Fat Fritz analysis: A matter of technique

by Albert Silver
9/24/2019 – Although it is not hard to pick out moves that Fat Fritz finds more easily than Stockfish, or vice versa, sometimes a move stands out that just begs to be shared. This happened to Fat Fritz creator, Albert Silver, while he was casually following the recent Sinquefield Cup won by Ding Liren. Enjoy this example of positional judgement and technique. | Photo: Lennart Ootes / Grand Chess Tour

Komodo Chess 14 Komodo Chess 14

Last year Komodo won the world championship title on two occasions and can call itself "2019 World Computer Chess Champion" and "2019 World Chess Software Champion". And the current Komodo 14 has been clearly improved over its predecessor!


Insights into Anand's game

Like many readers here, I am foremost a chess player and chess fan, and I was enjoying the battles from the Sinquefield Cup with great relish, in spite of the disappointing number of games that failed to produce the deserved decisive result. One of the most thrilling games was just such a case: a near win by Vishy Anand over Ding Liren that ended in a draw.

Most players and pundits will remember the game mostly for the missed 26…e4!! which would have won the game on the spot. 


To be fair, while Vishy himself was undoubtedly gnashing his knuckles over this later that night, Ding Liren readily admitted in the post-game interview that he himself had not even suspected the move during the game.

When the gods disagree

However, there was a crushing move missed two moves earlier that was not mentioned by the commentators because they were fixated on the opinion of Stockfish. Here is the position two moves earlier after 24.f3


In this position, there wasn’t a lot of comment about Black’s choice 24…h5 as this was also Stockfish's choice, so feedback was along the lines of “he found the best move”. I sat up at this point because on my laptop, with a fairly middling GPU, I had been following the game with a beta of Fat Fritz, and it was practically chanting victory in the case of 24…c4!! 

Here is what I was seeing

In fact, Fat Fritz was to point out that ...c4 was a thematic play on more than one occasion after this, including when Anand played it during the game, though by then it needed very precise play to guarantee the win. After the game, I asked Anand about this, and he said that ...♛c4 was 'obvious technique', and as such was a move he analysed numerous times. 

The curious thing was that when I consulted Stockfish on my i7 processor, even after several minutes of analysis it ranked 24…c4 a distant third, just better for Black, but hardly crushing.

So now the question remained: who was right? Fat Fritz with its evaluation that ♛c4 is crushing, or Stockfish that thinks it is not bad, but not especially great either. I fed Stockfish the moves chosen by Fat Fritz to see if it could refute them, especially as the position is still very complex. 

The spectacular key move appears just three moves later after 24.f3 c4 25.xc4 dxc4 26.g5 d5 27.a5:


Don't forget you can move the pieces above!

Black's killer blow here is 27…a7!! With the idea of 28…a8. Taking the bishop leads to a quick demise after 28.xa7? a8! 29.b5 xa5 recovering the piece with a pawn to boot. 

Following the analysis by both engines led to many fascinating lines, but all were clearly going to win for Black. What was striking was that Stockfish was extremely slow to credit black with even a -2 advantage even 10 moves later, in spite of the extra exchange.

For example, here is a position after one line of analysis over ten moves later:


By now Fat Fritz is in the clear stratosphere judging itself at +13 or so, which just means it thinks this is a mere matter of technique. Technique it is confident it has needless to say. Here Stockfish says it plans 36.g4, expecting 36...f4 in reply, completely locking down the position, and only after two billion nodes does its evaluation begin to climb past -2 and -3 for Black. As a side note, trying to sit on the position with a move like 36.b3 fails as Black bulldozes through with 36...b5!

Clearly, this was not just a case of a missed tactic, since even after the tricky moves were past, it still struggled to perceive this as won.

Declining the exchange

Assuming there is no argument that 25.xc4 will lose for White, what if he declines the queen exchange with 25.d1!? 


Black's obvious continuation is 25...b4 with the nasty dual threat of ...xb2 and ...d2. What can white do? Vishy said that he had seen this line as well, but ran into a powerful resource he failed to see a way to overcome. Here he analysed — in complete agreement with Fat Fritz it should be noted — 25...b4 26.b3 d2... 


The answer is 27.f2!! a superb shot that protects the queen and threatens 28.xd2 where the pawn will be a target. Unable to see how to continue here, this ended up being the line that prevented the former World Champion from playing it. Although there is a winning continuation here, it is hardly obvious. The continuation goes 27...d7 28.c5 xc5 29.dxc5 d4!


And now the attempt to confuse the issue with 30.xd2 exd2 31.xd4 d8 32.e3 fails to the elegant 32...e5 33.f1 xd4 34.exd4:


And here Black wins with the simple 34...e8! 

Aside from the impressive evaluation by the neural network, the attractive lines punctuated by a few beautiful moves really did not deserve to be lost into oblivion — hence this article. As to Anand missing this possibility, the complexity is such that even a great player would have no reason to moan about missing it. Since he saw the power of the idea, his judgement was clearly not inferior, the problem was making the lines work.

