.01/7 31} Qc1+ {0.00/51 17} 41. Qd1 { 0.01/5 0} Qe3+ {0.00/52 49} 42. Qe2 {0.01/3 51} Qc1+ {0.00/58 85} 43. Qd1 { 0.00/2 0 Remis angenommen} 1/2-1/2

A Karpovian masterpiece

There are many good games in this match, and the Stockfish 10 wins are no less impressive, yet we are talking about FF, so the following positional masterpiece cannot go unnoticed. After obtaining a good positional advantage out of the opening, FF took his time until he realized his queenside breakthrough suffices, and he followed his play with a fantastic pseudo sacrifice. I will let the moves speak for themselves here:

[Event "Rapid match 20m+10s"] [Site ""] [Date "2019.10.25"] [Round "45"] [White "Fat Fritz"] [Black "Stockfish 10 64 POPCNT"] [Result "1-0"] [ECO "B90"] [Annotator "Elshan"] [PlyCount "135"] [EventDate "2019.??.??"] [TimeControl "1200+10"] 1. e4 c5 2. Nf3 d6 3. d4 cxd4 4. Nxd4 Nf6 5. Nc3 a6 {Last Book move} 6. f3 { 0.43/14 19} e5 {0.00/31 40} 7. Nb3 {0.40/16 19} Be6 {0.16/34 81 (Le7)} 8. Be3 { 0.41/15 26} Be7 {0.30/33 50 (h5)} 9. Qd2 {0.43/20 30} O-O {0.15/33 48} 10. O-O-O {0.46/24 25} Nbd7 {0.17/35 24} 11. g4 {0.47/23 18} b5 {0.05/32 8} 12. g5 {0.46/22 23} Nh5 {0.00/35 19} 13. Kb1 {0.47/21 26} Nb6 {0.00/36 41} 14. Na5 { 0.48/21 19 (Sd5)} Rc8 {0.00/37 36 (Dc7)} 15. a3 {0.51/21 58 (Sd5)} g6 {0.00/35 21 (Dc7)} 16. h4 {0.54/18 64} Qc7 {0.00/37 10} 17. Rh2 {0.51/21 32} Rfe8 { 0.00/39 18} 18. Nd5 {0.60/20 67} Nxd5 {0.00/41 18} 19. exd5 {0.58/27 0} Bd7 { 0.00/43 19} 20. Qb4 {0.44/22 104 (Lh3)} Bf5 {0.00/38 41 (Sg3)} 21. Nc6 { 0.79/15 36} Bf8 {0.00/39 28} 22. Bd3 {0.75/15 27 (Lh3)} Bxd3 {0.00/40 22} 23. Rxd3 {0.73/16 11} f5 {0.00/40 17 (Dd7)} 24. Rb3 {0.71/16 50 (Td1)} Ng7 { 0.00/39 51 (f4)} 25. Re2 {0.77/16 76 (Th1)} f4 {0.00/38 30 (Df7)} 26. Bf2 { 0.78/14 18} Qf7 {0.00/36 13 (Sf5)} 27. Qe4 {0.83/14 29 (Td2)} Nf5 {0.00/38 53} 28. Rd2 {0.85/16 22} Bg7 {0.00/35 52 (h6)} 29. Rd1 {0.94/14 40 (c4)} Nd4 { 0.00/38 16} 30. Rc3 {0.92/13 47 (Tbd3)} Rc7 {0.11/39 187 (Sf5)} 31. Rcd3 { 0.89/13 43 (a4)} Nf5 {0.00/40 19 (Tcc8)} 32. Rb3 {0.84/13 57} Rb7 {0.00/43 128 (Se7)} 33. Rb4 {0.99/10 28 (Sa5)} Rf8 {0.00/37 20} 34. Rb3 {0.96/11 23} Qd7 { 0.00/35 23 (Te8)} 35. Rc1 {0.94/12 57 (Tc3)} Re8 {0.00/36 22 (Sg3)} 36. Rb4 { 0.97/10 37 (Tg1)} Kh8 {0.00/36 41 (Tf8)} 37. Rg1 {1.07/10 25} Rc8 {0.00/36 23 (Df7)} 38. Na5 {0.80/16 44 (Th1)} Rbb8 {0.00/40 12} 39. Nc6 {0.81/22 9} Rb7 { 0.00/40 22} 40. Rc1 {0.75/22 46 (Th1)} Rf8 {0.00/38 47 (Se7)} 41. Rb3 {1.02/12 24 (Th1)} Rc7 {0.00/38 21 (Df7)} 42. Ka2 {1.04/11 34 (Dd3)} Rb7 {0.00/35 19} 43. Rg1 {1.01/11 32 (Kb1)} Rc8 {0.00/38 32 (Df7)} 44. Rc1 {1.01/12 26 (Th1)} Qf7 {0.02/37 104 [#] (Tf8)} 45. c4 {1.24/17 29} Ng3 {0.54/37 100 (bxc4)} 46. Qd3 {1.24/18 28} bxc4 {0.59/31 13} 47. Qxc4 {1.19/18 13} Rxb3 {0.51/35 21} 48. Qxb3 {1.19/14 14} Re8 {0.45/34 8 (Dd7)} 49. Bb6 {1.41/19 39} Qd7 {0.52/34 15} 50. Nd8 $3 {1.62/21 19 It is not only for the depth but the briliant idea behind this move: Black must take and then:} Rxd8 {1.80/33 48} 51. Bxd8 { 1.73/22 14} Qxd8 {3.61/32 15} 52. Qb7 {1.78/21 9} Qg8 {3.99/35 20} 53. a4 $1 { 1.86/22 21 Black's two minor pieces are rather clumsy. Black loses one way or another.} Nf5 {4.13/34 15 (Lf8)} 54. Qxa6 {2.73/14 43} Qxd5+ {4.22/33 4} 55. Qc4 {3.02/16 25} Qg8 {4.47/37 13} 56. Qxg8+ {3.19/16 17 (a5)} Kxg8 {0.45/1 0} 57. a5 {3.29/15 8} e4 {4.49/32 9} 58. a6 {3.33/14 19} Bd4 {4.34/34 6} 59. Rc4 { 3.41/13 8 Threatening Rxd4 followed by a7-a8} Bg1 {4.85/35 13} 60. Rxe4 { 3.46/12 26} Kf7 {5.07/37 22 (d5)} 61. Rxf4 {4.33/11 32} d5 {5.38/36 12} 62. Rb4 {4.62/11 17} Ke6 {5.78/37 10 (Sd6)} 63. Rb7 {4.78/11 21} Nxh4 {6.29/37 10} 64. f4 {4.77/12 10 (b4)} Ng2 {5.04/25 3} 65. Rxh7 {4.77/12 35} Nxf4 {7.18/30 18} 66. a7 {4.99/12 1 (b4)} Bxa7 {7.00/26 3} 67. Rxa7 {4.69/10 8} d4 {9.25/29 17 (Sh3)} 68. Ra5 {5.61/9 26} 1-0

