Fat Fritz – What on Earth is that?

by Albert Silver
8/13/2019 – It's a semi-secret development, an AlphaZero clone, engineered over the past nine months for ChessBase. Fat Fritz was tested by some of the best players in the world, who expressed unmitigated delight over the ideas and improvements it came up with. Now the program is publicly available on the ChessBase Engine Cloud. And it is running on awesome hardware. ALBERT SILVER explains.

ChessBase 15 - Upgrade ChessBase 15 - Upgrade

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. Start your personal success story with ChessBase and enjoy the game even more.

More...

Genesis

You probably know AlphaZero. It is a computer program that was developed by DeepMind in late 2017, an artificial intelligence research company founded by Demis Hassabis and subsequently purchased by Google. This AI program used awe inspiring hardware — 5000 custom tensor processing units (TPUs) to self-play tens of millions of games, and 64 second-generation TPUs to train the neural networks. It used no form of external chess knowledge but worked everything out from zero (hence the name). After four hours of work AlphaZero was playing better than the best brute force programs in the world. Here's a ChessBase description of the development: The future is here — AlphaZero learns chess.

Demis has been a friend of ChessBase co-founder Frederic Friedel since the mid 1990s, and his company was quite relaxed about the project and revealing the methods used to generate AlphaZero.

One of the project managers, Thore Gräpel [above in a blue shirt in the middle], research lead at Google DeepMind and Professor of Computer Science at the University College London, came to visit ChessBase in Hamburg, where he held a talk for half a dozen of our talented programmers. They went away inspired, determined to learn more about this kind of computer intelligence, and embark on AI projects themselves.

The Woosh

Back in late September 2018, I had been working on a program called “Deus X”, which was designed to explore and challenge the idea that learning from the best humans and engines, other than itself, would lead to a worse result. Using the open-source project Leela Chess as a foundation, which itself was an attempt to reproduce AlphaZero for the PC, I trained a neural network using millions of top human games from Mega Database and Correspondence Database, as well as engine vs engine games from the Playchess server and the CCRL ratings list. However, I did not reject self-play as a resource. I was only trying to prove that self-play was not the only valid source of content.

I wanted to round this off with self-play in its mature state, but lacked the resources to do so. I then spoke with my old friend Frederic Friedel, who expressed great interest in my ideas and agreed to help me. In early October 2018 he asked his son Tommy and nephew Noah to build him a really powerful computing machine. They bought the components, consisting of a 12-core processor and two state-of-the-art graphics cards that had just been released. These cards have thousands of graphic and tensor core processor units (GPUs and TPU), originally intended to power 3D video display in games as well as ray-tracing lighting effects. But it turns out that the processors are eminently suited for neural network development.

It is interesting to consider that if this machine had existed around the year 2000 it would have been the most powerful computer in the world!

After it was installed, Fred had a very powerful AI machine humming in his home office. Humming? Actually it is a fairly loud whirring sound of multiple fans, dissipating the heat from the 600 watts of energy the computer consumes. That heated the room during the cold Hamburg winter to a very comfortable 23°C, with central heating turned off. The constant sounds of the fans led it to be fondly referred to as “The Woosh”.

So what do we do with the super-machine? Fred offered me the chance to upload tools that were needed to build a neural network for chess, and once that was completed the machine went to work, playing an average 50,000 games per day against itself. I built myself a second, comparable machine which downloaded the games from the Woosh and learned from them – and from other games. Despite this sizable personal investment, and two strong computers, the project still had a thousand-fold hardware disadvantage when compared to AlphaZero, and at least hundred-fold compared to the community-driven effort Leela Chess. Nevertheless the hybrid mix soon came together in ways even our most optimistic thinking could not have predicted. Not only was it playing at superhuman levels of chess strength, but more importantly it did so in a completely new way, not with brute force tactics but with positional ideas that it has come up with, after studying millions of games and billions of positions.

Learning from human and computer games

To check the overall playing strength we entered an early version to compete in a well-respected computer tournament held in Leiden, Netherlands, and run by Jan Krabbenbos. It brought together not just the best PC programs around, but also unique efforts such as Jonny, a project that runs on a university server backed by 2400 CPU cores. As a matter of fact Jonny was using a hybrid neural network itself, as part of its makeup, and had beaten Leela in their individual game at the World Computer Championship. So it was not to be underestimated.

