Different behavior with different hash sizes. (Stockfish 1.9

[jwes]

The obvious example where it helps is fine 70

Indeed I just did infinite analysis on Fine #70 with Rybka 4 (1-cpu for reproducibility) with 512 MB hash, then 256 MB hash.
Of course, the PVs and nodes were different.
I have yet to see a position where, in infinite analysis, PVs and nodes are the same for different hash sizes!

You should virtually never see the same node count with different hash sizes, but I am surprised that you do not see the same PVs.

[bob]

The obvious example where it helps is fine 70

Indeed I just did infinite analysis on Fine #70 with Rybka 4 (1-cpu for reproducibility) with 512 MB hash, then 256 MB hash.
Of course, the PVs and nodes were different.
I have yet to see a position where, in infinite analysis, PVs and nodes are the same for different hash sizes!

You should virtually never see the same node count with different hash sizes, but I am surprised that you do not see the same PVs.

I would agree, if you add "most of the time." If nodes change, then scores and PV can change. Although if you look at the order I just mentioned, that is given in order of decreasing probability.

[ernest]

You should virtually never see the same node count with different hash sizes,

That's what I mean, principally!

[Richard Allbert]

The hash code in 66 has a bad bug, IIRC, when looking retrieving hash table scores.

Maybe you fixed this in your v67

I know I did for a version I have (called v72), but rather than release it, I decided to start again from scratch with Jabba.

Regards

Richard

[Arsha Mahdavi]

The hash code in 66 has a bad bug, IIRC, when looking retrieving hash table scores.

Maybe you fixed this in your v67

I know I did for a version I have (called v72), but rather than release it, I decided to start again from scratch with Jabba.

Regards

Richard

Hello

Yes, I rewrote the hash code but it's a long time since I touched the Lime 67 code. I think it's 150 Elo better.
But I'm not sure.

PS: And about Jabba, When is your next release due?

[Richard Allbert]

Soon!

I sent Jabba 1.3 to Leo for WBEC, but it's a bit of a mess, so I need to clean things up a bit for the release so the code is something approaching readable.

Also, I have added threads over the Weekend for input (so no peeknamedpipe() interrupt code needed ) but only in windows, I need to figure it out how to do that in Linux before releasing. I took ages to understand the MS documentation.

It'll definitely be close to the best "engine with most code for least elo ".

It reads databases, generates stats, has auto epd testing, added Polyglot book format, magics, evaluation terms all in an array for autotuning, and a few other things.

All of which I don't use due to other things in life & lack of time

Hopefully by December.

Thanks for asking

Richard