This article supplements a series of videos about Komodo Dragon’s opening repertoire which appeared on the “Silicon Road to Chess Improvement” YouTube site (https://www.youtube.com/c/SiliconRoadChess) on 10th July 2022. This article provides an overview of Komodo Dragon’s main openings; for more detail take a look at the video series!
Introduction to the opening play of engines
Until late 2017, strong human players could still claim ascendancy over the best engines in a few areas of the game. This was especially true of opening play. Engines like Stockfish Classical and Komodo Classical were of disappointing value in openings where tactics were not the primary feature. In essence they lacked the URGENCY that strong human players have in the opening phase: the sense that development should be completed as efficiently as possible, the sense of opportunities to complicate that arise while both sides are still mobilising. Setting engines of that period on an opening position for 6 hours was to produce a mass of difficult-to-understand variations, many of which contained strange and unexplainable tempo losses. Of course some gems of insight were buried within these lines but it was an intensive and sometimes distressing experience to extract them! This all changed for me in early 2018 when I got to witness AlphaZero’s games against Stockfish 8. There were many impressive facets to AlphaZero’s play, but especially its opening play was on a different level to anything else in the engine world. Just like human players, it was looking to MAKE THINGS HAPPEN from the earliest moment which was reflected in its discovery and espousal of some super-sharp lines against Stockfish 8’s favourite Queen’s Indian. For example:
1.d4 Nf6 2.c4 e6 3.Nf3 b6 4.g3 Bb7 5.Bg2 Be7 6.d5 exd5 7.Nd4
1.d4 Nf6 2.c4 e6 3.Nf3 b6 4.g3 Ba6 5.Qc2 c5 6.d5 exd5 7.cxd5 Bb7 8.Bg2 Nxd5 9.0–0
Leela – AlphaZero’s “spiritual successor” – displayed the same aptitude for the opening phase and with the introduction of NNUE technology (neural net technology though different to AlphaZero and Leela Zero) first into Stockfish and then into a plethora of other engines – such as Komodo Dragon which is the focal point of this first series – we saw the same quantum leap in strength in all the other engines. Nowadays, strong players look to their engines for ideas in every type of opening!
Why and How
It struck me that it would be intriguing to look at the opening choices of the latest engines and to compare them to conventional wisdom about the opening. When I talk about “conventional wisdom” I’m not talking about the current thinking of elite players about the opening but more about the orthodoxy which was prevalent until about 10 or 15 years ago and which I think is still how the average non-professional chess player thinks about openings.
In a series of videos on the “Silicon Road to Chess Improvement” YouTube site (https://www.youtube.com/c/SiliconRoadChess) we examine the opening choices of Komodo Dragon. Komodo Dragon is one of the top 3 engines, most probably still 3rd behind Leela Zero. The first video is an overview of Komodo Dragon’s choices against the main opening moves. In subsequent videos, I zoom into Komodo Dragon’s choices in a little more detail.
I used Komodo Dragon v3 – the latest version commercially available. To reach my conclusions, I let Komodo Dragon run for vast amounts of time on my most powerful hardware, typically to a depth of 54, searching a couple of hundred thousand million ply (nodes). I have noted these values next to most variations using d= to signify depth and Mn to specify millions of nodes.
Just a few words of caution / context when looking at these conclusions.
1. We are often talking about differences of 0.01-0.02 between variations over billions of analysed moves so we are often not talking about massive differences between primary and secondary lines. You might see some differences if you ran these tests multiple times.
2. Note also that the last few moves of a line of analysis given by an engine are often not optimal. In a long variation where tactics are not at a premium, I normally cut off the analysis a couple of moves early.
3. Komodo Dragon’s evaluations are on a similar scale to Stockfish’s evaluations: if you are used to those, then Komodo Dragon’s evaluations mean pretty much the same.
4. Bear in mind, that the opening position is just a draw between engines of the level of Komodo Dragon. So don’t be surprised if none of the lines given here lead to a decisive advantage!
The pgn files which I used to produce the videos are shared here: http://cloudserver.chessbase.com/MTIyMTYx/replay.html
Let’s start off with Komodo Dragon’s view of best play for both sides from the starting position. I have to warn you… it isn’t pretty!