And finally, game 49 is another example of why NNs are so strong in positional chess.

[Event "Rapid match 20m+10s"] [Site ""] [Date "2019.10.25"] [Round "49"] [White "Fat Fritz"] [Black "Stockfish 10 64 POPCNT"] [Result "1-0"] [ECO "C02"] [Annotator "Elshan"] [PlyCount "145"] [EventDate "2019.??.??"] [TimeControl "1200+10"] 1. e4 e6 2. d4 d5 3. e5 c5 4. c3 Nc6 5. Nf3 Qb6 6. a3 c4 {Last book move.} 7. Nbd2 {0.35/14 20} Na5 {-0.25/33 77} 8. Rb1 {0.35/14 22} Bd7 {-0.17/31 15} 9. Be2 {0.37/15 31} Ne7 {-0.32/33 30 (0-0-0)} 10. h4 {0.39/14 30 (0-0)} O-O-O { -0.27/33 66} 11. h5 {0.37/16 21 (Sf1)} Kb8 {-0.25/33 29 (f6)} 12. Nf1 {0.47/13 41} h6 {-0.19/31 23} 13. Bf4 {0.49/12 31} Nb3 {-0.11/32 57 (a6)} 14. Ne3 { 0.63/12 67 (S3d2)} Ba4 {-0.04/33 23 (Ka8)} 15. Nd2 {0.70/12 32} Nc6 {0.00/31 30 (Sxd2)} 16. Nxb3 {0.70/13 35 (Lg3)} Na5 {0.00/34 72} 17. Kf1 {0.65/13 66 (Lg3)} Bxb3 {0.00/40 26 (a6)} 18. Qd2 {0.68/14 20} Ba4 {0.00/39 11} 19. Bd1 { 0.67/13 20 (De1)} Bxd1 {0.00/35 17} 20. Rxd1 {0.66/12 25} a6 {0.00/37 16 (Le7)} 21. Rh3 {0.64/10 92 (Lg3)} Be7 {0.00/37 36} 22. Kg1 {0.62/11 41 (Lg3)} Rdg8 { 0.00/35 13} 23. Bh2 {0.59/11 36 (De2)} Qd8 {0.00/35 34 (Ka7)} 24. Qe2 {0.56/10 43 (Lg3)} Ka7 {0.00/37 39} 25. f4 {0.56/11 19 (Df3)} g6 {0.00/29 16} 26. g4 { 0.62/12 24 (Dg4)} Nb3 {0.00/33 56 (Sc6)} 27. Kh1 {0.72/11 42} Qe8 {0.00/38 41 (Dd7)} 28. Qf3 {0.73/14 31} Bd8 {0.00/37 35 (Th7)} 29. hxg6 {0.77/18 34} Rxg6 { 0.09/32 31 (fxg6)} 30. f5 {1.86/13 33 (Lg3)} Rg7 {0.00/38 34 (Tgg8)} 31. f6 { 1.89/13 24 (Lf4)} Rg6 {0.97/34 66} 32. Bf4 {1.89/11 22} Bc7 {1.00/35 28} 33. Ng2 {1.95/11 53 (Kg2)} Qg8 {1.34/26 82} 34. Rh4 {1.96/11 20} Qd8 {1.82/30 158 (Dh7)} 35. Be3 {2.27/11 49 (Kh2)} Na5 {2.46/30 84 (Dg8)} 36. Rd2 {2.40/11 39 (Sf4)} Nb3 {2.44/32 17 (Sc6)} 37. Rf2 {2.57/11 39 (Td1)} Rgg8 {3.38/30 83 (Dg8) } 38. Bxh6 {2.89/10 41 (Sf4)} Qd7 {3.65/26 36 (Lb6)} 39. Be3 {3.16/9 51 (g5)} Rxh4+ {4.29/30 54 (Sa5)} 40. Nxh4 {3.37/10 28} Rh8 {4.55/29 14} 41. Rh2 { 3.57/10 27} Na5 {5.06/34 56 (Da4)} 42. g5 {3.92/9 43 (Sg2)} Nc6 {5.17/32 22} 43. Ng2 {4.05/9 36} Rg8 {5.35/32 15 (De8)} 44. Rh7 {4.21/8 49} Nd8 {5.45/31 15 (Lb6)} 45. Qd1 {4.33/8 32 (Tg7)} Qb5 {4.48/27 19 (De8)} 46. Qc2 {4.56/8 24 (De2)} Qb6 {5.14/32 58 (De8)} 47. Kg1 {5.08/8 45 (De2)} Kb8 {5.43/31 29 (Lb8)} 48. Kf1 {5.45/8 35 (De2)} Ka7 {5.47/29 15} 49. Kf2 {5.48/8 25 (Kg1)} Qc6 { 6.00/29 16 (Sc6)} 50. Nh4 {5.36/7 26} Rf8 {6.40/31 10 (De8)} 51. Nf3 {6.20/7 21 (g6)} Rg8 {6.60/29 10 (De8)} 52. Kg3 {5.99/7 26 (Sh4)} Ba5 {5.51/26 10 (De8) } 53. Bf4 {6.00/7 23 (Sh4)} Qb5 {5.56/29 10 (De8)} 54. Rh6 {5.74/7 21 (Sh4)} Rf8 {5.74/29 10 (De8)} 55. Rh7 {5.85/7 19} Rg8 {5.51/28 5} 56. Bd2 {5.44/7 18 (Sh4)} Bc7 {5.61/28 10 (De8)} 57. a4 {5.43/7 19 (Sh4)} Qc6 {5.20/24 7 (De8)} 58. Be3 {5.43/7 17 (Sh4)} Kb8 {5.66/29 18 (De8)} 59. Kg4 {5.44/7 16 (Sh4)} b5 { 5.38/24 7 (De8)} 60. axb5 {6.18/8 9} Qxb5 {6.24/28 12 (axb5)} 61. Rh2 {6.69/7 18 (Sh4)} Qe8 {6.27/28 10 (a5)} 62. Kg3 {6.81/7 15 (Dh7)} a5 {5.58/24 4 (Th8)} 63. Rh7 {5.96/7 15 (Dh7)} a4 {5.34/25 6} 64. Rh1 {6.03/7 9 (Th6)} Ba5 {5.60/28 19 (Db5)} 65. Ra1 {7.48/7 11 (Th6)} Bb6 {5.85/24 11 (Th8)} 66. Qxa4 {11.19/7 11 (Txa4)} Qxa4 {5.22/28 2} 67. Rxa4 {11.46/8 8} Kc7 {6.13/34 18 (Sc6)} 68. Ra1 {16.83/7 22} Rh8 {6.59/31 7 (Sc6)} 69. Rg1 {15.38/7 13 (g6)} Kc6 {6.30/31 8 (Kd7)} 70. Rb1 {14.77/7 11 (Tf1)} Kd7 {6.68/31 15 (Lxd4)} 71. Bf4 {14.01/7 11 (Tf1)} Ba5 {6.65/31 9 (Lxd4)} 72. Ra1 {13.33/7 11 (g6)} Bc7 {6.69/34 5 (Lxc3)} 73. Kg4 {13.55/7 11 (g6)} 1-0

