Lack of up-to-date engines

[IWB]

Hello Tord

Oh dear. I was hoping that the list of fully-featured engines would increase, but with with Stockfish's removal it's down to three. One of which, Naum, is no longer in full-time development.

I would say that including Nalimov tablebases as a requirement for a "fully-featured engine" is rather unfair, as a significant fraction of all engines are simply not allowed to use them.

I would be much more satified with full 960 support

Bye
Ingo

[hgm]

I just want to bemoan the lack of up-to-date engines with the latest features.

How many current engines are multi-processor, 64-bit, fully UCI-compliant (such that they can be used for IDeA in Aquarium) and use Nalimov tablebases?

I know of only four: Rybka, Naum, Shredder, and StockFish.

Four!

Perhaps one of the Togas, but which?

Any others?

HIARCS - not currently 64-bit
Onno - not yet mp
Zappa, Junior - not fully UCI
Fritz, Crafty - opposed to UCI

Actualy, the number of fully up-to-date engines is zero. None of the engines you mention is written in Java, while this definitely is a modern trend!

[hgm]

I would be much more satified with full 960 support

Not to mention Gothic Chess!

[Trahald]

The bug with Chess960 in Stockfish has been fixed for the next version. For me too 960 support is the no 1 requirement for a 'full' chess engine.

I only wish Pro Deo could play it. I love it's style!

[rainhaus]

Oh dear. I was hoping that the list of fully-featured engines would increase, but with with Stockfish's removal it's down to three. One of which, Naum, is no longer in full-time development.

I would say that including Nalimov tablebases as a requirement for a "fully-featured engine" is rather unfair, as a significant fraction of all engines are simply not allowed to use them.

Some thoughts about it,

"full-featured engine" means the highest standard of an engine's equipment. This term was not yet in my vocabulary and I will certainly use it more often:) Of course it must be allowed and it's even necessary to compare all available engines on this high level, regardless of their various lacks and handicaps. The rating lists do the same. Therefore I wouldn't say it is unfair using an absolute level but it would be fair to name simultaneously the different starting positions for building a strong and comfortable engine.
From another point of view, if you want a full-featured engine you have to pay for it. Is there anyone who knows the price for a Nalimov license?
When the free engines and free GUIs become more and more stronger and are getting always more features, there must remain a little distinction to the commercials. Otherwise, one of them, free or commercial, reaches the point of absurdity.
Not concerning Marek, but when the World Wide Web was unleashed upon mankind there came a mentality "of getting all for nothing" over us. The user masses sometimes behave like an insatiable moloch who sucks and sucks and there is no reimbursement at all.
Back to the full-featured engine. From all engines I've used there is Spartan Thinker my favourite candidate for champion of the low-featured engines. Ok, it is available with 64 Bit, but it is also a master of camouflage. Why does he do so? Doesn't he like his chatty competitors ?
To conclude my long speech, a final word about the Tablebases. I may quote myself:
they seem to be essential in official tournaments where one game could decide the ranking but they probably have no significant statistical value for the rating lists. But there are no sufficient experiments or even scientific studies about it.

Rainer

[Marek Soszynski]

Lack of tablebase support affected one engine only (Stockfish). I was even prepared to suspend the requirement.

I believe that the current Sjeng isn't UCI compliant enough to run IDeA (though the next version will be).

My point was that despite all the engines out there, very few indeed incorporate all the latest, performance-improving, analysis-enhancing features. Remember this when you view the long lists of engines in the ratings tables.

[Mike S.]

I think that bitbases are a good alternative. The only important downsides I see is that mate distances are missing in the output, and (afaik) there are no 6-piece bb. yet. I read that 6-piece Shredderbases are almost ready but not released yet.

I use the Scorpio bb. (3+4+5) with Bright and Toga and my impressions from it are very good. From the engines I have (with many top engines missing though), Bright 0.4a with bb. scored best in the PET endgame test. - Stockfish scored very good too, even without any endgame tables, with the exception that I have multicore problems with Stockfish on my Pentium D CPU. But singlecore works fine and I don't go for top performances anyway.

Also, I think it's a good idea to have an engine use bitbases for the search as long as more pieces are on the board still, and if 6 or 5 pieces are left only, the interface can switch to Nalimovs, as the Fritz GUI does (with an option to deactivate that). I think the Shredder engine and -interface does it too, in a similar way.

As for the general set of engine features, I would include multi-pv. For me, this is more important than multicore and 64 bit-capabilities, which are more directed at "high performance demand" like online engine competition or Freestyle tournaments. - Analyses, but also GM chess kibitzing, is much more enjoyable if an engine can display e.g. 2 or 3 alternatives at the same time without requiring user interactions like "next best" etc. If human games are kibitzed, maximized engine speed is not required anymore with any "normal" computer. And for the analysis of my own patzer games neither!

[mcostalba]

I use the Scorpio bb. (3+4+5) with Bright and Toga and my impressions from it are very good. From the engines I have (with many top engines missing though), Bright 0.4a with bb. scored best in the PET endgame test. - Stockfish scored very good too, even without any endgame tables

Stockfish has large endgame knowledge hardcoded in, this allow to play good endgames also without bitbases.

Perhaps with bitbases support it would be even stronger, but is not so obvious because in comparison with, say, Toga, SF has already the necessary knowledge to play endgames perhaps not with 100% accuracy, but in many cases a good knowledge plus a deep search end up to be enough to win an endgame starting from an advantage position.

[jdart]

Arasan version 10.0 and higher supports UCI, SMP, 64-bit, and Nalimov tablebases.

--Jon

[jdart]

I think lack of distance to mate is a significant limitation, although how important it is in practice is not clear.

My understanding is Bruce Moreland did a cleanroom implementation of tablebase code for Ferret that didn't depend on Nalimov's code. So that's not impossible but it would be a difficult task.

--Jon