If you let Komodo Dragon analyse from the starting position, then I think you can guess (unfortunately) which line we get… the Berlin! [d=58 732513Mn]
2.Nf3 Nc6 3.Bb5 Nf6 4.0–0 Nxe4
Why is the Berlin so popular with neural net engines? I think it’s a product of this URGENCY I mentioned before: the neural nets are striving to redress the balance as quickly as possible, if necessary with direct action and the Berlin is the quickest opportunity for this, capturing one of White’s centre pawns and flattening out the position by forcing exchanges and – in many lines – a symmetrical pawn structure.
5.Re1 Nd6 6.Nxe5 Nxe5 7.Rxe5+ Be7 8.Bf1 0–0 9.d4 Bf6 10.Re1 Re8 11.c3 Rxe1 12.Qxe1 Ne8 13.Bf4 d5
Well… you get the idea! Komodo Dragon ends its line at move 29 with a 0.29 evaluation but in principle this is completely drawn.
14.Nd2 Bf5 15.Qe2 c6 16.Re1 Nd6 17.Qf3 Ne4 18.Nxe4 Bxe4 19.Qg3 Bg6 20.Bc7 Qd7 21.Qd6 Qxd6 22.Bxd6 Bf5 23.f3 h5 24.Kf2 g6 25.g3 a5 26.Bf4 g5 27.Bd6 Kg7 28.h3 h4 29.g4 Be6
Things get more interesting when you look at Komodo Dragon vs 1.d4. Actually, I nearly fell off my chair when I saw them 😊
1.d4 Nf6 2.Nf3 d5
2…e6 3.Bf4 Here once again, Komodo Dragon prefers the London System to main line d4–c4 systems! This conclusion is also reached after a great deal of analysis [d=56, 233789Mn] An amazing conclusion! There’s more to come!
The London System! [d=56, 178373Mn] Indeed, Komodo Dragon considers it more promising than Main Line d4–c4 systems!
3…c5 4.e3 e6 5.Nbd2 Qb6 6.Rb1 Bd6 7.dxc5 Qxc5 8.Bd3
8.Bxd6 Komodo’s choice with slightly less depth [d=55, 85262Mn] 8…Qxd6 9.c4 dxc4N (Predecessor: 9…0–0 10.b4 b6 11.Be2 Bb7 12.0–0 Qe7 13.Qb3 Nbd7 14.Rfc1 Rfc8 15.h3 h6 16.a3 a5 17.cxd5 Bxd5 18.Rxc8+ Rxc8 19.Qb2 axb4 20.axb4 Ne4 21.Nxe4 Bxe4 22.Rd1 Nf6 23.Qd4 Qc7 24.Ne5 Bd5 25.Ba6 Ra8 26.Bd3 Bb3 27.Rb1 Rd8 28.Qb2 Bd5 29.Rc1 Qd6 30.Bf1 Bb7 31.b5 Ne4 32.Nc6 Ra8 33.Qc2 Nc5 34.Rd1 Qc7 35.Be2 Bxc6 36.bxc6 Qxc6 37.Bf3 Qa4 38.Qd2 Rc8 39.Qd6 Sunilduth Lyna,N (2658)-Indjic,A (2605) Chess.com INT 2022 0–1 (102)) 10.Bxc4 0–0 11.0–0 b6 12.Rc1 Bb7 13.Be2 Nc6 14.Nc4 Qe7 15.Qd6 Rfd8 16.Qxe7 Nxe7 17.Nce5 Ne8 18.Ng5 f6 19.Nxe6 fxe5 20.Nxd8 Rxd8 21.Bc4+ Kf8 22.Rfd1 Nc6 23.Rxd8 Nxd8 24.f4 exf4 25.exf4 Bc6 26.g3 Nf6 27.Be2 Ke7 28.b4 Kd6 29.Rd1+ Ke7 is assessed as 0.29 by Komodo Dragon.
