GM preparation with Fat Fritz

by Frederic Friedel
9/12/2019 – It is inspired by the remarkable Google/DeepMind development that shocked the world: AlphaZero. Using the open source code of Leela we generated a neural network based on millions of auto-play games — but also added millions of high level human and computer-vs-computer games, and endgame tablebases. The result: a program that strong GMs are experimenting with — and they are fascinated with the ideas our engine produces. Even very young chess geniuses are fans of Fat Fritz. | Photo Frederic Friedel

ChessBase 17 - Mega package ChessBase 17 - Mega package

ChessBase is a personal, stand-alone chess database that has become the standard throughout the world. Everyone uses ChessBase, from the World Champion to the amateur next door. It is the program of choice for anyone who loves the game and wants to know more about it.


Nihal and Srinath

A while ago we had a visitor, one of my favourite Young Super-Talents, Nihal Sarin. The lad was fourteen and already equipped with a 2600+ Elo rating. Nihal had been to my house two years ago, and we had a lot of fun with him. This time was no different: fun, pranks and great discussions. He had his second, GM Srinath Narayanan, who is competing to be be one of the nicest people in chess. For the weekend I planned outings with the grandchildren, whom Nihal knew from the first visit, to the forest and maybe the Baltic Sea.

Srinath Narayanan and Nihal Sarin

However, on the Friday evening I made the fatal mistake of giving the two access to a developmental version of Fat Fritz. They accessed it on the engine cloud and then sat there, glued to the screen, trying line after line, position after position, with the neural network engine.

At some stage I had to say: "Guys, it's close to midnight, we have a long day ahead of us," and sent them off to bed. But the next morning had me calling around cancelling the outings. No way these GMs were going to go out to feed ducks and goats, or wade into the sea; they wanted more, more, more...

From then on, Srinath and Nihal spent at least six hours a day going through games with the new neural net engine. Breakfast, lunch and dinner were taken with the computer running next to them.

Typically they would have two engines on, the Fat Fritz development version and Stockfish or Komodo. There were many positions where the programs would deviate — showing different moves or substantially different evaluations. Later I asked Srinath to send me some examples, and he wrote:

I scoured around for the positions. We have a lot of ideas generated with the NN engine. However, a lot of them haven't yet been used in practice, and for obvious reasons I want to reserve it for Nihal to score points. We have had a bit of an advantage here, which I expect to last for the next 6-12 months with the opposition that we are expected to face. With that in mind I have attached one position...

Here's the new line Srinath sent us. It represents an improvement they found in a game that Nihal had lost:

[Event "TCh-TUR Super League"] [Site "Konya TUR"] [Date "2019.07.17"] [Round "1.3"] [White "Demchenko, A..."] [Black "Nihal, Sarin"] [Result "1-0"] [ECO "C50"] [WhiteElo "2665"] [BlackElo "2610"] [Annotator "Srinath,N"] [PlyCount "137"] [EventDate "2019.07.17"] [EventType "team"] [EventRounds "13"] [EventCountry "TUR"] 1. e4 e5 2. Nf3 Nc6 3. Bc4 Bc5 4. d3 Nf6 5. O-O O-O 6. Re1 {Literally the only move we hadn't deeply prepared against before this game. The Turkish League was one of the few tournaments where Nihal was at the receiving end of the negative side of chance.} Ng4 7. Re2 Kh8 8. h3 f5 9. Bg5 Nf6 10. exf5 d5 11. Bb5 {[#] This is the critical point where Black has to decide between Nd4 and e4. While a brute force engine would point towards e4, a NN engine points toward Nd4 right away!} Nd4 (11... e4 {A brute force move which is very concrete.} 12. dxe4 dxe4 13. Qxd8 Rxd8 14. Bxc6 exf3 (14... bxc6) 15. Bxf3 Rd1+ 16. Kh2 Bxf5 17. Bf4 Bd6 18. Bxd6 cxd6 $11 {is -0.00 according to Stockfish. For good reason, as the position is quite close to equal. However, it is not easy to get here. It is also not a draw yet, as World Champion Magnus has been teaching the chess world over the last decade. The game could continue} 19. Rd2 Rc1 20. Rd1 Rxc2 21. Rxd6 Rxb2 22. Nd2 {A 0.00 like position where I would hold very confidently in a correspondence game. In an OTB game, I wouldn't be completely sure against a very strong opponent. Such equalish positions can very easily turn into suffering.}) 12. Nxd4 Bxd4 (12... exd4 $1 {A very deep idea which the NN engine sees instantly. Brute force engines also get there, but you have to bring them to it and wait for them to comprehend. This used to be the way we found hidden ideas when analyzing with powerful brute force engines - make moves ourselves and let the engine catch up. However, with a NN engine, the percentage of such 'blind shots' increase substantially, and it also saves a lot of time. White's b5 bishop is totally sidelined now and Black gets full compensation due to the much better functioning of his pieces, despite being a pawn down and the structure being unusual. This exact position from the opening is not very likely to reoccur, but I was very impressed by this unusual strategic idea. These are the building blocks of evolution of human chess understanding.} 13. g4 c6 14. Ba4 a5 15. a3 Qd6 {Black's compensation is clear and the position frightening. If this happened over the board, Black has excellent practical chances.}) 13. c3 Bb6 14. Rxe5 c6 15. Ba4 Bc7 16. d4 $1 $14 Qd7 17. Bxf6 Rxf6 18. g4 Bxe5 19. dxe5 Rf8 20. Nd2 Qc7 21. Nf3 Qe7 22. Bc2 Bd7 23. Qd2 Rae8 24. Re1 Bc8 25. e6 Qf6 26. Ne5 Kg8 27. f4 Bxe6 28. g5 Qe7 29. fxe6 Qxe6 30. Kg2 Qd6 31. h4 g6 32. Kg3 Re7 33. b4 a6 34. c4 Rd8 35. cxd5 Qxd5 36. Qxd5+ cxd5 37. Bb3 Kg7 38. Rc1 d4 39. Kf3 Ra8 40. Ke4 Rd8 41. h5 d3 42. h6+ Kh8 43. Rd1 d2 44. Ke3 a5 45. bxa5 b6 46. a6 b5 47. Rxd2 Rxd2 48. Kxd2 Ra7 49. Nf7+ Kg8 50. Kc3 Rxa6 51. Kb4 Rb6 52. Kc5 Rb8 53. Bd5 b4 54. Bc4 Rc8+ 55. Kxb4 Rb8+ 56. Kc3 Ra8 57. Bb3 Rc8+ 58. Kb2 Rb8 59. Kc3 Rc8+ 60. Kd4 Rb8 61. Ke5 Rb5+ 62. Bd5 Ra5 63. Ke6 Ra6+ 64. Ke7 Ra4 65. f5 Ra5 66. Be6 gxf5 67. Nd6+ Kh8 68. Kf8 Ra8+ 69. Bc8 1-0

Srinath adds:

Brute force engines also get [to the line], but you have to bring them to it and wait for them to comprehend. This used to be the way we found hidden ideas when analyzing with powerful brute force engines — make moves ourselves and let the engine catch up. However, with a NN engine, the percentage of such 'blind shots' increase substantially, and it also saves a lot of time...These are the building blocks of evolution of human chess understanding.

Srinath sent us a second example of his work with a Fat Fritz prototype.

I have shown the annotations given below to the Indian team, who preferred to have some of the comments about our preparation removed. Since it's not quite relevant to the topic, I removed them and kept only the parts dealing with neural network and brute force engines. The other comments were too sensitive to reveal to the general public."

