Komodo 1339 64-bit 8CPU v Stockfish 221214 64-bit 8CPU Match

[Modern Times]

I think in TCEC the main reason Komodo beat SF was not due to EGTB's but rather due to the fact that at LTC Komodo seems to scale better.

This was the same time control as TCEC. But on half the cores, 8 compared to 16, and the Intel TCEC hardware is better than AMD of course.

[Modern Times]

The most likely explanation for Komodo's advantage in TCEC is that its SMP implementation makes better use of 16 cores than Stockfish.

[Hai]

I have the feeling that Stockfish on 8 cores is 35-45 elo ahead of Komodo.
And Komodo on 16 cores is 35-45 elo ahead of Stockfish.

[M ANSARI]

I do think running all rounds of TCEC without TB was silly.

Silly, or downright stupid ???

I think it is harsh to call it stupid as there is something to have an engine play well without the help of EGTB's. But I would have also preferred that EGTB's should have been used, not because I think it would help one engine more than another ... but rather because I think it would push forward better implementations in engines to be more efficient in handling EGTB's. You have to give engine developers a reason to work hard in employing better code to use EGTB's and the hardware required for EGTB's today is really nothing special. I would also want to encourage development of use of daughterboards that can understand fortresses and find on the fly solutions to 7 or even 8 piece EG's via brute force or Monte Carlo. Engines that improve in such a manner are very useful in analysis of chess positions.

[Modern Times]

I have the feeling that Stockfish on 8 cores is 35-45 elo ahead of Komodo.
And Komodo on 16 cores is 35-45 elo ahead of Stockfish.

I'm not sure about the numbers, but in principle yes I agree. If Stockfish 6 and Komodo 9 were released today, I think Stockfish would top most of the ratings lists, because most are less then 8 cores.

[beram]

I think in TCEC the main reason Komodo beat SF was not due to EGTB's but rather due to the fact that at LTC Komodo seems to scale better. Also it is possible that given more time Komodo simply has more time to work out the tactics that SF throws at it. There were a few games in TCEC where it seemed that SF was going to tactically blow Komodo out of the water, but somehow Komodo managed to find incredible defensive resources that saved the game and in some cases reversed the result. It just seems that SF strength is tactics while Komodo is more of a positional "squeeze" player ... SF can go deep in a position very very quickly and that can be lethal even against a very strong positional player. All it takes is to see a tactical shot one ply deeper and that is enough advantage to win a game with any of these powerful engines. But given enough think time on strong hardware, Komodo seems more than capable of blunting any tactical fireworks.

When going through the won games you cannot say less than that SF strength is especially in kingside attack (and endgame) and that Komodo has been outplayed in this match.
SF lost four games in this match and just one was because of a kingside attack while on the other hand SF won six games with a kingside attack.
And most of these attacks Komodo didnt saw coming not just one move but three moves later, see for instance game 21 and 41 both won in nice attacking way with black by SF.
It is not about seeing a tactical shot one ply deeper, but about building up an attack against the king.
There is just one nicely deserved win by Komodo and that is with the kingside attack in game 2. The other three where because of opportunistic exchange sac by SF in game 5. An unlucky endgame loss game 23 (with both two rooks queens and two connected passers) and because of bad opening in game 19

It is really fun to replay, thx again Ray and Graham

[M ANSARI]

I agree that in this match SF was stronger as it was able to tactically outplay Komodo, but in TCEC this was not the case. Komodo clearly felt like the stronger engine. Not sure if it is due to better scaling to 16 cores or due to being able to "see" deep enough in a position to negate SF superior tactical ability. I see positional chess as a translation or short cut to deep tactics, but you still have to see deep enough to avoid any tactical tricks that can destroy your positional advantage. What you call impressive king attacks is just deep tactics for the engine. On 16 cores Komodo seemed just as capable at handling deep tactics as SF, but I felt it had a superior static evaluation that allowed it to win more games and thus be stronger. This obviously is not the case on 8 cores.

[Matthias Gemuh]

... What you call impressive king attacks is just deep tactics for the engine. ...

No ! Every strong engine has static or dynamic king attack affinity values.

[Matthias Gemuh]

... What you call impressive king attacks is just deep tactics for the engine. ...

No ! Every strong engine has static or dynamic king attack affinity values.

[fenchel]

Exciting stuff, many thanks to you for the commentary, and to Ray for running this.

Comparing this and your previous tournament to TCEC, it seems the main things distinguishing SF and K are (a) 16 core play, (b) endgame play (with/without TB), (c) imbalance eval.

(Of course, small sample blah blah.)

I do find it odd that tablebases were not used in the TCEC final. It is the de-facto world championship, showcasing the two top engines at their very best. Sure leave them out in some of the intermediate rounds to satisfy some people's curiosity, but not the final. Did it affect the result ? Who knows.

Ray, do you happen to know of a any test between TB and no-TB engines? E.g., SF dev with no syzygy vs SF dev + 6-man vs SF dev + 5-man?

By test I mean really anything; e.g., a nice tournament like the one here, or a ELO-precise test of many quick games, or whatever.

For some reason, I can only find anecdotes about this comparison?

(Thanks either way, and sorry if I've missed some basic homework.)