Doubling of time control

[Laskos]

Great stuff. Maybe you could do the same for number of Threads to see at which point increasing processors has no elo gain.

This links to Stockfish and Komodo thread doubling runs:

http://www.fastgm.de/threads.html

I fitted for Komodo 9.3. I used Amdahl's law of speedup of a task at fixed workload: https://en.wikipedia.org/wiki/Amdahl%27s_law



The improvement practically stops at 2^6=64 threads. Going to 2^10=1024 threads, improvement is an insignificant 10 ELO points above 64 threads. I didn't take into account such things as NUMA with large number of nodes, which can deteriorate performance.

[Jouni]

Surprisingly the ancient rule of 75 ELO for double speed is still valid for 640 + 6,4 vs 320 + 3,2! But there was also theory that self-playing exaggerates differences - not valid anymore?

[duncan]

It says that at this 3700-3800 CCRL ELO level the doubling won't give any gain and draw rate becomes 100% for Komodo in self-play.

I assume even if this happens it is no evidence that chess is not a win in 75 (55?)moves.

[Laskos]

It says that at this 3700-3800 CCRL ELO level the doubling won't give any gain and draw rate becomes 100% for Komodo in self-play.

I assume even if this happens it is no evidence that chess is not a win in 75 (55?)moves.

Well, theoretically Chess might be even a Black Win from starting position. All I can say is that this is unlikely. The paradigm of Chess seems to follow closely the paradigm of Checkers. When Chinook started having 95%+ draw rates against top humans and 99% draw rate in self-play from the starting position, it took 10 or so more years to weakly solve Checkers as draw. It is more likely, if this capping of Chess at 400-500 more ELO points to current top engines is correct, that engines like Stockfish and Komodo already play non-losing Chess in say 5-10% of games from starting position. I don't believe the ways to win in perfect play the game of Chess are very rare or unique, more likely they are none. And the higher draw rate might indicate a real progress in solving Chess (again, like in Checkers). So, the capping in current paradigm might be not due to current paradigm, but to real progress in strength. It might be that this is the limit to weakly solved Chess as draw ftom starting position. It surely will take much longer than in Checkers, but I don't see a fundamental difference.

[Nordlandia]

Is it likely or unlikely white find himself in zugzwang from starting position?

[cdani]

The paradigm of Chess seems to follow closely the paradigm of Checkers. When Chinook started having 95%+ draw rates against top humans and 99% draw rate in self-play from the starting position, it took 10 or so more years to weakly solve Checkers as draw. It is more likely, if this capping of Chess at 400-500 more ELO points to current top engines is correct, that engines like Stockfish and Komodo already play non-losing Chess in say 5-10% of games from starting position. I don't believe the ways to win in perfect play the game of Chess are very rare or unique, more likely they are none. And the higher draw rate might indicate a real progress in solving Chess (again, like in Checkers). So, the capping in current paradigm might be not due to current paradigm, but to real progress in strength. It might be that this is the limit to weakly solved Chess as draw ftom starting position. It surely will take much longer than in Checkers, but I don't see a fundamental difference.

As with very long time control Stockfish & co rarely loses a game, will be nice to make an artificially stronger engine to test what can happen in the future.
So for example a version of Stockfish can be done that it reduces clearly less, so it will find more moves that standard Stockfish will ignore. Then we do a long time control match giving more time to the modified Stockfish to compensate, and see what happens. Ideally this match should be run in SMP mode, to be able to see if even with the widening of lazy smp the engines overlook many important moves.

To compensate, I suppose that one should give the modified version enough time to reach similar depths, but maybe there are better ways.

In fact I can do all this myself, but I think that explaining it here we can mature more the idea. Anyone interested on giving an opinion?

[duncan]

It surely will take much longer than in Checkers, but I don't see a fundamental difference.

more longer as in a couple of hundred/thousand years longer?

[duncan]

It says that at this 3700-3800 CCRL ELO level the doubling won't give any gain and draw rate becomes 100% for Komodo in self-play.

would you know about how many extra ply is needed to get to 3700-3800 CCRL ELO level with same evaluation. ?

[Robert Pope]

Is it likely or unlikely white find himself in zugzwang from starting position?

Very unlikely, but that's not at all the same as "isn't".

[MikeB]

This testing clearly shows a fundamental flaw in the Elo model.
The engine does not get weaker at long time control but stronger.
When given an hour to think, compared to one second, the move chosen with more time allowed will clearly be a much better move.

The increase in draws probably just shows that more careful chess is played by both sides at slower time control.

I've been crying myself hoarse for the last few years that testing through Blitz/STC games is NO SUBSTITUTE for testing through LTC games.
Glad to be vindicated.
Its not just a matter of engines being "more careful"....LTC games ruthlessly expose any weakness in the Evaluation and Search patterns of the Engine.
That is why in the TCEC, Houdini winning the Rapids is no guarantee that it will beat SF in the super-final ( though I hope it does, just to discomfit Adam ).
Adam has a point though, when he says that the Super-Final will be a different kind of ballgame.
Houdini can win only if it is REALLY the better Engine.

+1 I'm with you - the beauty in chess are in the deep searches...

edit: sorry - I think I just responded to a very old post that I thought was recent- just ignore - my bad ...