[Event "12th World Teams 2019"] [Site "Astana KAZ"] [Date "2019.03.05"] [Round "1.1"] [White "Adhiban, Baskaran"] [Black "Grandelius, N...."] [Result "1-0"] [ECO "C50"] [WhiteElo "2683"] [BlackElo "2694"] [Annotator "Srinath,N"] [PlyCount "61"] [EventDate "2019.03.05"] [EventType "team"] [EventRounds "9"] [EventCountry "KAZ"] [SourceTitle "The Week in Chess 1270"] [Source "Mark Crowther"] [SourceDate "2019.03.11"] [SourceVersion "1"] [SourceVersionDate "2019.03.11"] [SourceQuality "2"] [WhiteTeam "India"] [BlackTeam "Sweden"] [WhiteTeamCountry "IND"] [BlackTeamCountry "SWE"] 1. e4 e5 2. Nf3 Nc6 3. Bc4 Nf6 4. d3 Bc5 5. O-O d6 6. c3 h6 7. Re1 O-O 8. Nbd2 a5 9. Nf1 Be6 10. Bb5 Ne7 11. d4 exd4 12. cxd4 Bb6 13. Ng3 d5 14. e5 Ne4 15. Nxe4 dxe4 16. Rxe4 {[#]} c5 {The position is evaluated close to the 0.00/0.15 range by a brute force engine. In practical terms it is a lot easier to play as White. Therefore, the NN based engine's evaluation is much more useful in practical terms.} (16... Qd5 17. Bd3 Bf5 {Nihal's game proceeded in this direction against Oleksiyenko in the French League. The game eventually ended in a draw, but despite being slightly outprepared, by a few moves, White was more comfortable and pressing throughout the game.} 18. Rh4 Qe6 19. h3 Rad8 20. Qe2 Bxd3 21. Qxd3 Ng6 22. Rg4 c5 23. Be3 cxd4 24. Bxd4 Bxd4 25. Rxd4 Rxd4 26. Qxd4 Re8 27. Re1 Qxa2 28. Qd7 Qe6 29. Qxb7 Rd8 30. Qe4 Qd5 31. g3 Qxe4 32. Rxe4 Rb8 33. Re2 Rb4 34. h4 Nf8 35. h5 Ne6 36. Kg2 Kf8 37. Nh4 Nd4 38. Rd2 f6 39. exf6 gxf6 40. Ng6+ Kf7 41. Nf4 Ne6 42. Nd5 Rb5 43. f4 f5 44. Kh3 Rb3 45. Kh4 a4 46. g4 fxg4 47. Kxg4 Nc5 48. Re2 Rd3 49. Nb4 Rd6 50. Re5 Nd7 51. Ra5 Rd4 52. Rxa4 Ne5+ 53. Kf5 Nc6 54. Nxc6 Rxa4 55. b4 Ra1 56. Ke5 Rh1 57. b5 Rxh5+ 58. f5 Rh1 59. b6 Re1+ 60. Kf4 Rf1+ 61. Kg4 Rg1+ 62. Kf4 Rf1+ 63. Ke4 Re1+ 64. Kf4 { 1/2-1/2 (64) Nihal,S (2598)-Oleksiyenko,M (2595) Brest FRA 2019}) 17. Bd3 Bf5 18. Rf4 Bxd3 19. Qxd3 cxd4 20. Rg4 Qd7 21. h3 Qe6 22. Bd2 Rad8 23. Re1 Nf5 { A powerful brute force engine remains contantly at 0.00 here. However, Fat Fritz hardly requires a second invitaiton to the feast here. Chess is of course 0.00 with perfect play. However, the practical chess played by humans is quite different to engine games played at a 3600 level. Humans have to solve problems that are incredibly complex within limited time. I don't think we'll ever get close to that level of perfection, and so I think chess will remain fun for a long time.} 24. Bg5 $1 hxg5 25. Nxg5 Qc8 26. Rf4 g6 27. g4 { [#] A very interesting position to study/ analyze with or without an engine.} Ne7 {The typical human response to pressure. Nils also had the added pressure of his team's position collapsing on the other boards. But even without external factors, defending this kind of a position against Adhiban is an unpleasant task for any human being.} (27... Ne3 28. e6 Qc6 $1 {This defence found by Stockfish is 'missed' by the NN engine. This actually has a rather useful practical point in the sense that the NN engine can indicate better what our human GM opponent is 'likely to miss' and set them complex practical problems.} (28... Qc2 {is a more natural human defensive move, but falls to} 29. exf7+ Kh8 (29... Rxf7 30. Qxc2 Nxc2 31. Nxf7 Nxe1 32. Nxd8 Bxd8 33. Kf1 Nc2 34. Ke2 $18) 30. Rxe3 Qc1+ 31. Re1 Qxe1+ 32. Kg2 Qc1 33. Qg3 $40 {Fat Fritz thinks White has a rather strong attack, whilst Stockfish thinks that White is just winning.}) 29. exf7+ Kh8 {[#] The point. g6 needs to be defended. The position remains complex with engines unanimously agreeing on the universal evaluation '0.00' which humans have referred to as 'unclear' for decades before the introduction of the weird, but now accepted centipawn evaluation system.}) 28. Qg3 Kg7 29. Qh4 Rh8 30. Rxf7+ Kg8 31. Rh7 {A fine game. The last decade saw a huge difference in how we look at openings. With brute force engine showing everything as 0.00, it was considered that White has no objective advantage anymore and it is 0.00 everywhere. The aim with White was to 'just get a position' which doesn't immediately draw. While this is largely still true from a strictly objective vantage, from a practical point of view, with the emergence of NN engines, I believe White can lay claim to 'edges' once again. Until the next rotation of the cycle.} 1-0

In conclusion Srinath writes:

"A fine game. The last decade saw a huge difference in how we look at openings. With brute force engine showing everything as 0.00, we considered that White has no objective advantage anymore and it is 0.00 everywhere. The aim with White was to 'just get a position' which doesn't immediately draw. While this is largely still true from a strictly objective vantage, from a practical point of view, with the emergence of NN engines, I believe White can lay claim to 'edges' once again. Until the next rotation of the cycle."

