Status Quo, Stockfish 170315 vs Komodo 1400

[nabildanial]

Komodo really got wrecked, its unbelievable. Current result is +11 =37 -5 to Stockfish.

[Hai]

Komodo really got wrecked, its unbelievable. Current result is +11 =37 -5 to Stockfish.

Points 29.5
Games 53
Winning percentage 55.66
Elo difference +40

PS. they need more machines, to much tests can explode the framework
http://tests.stockfishchess.org/tests

[zullil]

I wonder how the Stockfish developers may think about this thread.
Their "baby" plays an outstanding match performance, and so many complains about "blunders" and "fails"....
If I would be a Stockfish fan I would have composed its own hymn and dance around full of joy....

It's great to see SF finally making real use of more than 8 cores. The SMP improvements since SF 6 have been terrific. Kudos to the team, and Joona especially.

[Lyudmil Tsvetkov]

Your time control is also much better to view than 120' + 30", they should change this into 90' + 15" or even 60 ' + 15".

Noo, quality is quality, a precious commodity, and longer time control means exactly that: better chess games.

This match is a rapid challenge, very nice, especially with the big number of cores, but longer TC games should never be dropped. There is place and time for everything.

Specifically for this match, we already know that:

- according to Bertil's tests posted on SF forum, SF loses 35 elo points to Komodo going from 16 to 24 cores on Intel machines
- SF performs significantly better than Komodo at rapid TC, many rating lists in the past have shown this

Komodo usually performs better with longer time.

So, for this specific match, Komodo is favoured by the use of 24 real cores, while SF is maybe even more favoured by the rapid TC.

[Modern Times]

Your time control is also much better to view than TCEC 120' + 30", they should change this into 90' + 15" or even 60 ' + 15".

I agree, I find TCEC time control too long to watch live. But the point is you don't have to I guess - just play through them later. So you can choose what suits you best.

[beram]

Your time control is also much better to view than 120' + 30", they should change this into 90' + 15" or even 60 ' + 15".

Noo, quality is quality, a precious commodity, and longer time control means exactly that: better chess games.

This match is a rapid challenge, very nice, especially with the big number of cores, but longer TC games should never be dropped. There is place and time for everything.

Specifically for this match, we already know that:

- according to Bertil's tests posted on SF forum, SF loses 35 elo points to Komodo going from 16 to 24 cores on Intel machines
- SF performs significantly better than Komodo at rapid TC, many rating lists in the past have shown this

Komodo usually performs better with longer time.

So, for this specific match, Komodo is favoured by the use of 24 real cores, while SF is maybe even more favoured by the rapid TC.

I agree on fact that chess quality is always better at longer time control
But two corrections I want to make
1) This match is not rapid
2) Komodo doesnt perform better against SF at longer time control
The only thing that occurs at LTC is increasing draw rate so win percentage diminishes because of that
The win-loss ratio's (SF wins:losses without the draws) stay overall the same

Compare for instance the latest SF dev results by Stefan Pohl at blitz (win/loss ratio = 2) with this LTC match at 24 cores (after 52 games w/l=2) or with 40/20 CEGT games (w/l =2,67)

So in all these different TC SF vs K matches (apart for the draws) for every lost game SF wins 2 or more

[Lyudmil Tsvetkov]

Your time control is also much better to view than 120' + 30", they should change this into 90' + 15" or even 60 ' + 15".

Noo, quality is quality, a precious commodity, and longer time control means exactly that: better chess games.

This match is a rapid challenge, very nice, especially with the big number of cores, but longer TC games should never be dropped. There is place and time for everything.

Specifically for this match, we already know that:

- according to Bertil's tests posted on SF forum, SF loses 35 elo points to Komodo going from 16 to 24 cores on Intel machines
- SF performs significantly better than Komodo at rapid TC, many rating lists in the past have shown this

Komodo usually performs better with longer time.

So, for this specific match, Komodo is favoured by the use of 24 real cores, while SF is maybe even more favoured by the rapid TC.

I agree on fact that chess quality is always better at longer time control
But two corrections I want to make
1) This match is not rapid
2) Komodo doesnt perform better against SF at longer time control
The only thing that occurs at LTC is increasing draw rate so win percentage diminishes because of that
The win-loss ratio's (SF wins:losses without the draws) stay overall the same

Compare for instance the latest SF dev results by Stefan Pohl at blitz (win/loss ratio = 2) with this LTC match at 24 cores (after 52 games w/l=2) or with 40/20 CEGT games (w/l =2,67)

So in all these different TC SF vs K matches (apart for the draws) for every lost game SF wins 2 or more

Not quite, SF excels at rapid even more than at blitz, from what I have seen, this seems some kind of optimum TC for SF, but I have no reference lists at hand.

[Hai]

Positional play is more complex than tactical play.
Tactical is about: is there something or nothing (to exchange or win...).
Positional play is about: how much positional knowledge an engine has.
= You need more time for positional play.
= You need less time for tactics.
= An Engine which play more positional will get a nice boost from more time.

Example
60'+15" Stockfish is 100%/Komodo is 80% save against opponents tactics.
120'+30" Stockfish is 100%/Komodo is 100% save against opponents tactics.
=Tacticaly Stockfish gets nothing from more time.
=The engines can now only win with positional strength.

60'+15" Stockfish has 30% / Komodo 50% of perfect positional knowledge.
120'+30" Stockfish got +10% / Komodo got +20% of better positional play.
No wonder that Stockfish lost last TCEC with +4 -7.

[Lyudmil Tsvetkov]

Positional play is more complex than tactical play.
Tactical is about: is there something or nothing (to exchange or win...).
Positional play is about: how much positional knowledge an engine has.
= You need more time for positional play.
= You need less time for tactics.
= An Engine which play more positional will get a nice boost from more time.

Example
60'+15" Stockfish is 100%/Komodo is 80% save against opponents tactics.
120'+30" Stockfish is 100%/Komodo is 100% save against opponents tactics.
=Tacticaly Stockfish gets nothing from more time.
=The engines can now only win with positional strength.

60'+15" Stockfish has 30% / Komodo 50% of perfect positional knowledge.
120'+30" Stockfish got +10% / Komodo got +20% of better positional play.
No wonder that Stockfish lost last TCEC with +4 -7.

SF now improved SMP up to 16 cores, so there will not quite be a repetition of last TCEC.

But your main point is true.
SF, that relies more on attacks than positional subtleties, might lose some of its strength in relation to Komodo at very LTC.

But no development version is ever the last one.

At the end of the day, it is about which search and eval terms scale well at long and very long TC, and which not.

[M ANSARI]

Great to see SF finally addressing its SMP issue. No doubt this was the Achilles heel for SF during TCEC. The next TCEC should be very interesting but I have a feeling this time SF will win convincingly. Incredible how fast SF has gained strength and this does not look that it will stop anytime soon.