Replay the full analysis

[Event "7th Sinquefield Cup GCT 2019"] [Site "Saint Louis"] [Date "2019.08.24"] [Round "7.1"] [White "Ding, Liren"] [Black "Anand, Viswanathan"] [Result "1/2-1/2"] [ECO "A40"] [WhiteElo "2805"] [BlackElo "2756"] [Annotator "Albert"] [PlyCount "119"] [EventDate "2019.08.17"] [EventType "tourn"] [EventCountry "USA"] [SourceTitle "playchess.com"] [Source "ChessBase"] [SourceQuality "1"] 1. d4 Nf6 2. c4 e6 3. g3 Bb4+ 4. Bd2 Be7 5. Bg2 d5 6. Nf3 O-O 7. O-O Nbd7 8. Qc2 c6 9. a4 a5 10. Na3 Bd6 11. Ne1 Re8 12. Nd3 e5 13. cxd5 e4 14. Nf4 cxd5 15. Nb5 Bb8 16. Rac1 Ra6 17. Bh3 Rc6 18. Qb3 Nb6 19. Bxa5 Bxh3 20. Nxh3 Qc8 21. Kg2 Rxc1 22. Rxc1 Qxc1 23. Bxb6 e3 24. f3 h5 $2 {Although this does not compromise Black's chances in any obvious way, a stronger move was missed.} (24... Qc4 $3 {This is a hammer blow, and two lines need to be examined: declining to exchange queens, or accepting.} 25. Qxc4 {The first and foremost question is why is exchanging queens so bad. Black may be better, sure, but outright losing?} (25. Qd1 {is the next big question: can Black win if White refuses to play along?} Qb4 26. b3 Qd2 27. Nf2 $1 {A fantastic resource that seems to hold the position. Or does it?} Nd7 28. Bc5 Nxc5 29. dxc5 d4 30. b4 (30. Qxd2 exd2 31. Nxd4 Rd8 32. e3 Be5 33. Kf1 Bxd4 34. exd4 Re8 $1)) 25... dxc4 26. Ng5 Nd5 27. Ba5 {[#] White is trying to organize its position into a semblance of a defense, but Fat Fritz finds a lovely tactic here:} Ba7 $3 {Really deserving two exclamation points.} 28. Bc3 (28. Nxa7 {The obvious question is what happens if White takes the bishop. It is not the mainline as White's position falls apart very quickly.} Ra8 29. Nb5 Rxa5 {recouping the piece and a pawn to boot.}) 28... Ra8 $1 {This is the entire point of Ba7 and the winning evaluation it gave at the start. It brings the took into play where it will be able to wreak havoc. So long as White is able to lock the position up and prevent the rook from penetrating, the chances to hold are very good, but once it is able to leverage its long range ability, White has little hope.} 29. Ne4 Bb8 30. a5 {This is Fat Fritz's suggested move.} (30. Nc5 {Is Stockfish's preferred continuation, though neither save the position.} b6 31. Nd7 f5 32. Be1 Kf7 33. Nxb8 Rxb8 34. Nd6+ Ke6 35. Nxc4 g5 {Here, after over two billion nodes, Stockfish suggests it will play} 36. g4 {and after} ({Trying to just sit and wait with a move such as} 36. b3 {won't work as Black just bulldozes his way in with} b5 $1) 36... f4 {, and it is also the first situation where its eval finally breaks -2 and even -3 in Black's favor.}) 30... f5 31. Nc5 Nxc3 32. Nxc3 Rxa5 33. Nxb7 Ra6 34. f4 Rb6 35. Na5 Rxb2 {and now it is clear White's position is collapsing.} {The c4 pawn is untouchable since} 36. Nxc4 { loses a piece to} Rc2) 25. Ba5 h4 26. gxh4 Qc6 27. Be1 Nh5 28. Bg3 Qc4 29. Qxc4 dxc4 30. Ng5 Nf6 31. Na3 Bxg3 32. hxg3 Ra8 33. Nxc4 Rxa4 34. Nxe3 Rxd4 35. Kf2 Nd5 36. Nf5 Rb4 37. h5 Rxb2 38. h6 f6 39. Ne4 gxh6 40. Nxh6+ Kf8 41. Nf5 b5 42. Ne3 Nb6 43. Ke1 b4 44. Kd1 Na4 45. Nd5 f5 46. Nd2 b3 47. e4 fxe4 48. fxe4 Kf7 49. g4 Nc5 50. Nb4 Ke6 51. Kc1 Na4 52. g5 Ke7 53. Nc6+ Kf7 54. Nd4 Kg6 55. N4xb3 Ra2 56. Nf3 Nc3 57. Nbd2 Rxd2 58. Kxd2 Nxe4+ 59. Ke3 Nxg5 60. Nxg5 1/2-1/2


Did Fat Fritz see this all to a concrete end? Of course not. It certainly saw the mainline and the key shot 27...a7!! almost immediately, but further down the road it is really all about positional judgement. I think this is the sort of position that really shows just how different analysis with Fat Fritz can be, and some of the things you can expect to encounter and enjoy.