After four rounds, the leader was Jonny with a perfect score, followed by “Deus X” (the Fat Fritz prototype) with 3½ out of 4. We locked horns in round five, and after a tough battle, Deus X came out victorious with black, taking over the lead. After seven rounds, we had won the competition with 6 out of 7. That made it the first neural network to win a computer chess tournament. We showed the games to a couple of GM friends for their opinions, and one game stood out to them as utterly unique and beautiful in the attack the neural network built and conducted. See for yourself:

[Event "Leiden 2018"] [Site "?"] [Date "2018.12.01"] [Round "2"] [White "Rofchade"] [Black "DeusX Leiden"] [Result "0-1"] [ECO "C54"] [Annotator "Albert Silver"] [PlyCount "214"] [EventDate "2018.??.??"] {After a technical hiccup in which the main computer was unavailable in round one, this was the first game in which it would play at full strength. My book preparer was Erdogan Gunes.} 1. e4 {0} e5 {0.00/0 0} 2. Nf3 {11} Nc6 {0.00/0 0} 3. Bc4 {9} Bc5 {0.00/0 0} 4. c3 {9} Nf6 {0.00/0 0} 5. d3 {9} a5 {0.00/0 0} 6. O-O {8} d6 {0.00/0 0} 7. Bb3 {8} O-O {0.00/0 0} 8. Bg5 {114} h6 {0.00/0 0} 9. Bh4 {84} g5 {0.00/0 0} 10. Bg3 {149} g4 {0.00/0 0} 11. Nh4 {101} Nh5 {0.00/0 0} 12. Ng6 {522} Re8 {-0.18/18 76 This is the neural network's first move out of book.} 13. a4 {95 (Sh4)} Qg5 {-0.39/39 79} 14. Nh4 {310} Nxg3 {-0.29/29 1} 15. hxg3 {78} Ne7 {-0.29/29 0} 16. Na3 {68} Ng6 {-0.29/29 1} 17. Nxg6 {383 (d4)} Qxg6 {-0.64/64 77} 18. Nc2 {0} c6 {-0.62/62 1} 19. Qe2 {0 (Se3)} Kg7 {-1.16/ 116 98} 20. Qd2 {14 (Se3)} Ba7 {-0.94/94 103} 21. Ne3 {61} h5 {-0.86/86 1} 22. Rae1 {35} Be6 {-0.93/93 0} 23. Bxe6 {0} fxe6 {-0.99/99 0} 24. d4 {32 (f4) It is unclear what White thought of its position here, but Black was qutie happy and as of the next move will see a huge increase in its evaluation. [#]} Rh8 $3 {-1.42/142 72 This move brought no end of compliments by the grandmasters for both the depth of the concept and the elegance of the execution. And for the fact that Deus understood how good the move was. Black is now preparing the onslaught on White's king.} 25. dxe5 {77 (f4)} Raf8 $1 {-1.65/165 67 Black is not remotely concerned with material here, and again plays the most aggressive and active move.} 26. Qxd6 {47 [#] It is now down two pawns after Qxd6, but its evaluation just increases. This is symptomatic of its play, as it has learned to evaluate positions on what it feels offer the best scoring percentage, and not material counts.} Rf7 {-1.83/183 0} 27. c4 {22} h4 $1 { -2.78/22 65 So what does it think of its position exactly? -2.78} 28. c5 {13} Qh5 {-2.81/25 10} 29. Qxe6 {22} hxg3 {-2.86/30 0} 30. Qxg4+ {72} Qxg4 {-2.87/ 31 0} 31. Nxg4 {37} Bxc5 {-2.86/30 0} 32. Rc1 {134} ({After} 32. Ne3 {The attempt to cover the f2-pawn from the pin only serves to put a bigger target in the sights of the bishop. This doesn't worsen the situation for White (who is lost here), it just fails to improve it.} Kg6 33. e6 Rf4 34. e7 Re8 35. e5 Re4 36. Nd1 {Even after removing the knight from the deadly x-ray, the combination of bishop targeting and pinning f2 with the g3 pawn gives White no respite.