8…b6 9.0–0 Bb7 10.b4 Qc7 11.Bxd6 Qxd6 12.c4 0–0 13.a3N
13.cxd5 was played in a high-level encounter: 13…Qxd5 14.Nc4 Rd8 15.Qe2 Qh5 16.Rfd1 Nc6 17.a3 Ne7 18.Nfe5 Qxe2 19.Bxe2 Ne4 20.Rxd8+ Rxd8 21.Rd1 Rxd1+ 22.Bxd1 f6 23.Nd3 Bd5 24.Ndb2 e5 25.Bc2 Nc8 26.Kf1 1/2–1/2 (26) Xiong,J (2686)-Vidit,S (2727) Chess. com INT 2022 1/2–1/2 (26) Xiong,J (2686)-Vidit, S (2727) Chess.com INT 2022
13…Nbd7 14.cxd5 Nxd5 15.Nc4 Qc7 16.Rc1 Rac8 17.Qc2 h6 18.Qb2 Qb8 19.h3 N5f6 20.Be2 Bd5 21.Nd4 Rfd8 22.Rfd1 Ne5 23.Nxe5 Qxe5 24.Rxc8 Rxc8 25.Rc1 Qb8 26.Rc2 Rxc2 27.Qxc2 Qa8 28.Bf1 Qd8 29.Bd3 a5 30.Nb5 axb4 31.axb4 Bb7 32.Bf1
is evaluated as 0.19 by Komodo Dragon, but is obviously also drawn. This line was analysed at the following depth [d=56, 178373Mn]
The shocks continue when you look at what Komodo Dragon wants to do against the King’s Indian and Grunfeld!
1.d4 Nf6 2.Nf3 g6 3.c4 Bg7 4.e3
No London System here, but the quiet 4.e3 lines that have recently become extremely popular at the elite level! Komodo Dragon again considers this “offbeat” approach to be stronger than entering into the main lines with 4.Nc3 (after which it considers the Grunfeld to be more or less equal)
4…0–0 5.Be2 d6 6.Nc3 Bf5 7.d5
A very unusual move (7.0–0 has been played most often in elite games)
7…Ne4 8.Nxe4 Bxe4 9.0–0 Bxf3 10.Bxf3 a5 11.Rb1 a4 12.Bd2 Nd7 13.Qc2 Ne5 14.Be2 c6 15.dxc6N
15.e4 was played earlier in a correspondence game: 15…Qc7 16.Rfc1 Rfc8 17.Qd1 Qd8 18.Be3 h5 19.h3 Nd7 20.Rc2 Qe8 21.f4 cxd5 22.Qxd5 Nc5 23.Bf3 Rc7 24.Re2 Rb8 25.Rd1 b6 26.f5 Kh7 27.g4 Be5 28.fxg6+ fxg6 29.gxh5 e6 30.Qd2 gxh5 31.Rg2 Qf7 32.Qe2 Rg8 33.Bxh5 Rxg2+ 34.Kxg2 Qf6 35.Rf1 Rg7+ 36.Bg4 Qg6 37.Bxc5 1/2–1/2 (37) Kuntze,A (2384)-Hablizel,F (2393) ICCF email 2020
15…bxc6 16.c5 d5 17.b4 axb3 18.axb3 Qc7 19.Rfc1 Ra3 20.Bb4 Ra7 21.Bc3 Rfa8 22.b4 e6 23.h3 Ra2 24.Rb2 R2a7 25.Bd4 Nd7 26.Bxg7 Kxg7 27.b5 Ra5 28.Qd2 cxb5 29.Bxb5 Nf6 30.c6 Ne8 31.Qc3+ Kg8
is evaluated as 0.37 by Dragon.
How about 1.c4?
Just like AlphaZero, Komodo Dragon wants to meet 1.c4 with 1…e5, heading for a reversed Sicilian.
