Stockfish TCEC6 PA_GTB

[Jeremy Bernstein]

By popular demand... Stockfish TCEC6 PA_GTB is released, following in the venerable footsteps of its predecessors. Based on SF 24052014 (not exactly the version playing in TCEC6, but close), and with all of the PA_GTB features you've come to know and love:

- Persistent Analysis (back and forth browsing of a game provides stable evals)
- Gaviota Tablebase support
- Permanent Hash/Position Learning
- other minor stuff

No changes to the goodness, just updated to the the upstream master. Please let me know how you get on...

Jeremy

[User923005]

I am giddy with excitement over this.
Thank you for your efforts.

[User923005]

Look at how this new version accelerates analysis:

[ernest]

Look at how this new version accelerates analysis:

Can you explain a little more for all the morons ?

[ChessDrone]

Look at how this new version accelerates analysis:

Can you explain a little more for all the morons ?

User923005 that wants to point out is the difference, the comparison of the datas (Depth/times, Depth/nodes, nps) between SF-DD and SF-TCEC6 on same machine with same position and same conditions.
In the absence of this, we must do it alone.

[ernest]

In the absence of this, we must do it alone.

[User923005]

Look at how this new version accelerates analysis:

Can you explain a little more for all the morons ?

Notice from the image that I achieve 32 ply in one minute and 16 seconds.
My hardware is good, but not that good.

On the other hand, I have 10,000,000 positions analyzed to 20 ply or deeper in my stockfish.hsh file, which I imagine that other people probably do not have

Directory of C:\chess\hash
2014-05-24 06:23 AM 2,516,038 stockfish-TCEC6-64-PA_GTB.exe
2014-06-09 12:38 PM 536,870,912 stockfish.hsh

Using it in this way requires deep analysis. It takes 20-50 seconds to load the stockfish.hsh file (from SSD at that).
If you were doing analysis at 30 seconds, half or more of the positions would simply fail.

[brisas2k]

Hello fellow computer chess fans,

I have accumulated a "chess experience" in my dd_pa_gtb hash table of 20 mb, in about 8 months worth of rapid an long control times. With the tcec version it has become a godzilla of chess. (It probably helped speed rup the recent pawn and uci mods in the latest mods on stockfish, since i was beating the fishes with intel i7's, 4 and 6 cores at infinity chess, with my humble laptop at 1.8 ghz).

I would be interested in playing against other users of pa gtb versions of stockfish.

I use ancient hardware (AMD athlon X2, 2.4 ghz, very similar to CEGT test site)If anybody wants to test their hash "learning" against mine, we can meet at http://www.fics.com and test these "learning hashes".

While i have very old hardware, My rig routinely defeats i7 4 and 6 cores with 24 gb of ram, and 3.x ghz computers. i have to warn you: i have recently defeated an intel xeon 24 cores @ 3.4ghz, in 90-15 tc at http://www.infinitychess.com (that's the power of permanent hash: Thanks JB. Your engine is an unmatched gift).

So, we might get to watch some very interesting games. Plus the "refinement" that comes from having a permanent hash improved by other hashes "experience", is invaluable, for both analysis and computer engine matches.

If it sounds interesting, meet me at FICS, on saturdays and sundays from 1800 GMT. Are you game? Who knows, maybe we can field a nice tournament over there. You can find me there wit the nick centurionchess.

Greetings,

JMBrisas

[echai]

Post subject: Marco steps down as Stockfish maintainer Posted: Wed Jun 18, 2014 7:13 pm Reply to topic Reply with quote
Marco Costalba has stepped down as maintainer of Stockfish :

https://groups.google.com/forum/m/?from ... sYuyij6uqQ


"I have decided to step down as Stockfish maintainer

I have less and less time and I am tired on engaging in long and annoying discussions every time a patch passes LTC (or even when it doesn't).

Sometime I hope that a patch does not pass LTC for the trouble that it will get, of course this cannot be the right mind approach of a maintainer, so I am no more suitable for this job.

To maintain a project like SF requires a lot of work, especially if you want to remain strict on the quality of the stuff you apply, it requires much less work if you apply anything that passes, but this is not my aim and I will never accept it.

Lately this work became a real pain.

I ask you, as a community, to fork my repo and continue this effort on that. I will keep my repo active as private development, where I will add what I like according to my own judgement. Of course I will give up submitting tests on fishtest because, as the rule says, only official dev branch can be tested in fishtest. Possibly I will apply some good (according to my only judgment) patch from master repo, giving credit to the author, until the two repos will diverge so much that back porting without testing becomes too risky and eventually my repo will die of natural death.

Probably a new development model will raise, and I wish the new one to be better than the old one, although I strongly doubt that applying anything is a good policy in the long term. But this is no more a decision belonging to me, nor a decision I'd want to be involved.

Now Stockfish is the strongest engine in the world. This is much more than what I had dream of when I forked Glaurung in 2008, and I really cannot ask more from this journey in chess engine world.

Have fun
Marco
"

Some background may be found here with the details of his latest dispute with other Stockfish developers :

https://github.com/mcostalba/Stockfish/pull/243

[Jeremy Bernstein]

Yikes, but this conflict seems to have be preprogrammed, given the friction between Marco as the maintainer and several developers who are interested in seeing their ideas incorporated in SF's codebase. I think Marco is making the right choice here, eliminating the agony of having to say "no", and to justify that "no", at every crossroad. If the SF codebase is to remain clean and easy to understand (and it is), permitting complex and ugly patches like Ron Britvich's is a step in the wrong direction for sure**. The maintainer has a responsibility to the code, not just to the developers, and I think it's fully acceptable to raise the drawbridge, fill the moat with sharks and make that curatorial role explicit, especially when dealing with limited available time and energy for the project.

** It might be that Ron's patch increases Elo by +5-9, but at what cost? First of all, truncating the hash key to 16-bits is a huge, huge modification and needs massive testing, and not just self-testing at blitz TC. The other stuff like bitpacking, etc. might be all fine, but it significantly deterioriates the legibility of the code. I assume that Ron put a lot of work into the original development, revisions based on Marco's initial response, testing, etc., and I sympathize with his frustration, but I suspect that I would have had exactly the same reaction as MC did.

EDIT: looks like Gary Linscott has created a new official Stockfish fork and the initial battles for maintainer position(s) and whether support for Syzygy TBs will be integrated into the official repos are underway. I wish the team the best of luck moving forward.