Correspondence ICCF Champion Vs Top Engine ?

[jr66]

ICCF player can use too an engine 24 h for a move if he want ?
It is not my cup of tea but i know some players do it !

[jr66]

Shame a new tournament will start very soon.
I would be ready to make a little test with more time available because i have no pride...if i lost ?

[Uri Blass]

Well, it is sure the biggest advantage in favor of CC player is databases and also some others factors which are less determinant year after year ( for example, i improved often SF 9 play but it is much more difficult now )
If you consider several games with openings choices open with no Book for engine alone, i think however the ICCF champion will win the match ???

No human can win even a single game out of hundred against unassisted SFdev even on a very moderate hardware. Bear in mind SF would have 120 hours of thinking time per move at least. That is on average depth over 70 from a starting position. It would be simply 100 draws in a 100 game match.
And databases are so overrated. In 120 hours per move there is virtually 0 chance engine would make a subpar move in an opening. Chances are much higher for a human though.

I think that the question is what is the longest time control that it is possible to win against unassisted stockfish with a single core and no book.

Fixed time may be different for different hardware but if you use a single core fixed hash size and
a fixed version of stockfish then it is better to make a competition who can produce a win with more nodes per second for stockfish.

with 1,000,000 nodes per move it is no problem to do it.
I guess you can do it also with 1,000,000,000 nodes per move.
The question is if you can do it with 10,000,000,000 nodes per second or 100,000,000,000 nodes per second or 1000,000,000,000 nodes per second.

[jr66]

By 1 000 000 000 nodes, do you mean 1 000 000 Kn/s ?

[Uri Blass]

By 1 000 000 000 nodes, do you mean 1 000 000 Kn/s ?

I say per move.

per second is dependent on the hardware.

1 000 000 000 nodes are of course 1000 000 knodes.

[jr66]

By 1 000 000 000 nodes, do you mean 1 000 000 Kn/s ?

I say per move.

per second is dependent on the hardware.

1 000 000 000 nodes are of course 1000 000 knodes.

Ok sorry.
So, with 10 000 Kn/s for example, you have your 1 000 000 000 nodes in 1'40" ....
If i am not wrong, in this conditions, it need about 28h per move for 1 000 000 000 000 nodes ?

[Uri Blass]

By 1 000 000 000 nodes, do you mean 1 000 000 Kn/s ?

I say per move.

per second is dependent on the hardware.

1 000 000 000 nodes are of course 1000 000 knodes.

Ok sorry.
So, with 10 000 Kn/s for example, you have your 1 000 000 000 nodes in 1'40" ....
If i am not wrong, in this conditions, it need about 28h per move for 1 000 000 000 000 nodes ?

Yes

You are not wrong.

[jr66]

So, your idea above seem good !
Of course, we need to determinate the conditions for assisted human too ?

[Milos]

Well, it is sure the biggest advantage in favor of CC player is databases and also some others factors which are less determinant year after year ( for example, i improved often SF 9 play but it is much more difficult now )
If you consider several games with openings choices open with no Book for engine alone, i think however the ICCF champion will win the match ???

No human can win even a single game out of hundred against unassisted SFdev even on a very moderate hardware. Bear in mind SF would have 120 hours of thinking time per move at least. That is on average depth over 70 from a starting position. It would be simply 100 draws in a 100 game match.
And databases are so overrated. In 120 hours per move there is virtually 0 chance engine would make a subpar move in an opening. Chances are much higher for a human though.

I think that the question is what is the longest time control that it is possible to win against unassisted stockfish with a single core and no book.

Fixed time may be different for different hardware but if you use a single core fixed hash size and
a fixed version of stockfish then it is better to make a competition who can produce a win with more nodes per second for stockfish.

with 1,000,000 nodes per move it is no problem to do it.
I guess you can do it also with 1,000,000,000 nodes per move.
The question is if you can do it with 10,000,000,000 nodes per second or 100,000,000,000 nodes per second or 1000,000,000,000 nodes per second.

1Mnps and 1h/move for an engine is probably borderline to have one win in 100 matches against unassisted SFdev.

[jr66]

I don't understand Milos sorry because it seem you speak only about engine vs engine match which is not the subject of the topic ?