2.Nc3 is not Komodo’s main move but I was interested to know what it would do [d=49, 202677Mn] 2…Nf6 3.Nf3 Nc6 4.e4
A favourite of many engines and even AlphaZero too which I never really understood. For me, 4.e4 leads to very dull positions with few winning chances for White. 4…Bb4 5.d3 d6 6.a3 Bc5 7.b4 Bb6 8.Be3 Bg4 9.Be2 0–0 10.0–0 Bxf3 11.Bxf3 Nd4 12.Na4 Nxf3+ 13.Qxf3 Bxe3 14.Qxe3 Nh5 15.g3 f5 16.exf5 Rxf5 17.Nc3 Nf6 18.Rae1 (18.f3 was a high-rated correspondence game that ended in a loss…for White!) 18…a5 19.d4 axb4 20.axb4 Qd7 21.Qd3 Rh5 22.Re2 Ra3 23.dxe5 Rxe5 24.Rxe5 dxe5 25.Qxd7 Nxd7 26.Nb5 Rb3 27.Rd1 Nf8 28.Nxc7 Rxb4 29.Rd8 Kf7 30.Nd5 Rxc4 31.Rb8 Rc1+ 32.Kg2 Ne6 33.Rxb7+ Kf8 with an end evaluation of 0.10 (just a draw)
2…Nf6 3.Bg2 c6 4.d4 e4 5.Bg5 d5 6.Bxf6 Bb4+ 7.Nc3 Qxf6
8.e3 Be6 9.Nge2 0–0 10.cxd5 cxd5 11.0–0 Nd7
This is known theory and it’s pretty even!
12.Nf4 Bxc3 13.bxc3 Nb6 14.f3 exf3 15.Bxf3 Rac8 16.a4 Qg5 17.Qc1 Rce8 18.Re1 Nc4 19.e4 dxe4 20.Bxe4 Bd7 21.Nd3 Qxc1 22.Raxc1 b6 23.Bd5 Na5 24.Ne5 Bxa4 25.Ra1 Bc2 26.Nxf7 Rxe1+ 27.Rxe1 Bb3 28.Nh6+ Kh8 29.Bxb3 Nxb3 30.Ng4 Kg8 31.Re7 a5 32.Ne5 Nd2
with a 0.10 evaluation is the main line.
1.Nf3 obviously has many transpositional possibilities but also some original features:
Against the Reti, Komodo Dragon makes a choice that – in retrospect – is quite obvious!]
2.c4 d4 Again the most aggressive choice from Komodo Dragon, looking to exploit the drawbacks of White’s opening play. 3.e3 c5 4.b4 Nf6 5.b5 Qc7 6.g3 e5 7.Bg2 a6 8.0–0 Bd6 9.d3 0–0 10.a4 Re8 11.Na3 h6 12.Re1 Bf5 13.bxa6 Nxa6 14.Nb5 Qd7 15.exd4 cxd4 16.a5 Nc5 17.Ba3 Rxa5 18.Bxc5 Rxa1 19.Qxa1 Bxc5 20.Nxe5 Qd8 21.Qd1 g6 22.Nf3 Rxe1+ 23.Nxe1 b6 24.Nf3 Kg7 25.Ne5 h5 26.Nc6 Qd7 27.Ne5 Qd8 Komodo Dragon actually considers Black to be slightly better! (–0.07);
2.d4 would transpose back into Komodo Dragon’s 1.d4 lines.
2…c5 3.Bg2 Nf6 4.0–0 e6
A reversed 4.e3 system vs the King’s Indian!
5.c4 d4 6.e3 Nc6 7.d3 Be7 8.exd4 cxd4 9.Qe2 Nd7 10.Na3 e5 11.Nc2 0–0 12.a3 a5 13.Bd2 Bf6 14.b4 Re8 15.Qe4 h6 16.Rfe1 Be7 17.Qe2 axb4 18.axb4 Rxa1 19.Rxa1 e4 20.Nfxd4 Nxd4 21.Nxd4 exd3 22.Qxd3 Ne5 23.Qc3 Bxb4 24.Qxb4 Qxd4 25.Re1 Be6 26.Bc3 Nd3 27.Bxd4 Nxb4 28.Bxb7 Nc2 29.Rd1 Bxc4 30.Bc3 Na3 31.h3 Bb3 32.Ra1 Nb5 33.Ba5 Bc2
with a 0.13 evaluation which is of course just equal
Finally let’s round off this overview with a great line against all the 1.b3 cowboys out there!