} Rxa4 37. Re2 Rxe7 {and now, if} 38. Rfe1 {Black reminds White there is a little mating problem it needs to cover as well.} Rh4 $1 {with the idea of Reh7 and Rh1 mate!}) 32... Bd4 {-3.03/47 0} 33. Rc4 {32} c5 {-2.99/43 0} 34. Rc3 {39 White doesn't have many choices here, and a more passive attempt to just try to weather the storm is doomed to failure.} ({Ex:} 34. Rc2 Rf4 $1 { and the knight is just lost? Not because it has no squares to go to, but because if it moves, mate will follow on the h-file.} 35. Nf6 Rfh4 {and mate is unavoidable.}) 34... Bxc3 {-2.98/42 0} 35. bxc3 {106} gxf2+ {-2.94/38 0} 36. Nxf2 {11} Re8 {-2.93/37 0 Black simplifies and then converts with impeccable technique.} 37. Rb1 {65} Re6 {-2.91/35 1} 38. Rb5 {45 (Sg4)} b6 {-3.96/140 82} 39. Rb1 {63} Rf8 {-3.83/127 0} 40. Rb2 {39 (Tb3)} Kg8 {-3.69/113 95} 41. Rb1 { 39} Rg6 {-3.68/112 8} 42. Rb2 {11 (Sh3)} Re8 {-4.99/243 82} 43. Kh2 {41 (g4)} Rxe5 {-5.53/41 53} 44. g4 {10} Rf6 {-5.60/48 10} 45. Kg3 {15} c4 {-5.70/58 1} 46. Rb1 {37} Kf8 {-5.77/65 29} 47. Rd1 {41} Ke7 {-5.95/83 46} 48. Rd4 {32 (Td2) } Rc5 {-6.78/166 57} 49. e5 {38 (Td2)} Rxe5 {-9.00/132 73} 50. Ne4 {11} Rf1 { -9.13/145 13} 51. Nd6 {28} Re6 {-8.94/126 0} 52. Nf5+ {28} Kf6 {-8.80/112 0} 53. Rd8 {38} Ke5 {-8.82/114 0} 54. Nd4 {54 (Td4)} Ref6 {-10.37/13 71} 55. Re8+ {17} Kd6 {-10.07/239 28} 56. Nf5+ {92} Kd7 {-8.79/111 1} 57. Re4 {31 (Te7+)} Rc1 {-10.88/64 29} 58. Rxc4 {27} Rc6 {-10.44/20 2} 59. Rd4+ {29} Ke6 {-10.11/ 243 138} 60. Kf4 {27} R1xc3 {-9.85/217 1} 61. Re4+ {33} Kf7 {-9.91/223 41} 62. g5 {59} Rc1 {-10.75/51 0} 63. Kg4 {28} Rf1 {-10.90/66 1} 64. Rd4 {27} Ke6 { -13.52/72 40} 65. Ng3 {62} Rf2 {-13.15/35 1} 66. Rd3 {10 (Te4+)} Rc4+ {-18.12/ 20 67} 67. Kh5 {8} Rh2+ {-16.50/114 18} 68. Kg6 {10} Rxa4 {-16.05/69 9} 69. Rd1 {52 (Sf5)} Rg2 {-22.02/154 134} 70. Re1+ {11} Kd7 {-19.76/184 90} 71. Ne4 {30} Rc4 {-21.54/106 128} 72. Kf5 {10} a4 {-22.95/247 107} 73. g6 {10} a3 {-22.51/ 203 91} 74. Ra1 {30} Rxg6 {-29.16/100 81} 75. Kxg6 {103 (Txa3)} Rxe4 {-128.00/ 0 3} 76. Rxa3 {36} b5 {-279.95/91 0} 77. Kf5 {44} Rh4 {-279.96/92 0} 78. Ra6 { 20} b4 {-279.92/88 0} 79. Ke5 {14} Kc7 {-279.93/89 0} 80. Kd5 {62} b3 {-279.92/ 88 0} 81. Ra3 {46} Rb4 {-279.98/94 0} 82. Kc5 {9} b2 {-279.98/94 0} 83. Kxb4 { 13} b1=Q+ {-279.78/74 0} 84. Rb3 {21} Qe4+ {-279.79/75 0} 85. Kc3 {9} Kd6 { -279.80/76 0} 86. Kd2 {31} Kd5 {-279.81/77 0} 87. Re3 {20} Qf5 {-279.83/79 0} 88. Ke2 {21} Kd4 {-279.84/80 0} 89. Rg3 {8} Qe4+ {-279.85/81 0} 90. Kf2 {17} Qf4+ {-279.86/82 0} 91. Kg2 {8} Ke4 {-279.87/83 0} 92. Rh3 {11} Qg4+ {-279.90/ 86 0} 93. Kh2 {13} Kf4 {-279.93/89 0} 94. Rh6 {11} Ke3 {-279.94/90 0} 95. Rb6 { 9} Qg7 {-279.95/91 0} 96. Rb4 {11} Kf3 {-279.96/92 0} 97. Rb3+ {9} Kf2 { -279.97/93 0} 98. Kh3 {8} Qh6+ {-279.98/94 0} 99. Kg4 {10} Qe6+ {-279.99/95 0} 100. Kf4 {7} Qxb3 {-279.93/89 0} 101. Kf5 {11} Qe3 {-279.94/90 0} 102. Kf6 {11} Kf3 {-279.95/91 0} 103. Kf7 {8} Kf4 {-279.96/92 0} 104. Kf6 {7} Qe8 {-279.97/ 93 0} 105. Kg7 {6} Kf5 {-279.98/94 0} 106. Kh7 {6} Kf6 {-279.99/95 0} 107. Kh6 {8} Qg6# {-279.99/95 0} 0-1