Finally, here is the complete set of games:

Replay all games Fat Fritz vs Stockfish 8

 

Replay all games Fat Fritz vs Stockfish 10

 

Answering reader questions

There are obviously further experiments to run, and all top engines are under continuous development. Thanks for all your suggestions and feedback in the comments to part one.

To clarify what the "the Silver Openings Suite" is, as Albert Silver explained:

It is comprised of 50 positions, which are designed to cover the basic range of openings, and opening structures, in proportions roughly comparable to their popularity in Mega.

The match uses these 50 positions, all of which are played twice with colors reversed the second time. In other words, for every position Stockfish had white in one and black in the other, and same goes for Fat Fritz.


Check back for more exciting Fat Fritz news in the coming days!




Elshan Moradiabadi is a GM born and raised in Tehran, Iran. He moved to the US in 2012. Ever since, he has been active in US college chess scenes and in US chess. is a veteran instructor and teaches chess to every level, with students ranging from beginners to IM. He can be contacted for projects or teaching.
Discussion and Feedback Join the public discussion or submit your feedback to the editors


caliche2016 caliche2016 11/10/2019 03:26
Does anybody (besides Fat Fritz devs) really care if FF can or cannot beat Stockfish? I think the real test is beating Leela Chess Zero in finding the best move/plan in complicated strategic positions. If it can't beat LCZero, then this FF argument is basically over, because LCZero is FREE. Why don't you demostrate instead that your engine is stronger than LCZero using exactly the same hardware? That and only that would make a reasonable chess player want to buy it.
richardward64 richardward64 11/9/2019 10:30
Is Fat Fritz disadvantaged by interfacing with a GTX 1660 Ti gpu/graphics card? Or is RTX much better?
SunriseK SunriseK 11/9/2019 01:28
@sovaz1997: sorry, I suppose what I said about AlphaZero or FF is also valid for NN LeelaChess.
Just I don't know that software enough. At first I played some games with it (when it still was not too strong; so in a manner, I even gave my little contribution to the program). But then at a certain moment, the developers withdrew it from the website, thus making impossible for me to furtherly deepen my knowledge of it. I was not very happy with that outcome of course...
And I also believe the "Zero"attribute should not be used, because it means the NN is learning just by playing against itself, like AlphaZero.
Keshava Keshava 11/9/2019 01:08
@Karbuncle
They could still test it at c----.com/cccc
if they wanted to know the objective truth.
Albert Silver Albert Silver 11/8/2019 04:10
@princesserasta - https://en.chessbase.com/post/test-your-engines-the-silver-openings-suite
Tom Box Tom Box 11/8/2019 03:29
I think Chinguaro meant ChessBase did not write articles praising Stockfish, not that ChessBase did not support the engine.
KevinC KevinC 11/8/2019 12:38
chinguaro 11/7/2019 10:29
And suddenly Stockfish exists for chessbase ... before Fat Fritz arrived it didn't exist. Now we have to read day after day that Fat Fritz is the best chess program in history...

