Dragon by Komodo Chess

[lkaufman]

Larry,

Curious, have you given any thought of going up against a strong human GM with Dragon set to Armageddon mode as Black and the human as White? The GM would basically get draw odds with White, if he can get the draw, while Dragon would go all out attacking with Black. It would be an very realistic and exciting handicap.

Back in 2007, Rybka defeated GM Joel Benjamin with those conditions (draw and White odds) 6 to 2 in standard time limit games. Perhaps coupled with enormous time odds (bullet vs classical tc with no ponder) it might be reasonably fair vs. Magnus Carlsen if he wanted to play. I do think it is a very interesting handicap, it's just not enough to give even top GMs much of a chance without a big "kicker" as suggested.

[lkaufman]

Hi Larry, Is there any chance of NNUE specialized for losing positions for use in handicap matches?

We actually did include some such positions in the training which might help a bit in such matches, probably not a lot though. The problem is that if the handicap is high enough to give humans a good chance (like knight odds or even two pawns), the stronger side almost always wins in self play so it can't really learn how to trick the opponent. I suppose we could train it by playing against a crippled version of itself, but that might mess up normal play so we haven't tried that. We are considering possible opponents for such a match. I wonder, which players would readers most like to see facing Dragon in a handicap match (other than Magnus Carlsen who has never shown interest in playing such matches with engines)?

How does Dragon fare at giving knight odd in slow bullet? My dream is to see Hikaru Nakamura against Dragon at knight odds in many games 2 '+ 1 " in a row, like 12 or 24 games.
While perhaps Hikaru would still be the favorite, it could be a periodic event to track Komodo's progress.

I believe that Dragon is the best engine now for giving knight odds, but I think Hikaru is too strong for knight odds at 2' + 1". Maybe 1' + 1" knight odds might be about fair. We don't have much data yet, Dragon is too new. I can say that at Rapid (15' + 10"), I think it could maintain a chess.com rating close to 2500 giving knight odds, which is probably something like 2400 FIDE.

[Madeleine Birchfield]

Just read this on Twitter.

Matthew Sadler is wrong; Komodo hasn't been updated to the Dragon version for the FRC2 bonus according to that website.

[MikeB]

<snip>
...
Hikaru Nakamura is an obvious one; he played against Rybka and Houdini without handicaps a decade ago and managed to draw both of them. His streaming platform on Twitch would also bring more interest into computer chess and your new Dragon engine, and he is affiliated with chess.com.

You would think that the stars are aligned with all those points ...

[Chessqueen]

A short test Dragon NNUE. You still have problems with time, unfortunately the same bug is in Komodo.

http://www.mediafire.com/file/zwmyya23n ... E.pgn/file

The file shows a ten game match with Lc0 with a score of one win each and eight draws, and the sole loss was by adjudication, not time. Could you please explain what you think is the problem with Komodo or Dragon time management? It's not obvious from the file.

Please clarify does Dragon NNUE performs better in shorter time control Blitz, or the other way around using standard time control?

[lkaufman]

<snip>
...
Hikaru Nakamura is an obvious one; he played against Rybka and Houdini without handicaps a decade ago and managed to draw both of them. His streaming platform on Twitch would also bring more interest into computer chess and your new Dragon engine, and he is affiliated with chess.com.

You would think that the stars are aligned with all those points ...

I don't remember these games from a decade ago. Could you remind me of the details, like where played, how many games, time limits, score, hardware, and any other relevant factors?

[pohl4711]

We have released Dragon by Komodo Chess at komodochess.com today. It uses the NNUE technology (but not the code) developed for the game of shogi. The improvement in strength over the latest standard Komodo, version 14.1, is enormous, roughly 160 elo on four threads, 190 elo on one thread, based on 2' +1" matches.

7000 games testrun of KomodoDragon 1.0 is finished. Versus 7 engines, I can confirm that progress...
Look on my website:

https://www.sp-cc.de

Code: Select all

KomodoDragon 1.0 avx2    : 7000 (+3986,=2797,-217), 76.9 %

Slow Chess 2.4 popc        : 1000 (+665,=323,- 12), 82.7 %
Ethereal 12.75 avx2        : 1000 (+607,=381,- 12), 79.8 %
Nemorino 6.00 avx2         : 1000 (+493,=496,- 11), 74.1 %
Stockfish 12 200902        : 1000 (+ 65,=784,-151), 45.7 %
Xiphos 0.6 bmi2            : 1000 (+757,=235,-  8), 87.5 %
Fire 7.1 popc              : 1000 (+758,=235,-  7), 87.5 %
Houdini 6 pext             : 1000 (+641,=343,- 16), 81.3 %

[Cornfed]

Still about 80 points behind a similar current Stockfish. Not bad.

[mwyoung]

Still about 80 points behind a similar current Stockfish. Not bad.

At 5m+5s on a 32 thread 2950x. Dragon is doing much better then 80 Elo. Very good for the first release of Dragon.

Mark L. asked my to change the default setting for hash memory from default. I do not know if this changed the results. Default was 64, and was changed to 512. The suggested best setting for this many threads.

Here is the results playing the best Stockfish.

Dragon by Komodo Chess vs Stockfish 101120 (TC=5m+5s)

With 198 games played so far. Dragon vs Stockfish 101120 +6 =172 -20. 46.5% TP = -25 Elo.

Live Stream

[Chessqueen]

Still about 80 points behind a similar current Stockfish. Not bad.

At 5m+5s on a 32 thread 2950x. Dragon is doing much better then 80 Elo. Very good for the first release of Dragon.

Mark L. asked my to change the default setting for hash memory from default. I do not know if this changed the results. Default was 64, and was changed to 512. The suggested best setting for this many threads.

Here is the results playing the best Stockfish.

Dragon by Komodo Chess vs Stockfish 101120 (TC=5m+5s)

With 198 games played so far. Dragon vs Stockfish 101120 +6 =172 -20. 46.5% TP = -25 Elo.

Live Stream

I read on TCEC that they have NOT received the Komodo Dragon version which they ae interested in testing it on FRC very soon, I guess that Mr. Kaufman has NOT sent them the program yet.