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.


Rules for reader comments


Not registered yet? Register

Dendogodned Dendogodned 11/12/2019 08:38
Please tell me it can use more then 2 gpus
Kurt Utzinger Kurt Utzinger 9/25/2019 02:58
Insights into Anand's game: There is no Fat Fritz needed to see all these variations. Even Lc0 v0.22.0 id 42700 on a slow GeForce 840M finds all the best moves immediately.
chessstudent007 chessstudent007 9/25/2019 04:47
Many thanks. Your answer to my comment is exactly what I was looking for. Very nicely summarized. Thank you !
Bertman Bertman 9/25/2019 04:43
@chessstudent I apologize and you are completely correct. I promise to be more attentive to this in the future, and thank you for your support. Since it is a bit late to really throw this in there, I will summarize two main points here in the feedback:

- a key point of the queen exchange on c4 is that it prevents White from easily protecting the a4-pawn with b2-b3, making it ripe to become a target for the rook, which it does with that fancy Ba7-Ra8 maneuver.

- Winning a position like with an exchange up can sometimes be really problematic if the minor pieces can lock everything down and protect them, or fail to give the rook good options to wield its superior power. Qc4 Qxc4 dxc4 helps resolve this. White tries very hard to prevent the rook from entering, but enter it does, making the win that much easier to convert.
chessstudent007 chessstudent007 9/25/2019 04:33
Albert I really enjoy all of your writing however just a question or perhaps a suggestion. I am someone who enjoys chess for the beauty of it. I am not a great player myself but I do enjoy the game. For someone like me, it makes more sense to read broad theme that leads to victory. For instance , [not related to this game !] engine A's path to victory is to to get a passed pawn. Or Engine B's plan leads to trapped piece and ultimately winning it. If you can possibly add some comments to these effects, it will make the article more enjoyable. Even commentators sometime say "Oh this is clearly winning" ... sure for a club player its perhaps clear but for average person who is average, such as myself, I do not always get to see the "obvious" that they talk about ! Hope I am able to express myself. Not trying to criticize .... I do read every article you write.
Bertman Bertman 9/25/2019 01:25
@MrT Fat Fritz is extremely versatile, and the example chosen was specifically on a positional question. As to tablebases, my laptop does not have them, so this was pure internal knowledge and judgement. Remember also that the many of the games you see of AlphaZero were the most attractive ones and handpicked, which is quite normal. I have been seeking to highlight differences where it stands out compared to the conventional engines. Sometime not too far from now line I will run some longer matches to give such attacks, but my computer time is completely tied up in development for the moment. Two or three days running a lengthy match means two or three days where its progress is frozen.
caliche2016 caliche2016 9/25/2019 12:00
In my case, all 3 engines Stockfish 10, Komodo 13 and Houdini 6 find 24...Qc4! very quickly, however, all of them struggle quite a bit deciding which move is the strongest 24...Qc4 or 24...Qc6. On the other hand, Leela Chess Zero finds 24...Qc4! in a split second and keeps it as the strongest option. This and other tests with examples provided by Mr. Silver, make me believe the free engine Leela is as strong as Fat Fritz when it comes to chess analysis.
Mr TambourineMan Mr TambourineMan 9/24/2019 10:32
I looks as Fat Fritz plays as a defensive player in table tennis or the sort of clay tennis that was played in the 80's. No serve and volley - only basline returning of the ball until opponents get tired. Or in tabletennis eternal defensive blocks with backspin push. Okey nice to see the opposite to AlphaZero. But I do think it has been so as AlphaZero does win before endgames and also that Fatty uses Endgame Tablebase! Would be nice to have Albert comment on my shall we say observation?
SunriseK SunriseK 9/24/2019 10:01
I've always thought that Stockfish is not that good as it could apparently seem; for example my old Komodo or my old Houdini find 24..., Qc4!! already at depth 15 plies. Many times, watching superGM games and seeing how it evaluate the moves, I found its assessments at least "bizarre", and I'm sure also the developers know this well, as they must have "retuned" the engine many times.
chinguaro chinguaro 9/24/2019 10:00
If Fat Fritz is so good, the best thing to do is to make a tournament in equal conditions with Stockfish, Leela Chess Zero, Houdini, etc., to see who is the best and not to be launching advertising with certain positions.
Halflash Halflash 9/24/2019 09:41
I can't wait wondering about the next article : How Fat Fritz unbusted the King's Gambit.
Fulcrum2000 Fulcrum2000 9/24/2019 09:03
Why are you using a 1,5 year old Stockfish version for reference?. Please use a recent version.