What are you talking about??? I have been using Stockfish with ChessBase for quite a few years now. you create a UCI engine, and it works fine.
Karbuncle Karbuncle 11/8/2019 12:22
Albert, I believe I owe you an apology. The TCEC rules for uniqueness are allowing for ambiguous interpretation. The rejecting of Fat Fritz is based on the fact that it uses the Lc0 binary. However, this does not take into account developers using their own unique training nets. It would seem there's room for debate on this aspect of the TCEC rules in my opinion. I think everyone involved in running TCEC should discuss and consider the merits and conditions of Lc0 binaries using different NNs since they will most likely play differently.
Marselos Marselos 11/8/2019 12:00
TCEC is relative.
We talk about technological eras and records.
Friyz 5.32 is a record.
Deep Fritz 10 is a record, it beats Kramnik.
Deep Thought, and Alpha Zero are primates.
Leela, objectively, is a reflection, it does not mark an era.
The analysis is contextual, so the match concerns, directly,Alpha Zero.
Karbuncle Karbuncle 11/8/2019 11:50
I have to retract my suggesting about having Fat Fritz entering TCEC for an ubiased test of strength. I was informed TCEC has a rule about engine clones, and Dues X and Fat Fritz are considered too similar to Lc0 (parent code they make use of) with not enough uniqueness to make them stand out. It's a fair rule, because otherwise you'd have the roster filled with 5 million clones of Lc0. So in that regard, Lc0 will have to stand as the top NN engine entry in the TCEC until these clones develop a more unique adaptation and additional improvements.