Very encouraging! But that was not the main objective of the development — to make a program to compete against other chess engines. What we were aiming for was a new analytical tool that chess players could use to reevaluate their openings, and their general strategic ideas in the game. So it became important to have top GMs test the program, tell us what they thought, and whether the program was useful. That was more important than the score against Komodo or Stockfish.

So we gave access to half a dozen really deep players on the Engine Cloud. Some of them hovering around 2800 on the Elo scale, some were weaker tournament players. We wanted to find out whether the latter found the program helpful and useful to their preparation. Their reactions were, without exception, extremely positive — often exuberant. They all wanted permanent access to the program. We will be telling you about their reactions and giving you examples of their analysis in the following weeks. It is very interesting to see how just an hour or two with the program can change the evaluation of openings and positions of 2700+ players analysing with it.

Vishy

A number of top players will probably not allow us to reveal what they had found — the positions and moves they shared with us. "I cannot let you use them in your articles," one of them typically wrote. "They haven't yet been used in practice, and for obvious reasons I want to reserve them for future games, to score points." A number of them did not even want us to tell people that they had been working with the machine.

Five-time World Champion Vishy Anand was an exception, and gratefully so. In a friendly chat, after he had finished a tough game against a top player, the topic of neural networks came up. We gave him private access to a Fat Fritz prototype, running on the same machine now being offered on the ChessBase Engine Cloud. His reaction was unequivocally enthusiastic:

After two days preparing tournaments and games with it (Fat Fritz), I have come to the conclusion I will probably need to revisit my entire repertoire with it. Being able to work with such a powerful setup, thanks to the cloud, from my laptop from anywhere around the world, has been pure joy. It is very interesting to work with Fat Fritz.

Fat Fritz in the Engine Cloud

For a few days now we have silently launched the neural network program Anand is talking about on the ChessBase Engine Cloud. We decided to call it Fat Fritz, a tongue-in-cheek name that is easy to remember and talk about. Here's a summary description of the engine:

  • Fat Fritz is an advanced AlphaZero-type artificial intelligence chess engine that has been trained on billions of positions from human and computer-vs-computer games, self-play games, and endgame tablebases.

  • Hardware: Fat Fritz runs in the ChessBase Engine Cloud on computers equipped with two of the very latest top-of-the-line graphics cards (GPUs), making it hundreds of times faster than neural network engines running on CPUs.

The program is available for you to try out in the Engine Cloud. Some of you will be familiar with this service, which allows you to use a chess engine on a very powerful remote computer from within the ChessBase program, as if it were running on your local hardware. You will be able to try out Fat Fritz immediately — it costs around 30 Ducats per hour, which translates to €3.60/hour.