Currently Srinath and Nihal are in Khanty-Mansiysk, where Nihal is playing in the FIDE World Cup 2019. In the first round he was paired against the newly crowned Ibero-American Champion Jorge Cori, 66 points higher on the rating scale. Nihal knocked Jorge out seemingly without much effort and proceeds to round two. Take a look at the remarkable first game and draw your own conclusions.

Where and how to use Fat Fritz

Just like all the other engines available online for analysis, Fat Fritz can be found in the ChessBase Engine Cloud.  There are several quick ways to do this from within ChessBase 15, or Fritz program (such as Fritz 16, Komodo, or Houdini).

Another very easy way is just to open a game or board position and start an engine with Add Kibitzer.

Add kibitzer

When you do this you get a list of the engines you can open, all in your computer, but under the list is the Cloud button. Just click on it and this way too will take you to the Engine Cloud window.

Load engine

You will then be shown the Engine Cloud window with the list of all the engines available for use.

Engine list

Click or tap to enlarge

Just look for the name Fat Fritz and choose it.

Fat Fritz

For more detailed information, don't hesitate to look at the full guide.

Nihal Sarin Brand Ambassador of Akshayakalpa Organic Milk

Akshayakalpa, India’s first Organic Milk Company, announces the 15-year-old Chess Grandmaster Nihal Sarin as its Brand Ambassador to further the brand Akshayakalpa as the fast-rising young organic milk company on the global stage.

Nihal is the youngest Indian ever and third youngest player in history to break the 2600 barrier in chess. This is considered to be among the biggest sponsorship deals for an Indian chess player after Viswanathan Anand.

Signing the contract

Mr. Shashi Kumar, Co-founder and Managing Director Akshayakalpa, said:

We are happy to announce Nihal Sarin as the new face of Akshayakalpa Organic Milk. What separates Akshayakalpa is the superior milk procurement process we have which results in organic milk that retains its original nutrients. This is key for the cerebral and physical growth of young children. Nihal is a great way for us to communicate this. His commitment to excellence on and off the board makes him an ideal youth icon. He embodies the spirit of perseverance, mental discipline, and natural flair, which aligns with the qualities of Akshayakalpa Organic Milk.

Akshayakalpa Organic Milk products’ hallmark has been the superior process the organization follows for procurement of milk. Zero antibiotics and growth hormones are used, and the fodder is grown by using pure organic methods. This results in the milk products retaining all the essential nutrients as nature intended, which is key to a young child’s brain development, chiefly Vitamin B12 in the Milk, and Omega 3 Fatty Acids in the Ghee. In the process, the company directly pays Akshayakalpa farmers 69% of the price customers pay for its Milk Products.

Excited about joining hands with the company, Nihal Sarin commented. “I was surprised to learn that the food that we eat costs more to produce than what we pay to the farmers. Akshayakalpa is a model that can change that. In the process of procuring Organic Milk, the company is making real difference to the lives of young children like me. I am honoured to be Akshayakalpa’s Brand Ambassador and will strive to help the company transform lives by spreading its brand on the global stage by doing what I love to do – play chess.”

About Akshayakalpa

Akshayakalpa Organic Milk, founded by Shashi Kumar in 2010, is the first Organic Milk brand in India. The company manufactures, markets and sells Organic Milk products. The organic products include Farm Fresh Milk, A2 Milk, Curd, Ghee, Butter, Paneer, and Artisan Cheeses. The company sells its products in Bengaluru and most recently in Chennai with door delivery of Organic Milk Products to more than 25,000 customers.

About Nihal Sarin

Nihal Sarin (born 13 July 2004) is a 15-year-old Indian Chess Grandmaster with a FIDE rating of 2610, making him India No. 11 and World No. 206 at this tender age. He crossed the 2600 rating barrier at 14 years 10 months, becoming is the youngest Indian in history, and the third youngest player in history to do so. His blitz rating of 2685 already makes him World No. 49 and India No. 3 in the Blitz format. He won the 2019 Asian Continental Blitz Championships, becoming the youngest player in history to win the title. He is hailed by numerous chess pundits as one of the greatest natural talents in chess history. He was awarded the 2016 National Child Award For Exceptional Achievement by the then Indian President Pranab Mukherjee in a ceremony at the Rashtrapati Bhawan.


Editor-in-Chief emeritus of the ChessBase News page. Studied Philosophy and Linguistics at the University of Hamburg and Oxford, graduating with a thesis on speech act theory and moral language. He started a university career but switched to science journalism, producing documentaries for German TV. In 1986 he co-founded ChessBase.


Rules for reader comments


Not registered yet? Register