With that said, I still find it fascinating that Stockfish can only seem to win tournaments held by unbiased 3rd parties. Funny how that works out...
PrincesseRasta PrincesseRasta 11/8/2019 11:48
Ok, this might be a little off-topic for you but I'd like a link to get this Silver openings suite to match my own engines. Can't find it.
Best regards.
sovaz1997 sovaz1997 11/8/2019 08:30
@SunriseK And not a word about Leela...
Marselos Marselos 11/8/2019 07:14
Stockfish is tired, it is stop at the version 10.
Ok, good work. Engine guaranteed.
Put it on a beautiful record and send it to my home.
Make a price ...
Scorpion29 Scorpion29 11/8/2019 03:41
I guess SunriseK has properly summed up what Stockfish is!
SunriseK SunriseK 11/8/2019 01:43
There's one thing I particularly appreciate about this FF vs SF match: it has clearly and finally shown that SF is not so strong as many people (especially patzers all around the world!) believe; unless you are looking at the finger instead of the moon (as in fact you can see in various comments, also here).
I'm very happy to see finally come true what I was already thinking since many years ago: all conventional chess engines don't merit to be considered "AI" because they are just blind silly robots (only able to use brute force and also very questionable methods of choosing their "best move", which tipically become even ridiculous when coming to the endgame, unless they fall into tablebases of course - another blind silly method btw). Obviously their software cannot change (whatever it happens during the game) and of course they have zero real knowledge about chess; this is very easily spotted for example by looking at the sequence of moves they think should follow their "best move": it's just a pseudo - random sequence, with no real plan behind. This stupid method is only effective because of the amazing improvements in the hardware's speed, but every sentient being should find this fact as simply disgusting and outrageous!
Now finally, after AlphaZero (and FatFritz too, as it seems), we are breathing fresh air again!
And it should be obvious to everyone that the future of chess engines (and not only) stands in the neural networks, which are acting like the human brain, changing their virtual connections and learning from their errors; thus finally reaching a real human - like understanding of the game, real strategic plans (as we have seen in some games commented by GM Moradiabadi) and a real intelligence level, worth to be truly named AI.
sahista1981 sahista1981 11/8/2019 01:32
I kind of agree with Karbuncle on this. The whole Fat Fritz vs Stockfish thing should have been organised by a third party organizaton like TCEC, simply in order to avoid any potential issues involving conflict of interest so that the final results could be taken seriously. That's just my opinion.
Albert Silver Albert Silver 11/8/2019 01:01
@Karbuncle - You should take the time to read what I wrote below.
Karbuncle Karbuncle 11/8/2019 12:55
Here are the simple facts: 1. People that actually follow the latest tournaments in engine chess don't care about matches with old builds. Cutting edge is what interests fans in this field. 2. On that very basis of cutting edge, a recent dev build defeated AllieStein AND Leela Chess Zero in the most recent TCEC knockout cup. 3. SF also won season 16 ahead of both Leela Chess Zero and Allie Stein, and most recently won the Fisher Random Chess tournament on TCEC. It is now currently DOMINATING AllieStein in a direct Fisher Random Chess tournament. 4. Everyone is cutting jokes about how Fat Fritz is a clone of Lc0. So like others have said, if you want to PROVE your software is the best, enter it in season 17 of the TCEC and see how far you get. Otherwise, people will continue to dismiss this as shilling without any real substantive comparisons. Bring it to the cutting edge table.
Albert Silver Albert Silver 11/7/2019 11:10