For those of you who have never used a cloud engine, or who are not completely familiar with how this works, we will be publishing a full tutorial in the next days. It is useful to learn, and very easy — all top players work on the Engine Cloud constantly.

Things to remember
  • The original AlphaZero was rooted in the concept of pure reinforcement learning, hence the ‘zero’, meaning it learned everything purely from scratch. A wonderful and fascinating premise, but there is no reason to believe this cannot be enriched with all the material developed until now.

  • Our AI project, Fat Fritz, not only uses exclusively created reinforcement learning self-play games, but it has also learned from millions of master level games from Mega Database, millions of top engine games, and even from endgame tablebases. As a result this has given it a more comprehensive and balanced look than a purely ‘zero’ approach would have allowed.

  • With millions of self-play games alone, Fat Fritz would only be truly awesome and innovative in its favorite openings, those that were played and trained. Many openings that the engine might not like would only receive cursory attention. By adding the master level games the program gains insights into openings it did not originally like, and reevaluates them. It might still hate some, but at least it will be based on extensive study and exposure to them.

  • The hardware used by Fat Fritz on the Cloud is critical. Running a neural network engine on a CPU gives you only a tiny fraction of its true power. When you have two of the fastest GPUs available today, overclocked, you get an unparalleled experience that would cost thousands of Euros to put together.

  • Incidentally, when using Fat Fritz you should not be nervous about the small number of nodes per second the program displays. Remember that the evaluation function of a typical brute force program is processing a few thousand positional algorithms, while Fat Fritz is working with 27.9 million values to evaluate each position — a vastly greater knowledge base.

  • Finally it is worth mentioning that Fat Fritz is still very much a work in progress, meaning that it will be updated with newer and stronger versions on the ChessBase Engine Cloud.

Fat Fritz

Use it now with ChessBase 15 and Ducats!

Links




Born in the US, he grew up in Paris, France, where he completed his Baccalaureat, and after college moved to Rio de Janeiro, Brazil. He had a peak rating of 2240 FIDE, and was a key designer of Chess Assistant 6. In 2010 he joined the ChessBase family as an editor and writer at ChessBase News. He is also a passionate photographer with work appearing in numerous publications.
Discussion and Feedback Join the public discussion or submit your feedback to the editors


Discuss

Rules for reader comments

 
 

Not registered yet? Register

rav555 rav555 10/15/2019 12:23
When will Fat Fritz be available to the consumer? I suspect that a 32 core 7nm AMD EPYC CPU with a 7nm+ Radeon M150 Accelerator would be a very good workstation for Fat Fritz.

The computer that you built would not have been the "in 2000... the most powerful computer in the world". Perhaps the most powerful desktop but not computer.
Grimmell73 Grimmell73 8/16/2019 02:55
Was Fat Fritz using tablebases at the end of the game, or playing entirely from AI?
Jarman Jarman 8/15/2019 05:00
I would be most interested if it was possible to access the engine from a mobile platform (iOS/Android). Nice start, btw.
acepoint acepoint 8/15/2019 03:38
Somehow I feel reminded on Rybka/Fritz15. Maybe this is even worse if I consider how lc0/DeusX was handled by Albert Silver.
celeje celeje 8/15/2019 01:08
The main Lc0 author said this:

""DeusX" was Lc0 binary, Lc0 neural network architecture and implementation, trained with Lc0 scripts (so the only thing different is NN weights, it's not even "different neural net" as some people say).
I don't know what "Fat Fritz" is, but I expect it's the same."
combinatorialist combinatorialist 8/14/2019 10:11
Which "well-respected computer tournament held in Leiden" did Deus X win?

In contrast, Lc0 (using self-play only) won the TCEC superfinal this year and has won the TCEC cup title twice.

Given the careful statement made in www.chessdom.com/statements-by-deus-x-and-leela-chess-zero-authors/ about Deus X's use of Lc0, the lack of a similar clear statement here concerning Fat Fritz is both surprising and disappointing.
PeterP PeterP 8/14/2019 03:34
Albert Silver entered into the TCEC championships a tweeked version of Lc0 (Leela) software with a neural network he had trained using Lc0 and other Leela tools. He was far from upfront that it was not his own work and was widely regarded as lacking integrity for doing so.

