Syzygy / egbb discussion

[Ryan Benitez]

In the past I have considered 6 piece TBs to be impractical compared to handling in eval. I am open to data showing otherwise.

Today at TCEC Houdini drew a game it would have won with six-piece tablebases, at least according to the comments there.

But do we know if the probes would have caused the engine to not reach that position or slowed it down? Even 6 piece bitbases wont fit in ram so the time investment in probing is not trivial. A test between the same engine using 5 piece bitbases and 6 pieces TBs would be interesting data. In the past Fruit does better with 5 piece scorpio bitbases than 5 piece TBs but I did not have the drive space for 6 piece TBs.

[Milos]

In the past I have considered 6 piece TBs to be impractical compared to handling in eval. I am open to data showing otherwise.

Today at TCEC Houdini drew a game it would have won with six-piece tablebases, at least according to the comments there.

But do we know if the probes would have caused the engine to not reach that position or slowed it down? Even 6 piece bitbases wont fit in ram so the time investment in probing is not trivial. A test between the same engine using 5 piece bitbases and 6 pieces TBs would be interesting data. In the past Fruit does better with 5 piece scorpio bitbases than 5 piece TBs but I did not have the drive space for 6 piece TBs.

This was root position, there is no overhead at root.

[Dirt]

But do we know if the probes would have caused the engine to not reach that position or slowed it down? Even 6 piece bitbases wont fit in ram so the time investment in probing is not trivial. A test between the same engine using 5 piece bitbases and 6 pieces TBs would be interesting data. In the past Fruit does better with 5 piece scorpio bitbases than 5 piece TBs but I did not have the drive space for 6 piece TBs.

In this case all it had to do was probe at root in order to win. This doesn't take any time at all. For Sysygy bases it wouldn't have needed the WDL part.

These cases are very rare, so it's almost impossible to measure the Elo gained from them. But it does show that modern engines aren't good enough to always find the win on their own, and tablebases can help.