@mamago92 - While true in the strictest sense, it presents a misleading picture of how it works. Fat Fritz (and Leela obviously) has a huge file of values which represent its chess knowledge. The CPU is used, but only 2 cores for a single GPU (3 cores if more), and for every single position it searches, it asks the GPU to read the NN file and give it the evaluation. The GPU does not do any actual calculations, that is still just the CPU. If you give FF more than two cores, such as 6 or 8, the performance is actually *worse*. Stockfish on the other hand just gets stronger and stronger with more and more cores, while Fat Fritz does not. It only really gets stronger with a faster GPU so that it can read the NN file faster and boost its NPS. So it is more correct to say they are really just very different.
Albert Silver Albert Silver 11/7/2019 11:01
So in the order:
@Karbuncle - I replied to this in the feedback to Part One: I do not question the developer build is stronger, but there are new developer versions almost every day. When Stockfish 11 is released, and I do not doubt it will be a wonderful upgrade, it will be tested.
@LarryK - It trolls some, no question, but how much I cannot say.
@Raoul - Long story short: TCEC is the one who forbade the participation of Deus X and not the other way around. After an interesting first campaign, the rules of engagement were changed to prevent further participation. Take it up with them.
@LuckyLuke - (BTW, I loved those comics as a kid) Your viewpoint is very much how I see it. If Fat Fritz were just the exact same thing but 20 Elo better, it would be incredibly boring. I enjoy the difference in style and think this will provide players with wonderful contrasting analysis to explore positions and openings.
mamago92 mamago92 11/7/2019 10:58
FF uses CPU plus GPU? Stockfih uses only CPU?
sovaz1997 sovaz1997 11/7/2019 10:32
Fat Fritz does not exist. This is just a clone of Leela
chinguaro chinguaro 11/7/2019 10:29
And suddenly Stockfish exists for chessbase ... before Fat Fritz arrived it didn't exist. Now we have to read day after day that Fat Fritz is the best chess program in history...
LuckyLuke LuckyLuke 11/7/2019 09:41
Thx for the analysis and the games. Fat Fritz is stronger than Stockfish 8 and seems to be slightly stronger than Stockfish 10. The rating gap of your test hints that the development version of Fat Fritz is slightly weaker than the development version of Stockfish but further tests are needed to confirm this. But the truth about this shouldn't be so important because the engines are very strong and it's nice to have strong engines of different playing styles. I would like to see a match between Lc0 and Fat Fritz to compare them directly. Not so much to compare their strength but what makes Fat Fritz an engine "from an entirely different plane of existence" if so.
RaoulBertorello RaoulBertorello 11/7/2019 09:02
If you are so sure about Fat Fritz superiority against Stockfish, why don’t you, Fat Fritz makers, take part into next TCEC competition ? That would be the best way to clear the field from any dispute about which engine is stronger. TCEC plays about thrice a year, and I think it’s very easy for professionals like you to join it: just think what amateurs like Stockfish keepers have been able to achieve so far. Any reason put forward to justify Fat Fritz absence from that contest would be taken as an admission of obvious inferiority.
LarryK LarryK 11/7/2019 07:42
Question: Does Fat Fritz solve the "trolling" behavior of Lc0, meaning that when Lc0 gets an obviously winning position, it plays silly moves as if trying to make the game take as long as possible. Does Fat Fritz wrap up winning games quickly like all alpha-beta engines do, or does it act like Lc0?
Karbuncle Karbuncle 11/7/2019 07:13
You need to do these matches with the latest Stockfish builds. They are CONSIDERABLY stronger than even the official Stockfish 10, which is a year old now. I learned this myself the hard way, when I discovered some pretty nasty blind spots in SF 10, which the dev builds do not have a problem with.

Try the latest build from here: http://abrok.eu/stockfish/
1