Using tablebase is lock?

[Rubinus]

1+1
1 Deep Rybka 4.1 x64-b +41 +19/=58/-9 55.81% 48.0/86 (NalimovProbeFrequency=12)
2 Deep Rybka 4.1 x64 -41 +9/=58/-19 44.19% 38.0/86 (NalimovProbeFrequency=36 -default)
+-41ELO !

[mar]

Hi,

1) 1+1 => too fast
2) 86 => not enough games (means nothing)

Martin

[Rubinus]

Sure, but only free night.

[mar]

Sure, but only free night.

Ah ok I see

[lucasart]

Hi,

1) 1+1 => too fast
2) 86 => not enough games (means nothing)

Martin

Agreed

But in general, any good program should show almost zero elo difference with EGTB. I suggest you play 1,000 games with Houdini against itself, with and without EGTB.

Rybka however is an exception, because it's built in endgame knowledge is actually quite stupid. I remember seing it not even understand a KPK or KBPK draw without EGTB. Perhaps it has changed in the latest versions, anyway I don't use this program

So EGTBs are useless, but they keep the endgame play somewhat "cleaner". The point is that when you have a winning (very late) endgame, EGTB or not, you'll win it. And same for a draw. Counter examples exist, but their occurence frequency in real game is so minuscule that it doesn't give any elo.

PS: Don't forget EGTB are only used when there are up to 5 pieces on the board, including both kings. That's nothing... At this point the result is already decided anyway, and it's rare that engines will blunder at this point,