1…e5 2.Bb2 Nc6 3.e3 d5 4.Bb5 Bd6 5.f4
6.g3 Qe7 7.Nf3 f6 8.Nc3 Be6 9.fxe5 fxe5 10.d4 e4 11.Ne5
This fascinating idea has been seen in the TCEC already: I even wrote a theoretical article about it for New in Chess and there’s a video about it on the Game Changer YouTube site: https://youtu.be/ISF-rFatf9c
Predecessor: 11…Bxe5 12.Bxc6+ bxc6 13.Qh5+ Qf7 14.Qxf7+ Kxf7 15.dxe5 Bh3 16.Ne2 Ne7 17.Nf4 Bg4 18.0–0 Bf3 19.e6+ Kf8 20.Nh5 Rg8 21.Rxf3+ exf3 22.Rf1 Ng6 23.Ba3+ Ke8 24.Rxf3 Nf8 25.Rxf8+ Rxf8 26.Nxg7+ Kd8 27.e7+ 1–0 (27) Kozionov,K (2470)-Deac,B (2625) Chess.com INT 2021
12.Nxc6 a6 13.Ba4 Nf6 14.Ne5+
14.Ne5+ c6 15.b4 Qxe3+ 16.Ne2 Ng4 17.Nxg4 Bxg4 18.Qd2 Qxd2+ 19.Kxd2 0–0 20.h3 e3+ 21.Kd3 Bf3 22.Rhf1 Rae8 23.Rxf3 Rxf3 24.Bb3 b5 25.a4 Bb8 26.axb5 axb5 27.Ra6 Rf6 28.c3 Kf7 29.Bc1 Ree6 30.Ra8 Bc7 31.h4 Rf1 32.g4 Bd6 33.Ra6 Bb8 34.Ra8 Bc7 35.Ra7 Re7 36.Ra6 Re6 37.Ra7 adjudication 1/2–1/2 (37) StockfishNNUE20200704 (3801)-Stoofvlees II a14 (3755) https://tcec-chess.com/#season 2020;
14.Qd2 0–0 15.0–0–0 bxc6 16.Ne2 Bd7 17.Kb1 Qh6 18.Rdg1 Rf7 19.h4 Qg6 20.Nf4 Qf5 21.Rf1 Ng4 22.h5 Raf8 23.Bc1 g5 24.Ng6 Qxf1 25.Rxf1 Rxf1 26.Nxf8 Kxf8 27.b4 Bxg3 28.Qg2 Rf3 ½–½ (55), Stockfish 202007032109 3807 – Stoofvlees II a14 3749, TCEC Season 18 (24) Top Of The Top Bonus, https://tcec-chess.com/#season (7.1),
14…c6 15.b4 Qxe3+ 16.Qe2 Qxd4 17.Nxc6 bxc6 18.Bxc6+ Kf7 19.a3 Rac8 20.Rd1 Qe5 21.Rxd5
21…Bxd5 22.Bxd5+ Kg6 23.Bxe4+ Kf7 24.0–0 Rxc3 25.Bxc3 Qxc3 26.Bd5+ Kg6 27.Rxf6+ gxf6 28.Qg4+ Kh6 29.Qh3+ Kg6 30.Qg4+
with a 0.00 evaluation.
Hope you enjoyed that and remember there is masses of extra detail in my YouTube series: Budapest, Albin, Grunfeld, King’s Indian, Semi-Slav and even some Morra Gambit… to name but a few! See you there! https://www.youtube.com/c/SiliconRoadChess
2 Comments on “Komodo Dragon’s Opening Repertoire”
Hi GM Sadler!
These are fascinating findings! Who would have guessed the London System 🙂
Do you perhaps have a similar analysis of Stockfish NNUE’s opening repertoire? I think it would be very interesting to compare Komodo Dragon’s repertoire with Stockfish’s.
Thank you for sharing your analysis with us!
Hey Jacob… well watch this space 🙂 Coming soon 🙂 Glad you enjoyed the Komodo Dragon article too and don’t forget to take a look at the video series on https://www.youtube.com/c/SiliconRoadChess Best Wishes, Matthew