Now we have FatFritz and there must be a strong suspicion that this too is based upon Lc0 software. I would regard both Chessbase and Albert Silver as showing bad character if it were and they did not immediately declare it so. Whether or notr it would violate the Lc0 GPL is a whole other matter.
celeje celeje 8/14/2019 01:57
@ Mr TambourineMan:

Yes, but it looks like that wording ("Using ... Leela ... as a foundation") is meant to suggest that was only DeusX and not "Fat Fritz". If it's not in "Fat Fritz", someone would have had to have coded up their own version from scratch. Most people now think that's very unlikely. Either way, ChessBase should state what the truth is.

On clones: you and I can train our own networks if we have the computing power and use the Leela code. We'd need to agree on what a "clone" is before we could agree whether the result is a clone.

Again, if they did not use Leela code, someone would have had to do a lot of coding from scratch before they could even do the training.

If you have a lot of computer power and the Leela code, just training a different network is not an independent creative effort.
Mr TambourineMan Mr TambourineMan 8/14/2019 12:25
Thanks Urhen1978 and Celeje for the information. If so it reminds me of the Rybka scenario.

You ask Chessbase to come out and very clearly state whether this program is using Leela/Lc0 programming code or not? Albert Silver is clear about it here in the article "Using the open-source project Leela Chess as a foundation". But you want ChessBase to declare exactly what of the Leela code it has used. I hope theyll do. Also I hope theyll give applause for Leela project in up coming articles. But I feel you are saying that Fat Fritz / Deus X are only a clone. Albert/CB are clearly developing something as they train the neural network with games from Mega, Corr, Engin vs Engin database and also even "learns" from Endgame Tablebase. [what is there to "learn"; just input Lomanosov Tablebase 7 pieces and of we go to find ways to reach them end positions]

I was reading Urhens1978 link (thank you) and it said "Leela is still having many blind spots resulting in blunders from time to time mostly in endgame" Maybe its because of lack of AlphaBeta Search or just maybe I am right in my criticism that not using proven knowledge is a fundamental lack in the learning process. And that is as I see it also what Albert says in this article "I was only trying to prove that self-play was not the only valid source of content".

Sure its cool that we have find a way to self learning machines but maybe it could be even cooler if we could also use proven knowledge to get even better machines, and if so only using already proven knowledge means that we get them machines faster!
celeje celeje 8/14/2019 10:51
@urhen1978:

I think that would be only a start. ChessBase needs to declare exactly what of the Leela code it has used at any time in this "Fat Fritz" project.
urhen1978 urhen1978 8/14/2019 10:33
It would have been fair to do two things:

- Share a link of the lc0 project to for your readers;
- remind your readers that your initial effort on DeusEx was not considered fair by the Leela chess community https://blog.lczero.org/2018/08/update-from-lczero-core-devs-team-on.html;

This is not how you make good use of a community-driven project
celeje celeje 8/14/2019 09:44
It is ChessBase's ethical and legal responsibility to come out and very clearly state whether this program is using Leela/Lc0 programming code or not.

I would not like to continue supporting ChessBase financially if they are stealing open-source work of others without even saying so.
Magic_Knight Magic_Knight 8/14/2019 06:29
They should have called it Phat Fritz.
tsedawa tsedawa 8/14/2019 03:44
leela is the current world computer chess champion . she has beaten the latest stockfish 9 easily. this article is misleading people for profit.
Lachesis Lachesis 8/13/2019 11:29
Outstanding News!
blueforest blueforest 8/13/2019 10:32
Will fritz 17 have fat fritz as its engine?
Mr TambourineMan Mr TambourineMan 8/13/2019 07:09
Now we only got to see one game but of that and the explanatory text I think Fat Fritz places more emphasis on the endgame in strategy and tactics and it shows a fundamental lack of AlphaZero's learning process (correct me if I'm wrong) that it did not get to use or learn from Endgame Databases something Fat Fritz thus has and thus in a way can learn chess more effectively. Just imagine if you could get the same access to computing power with Fatty, where would that path lead?
Serse Serse 8/13/2019 05:31
Fascinating. I can't wait to give it a trial. Thanks for the initiative.
1