Glaurung 2 - epsilon/3

[Tord Romstad]

Hello Marc,

You are entirely right: It turns out that Glaurung 2 - ε/3 has a catastrophic pondering bug. It is very embarassing that I didn't notice this myself: Clearly I should play at least a single test game before releasing a new version.

The good news is that the bug is very easy to fix (it took me about 30 seconds to find and fix the bug). The bad news is that it may take a few days before I am able to release new binaries for all platforms. New source code and Mac OS X binaries will be ready very soon, Windows and Linux binaries might take some time (because I don't compile them myself).

The bug is only relevant when pondering is on. Without pondering, 2 - ε/3 should work fine.

Tord

[Edward German]

Glaurung 2 - ε/3, the latest development version of the program which will eventually be released as Glaurung 2, is now ready for download from my web site. At the moment, only the source code and executables for Mac OS X and Linux are included. 32-bit and 64-bit Windows executables will follow soon. The most important new feature is the addition of parallel search: Glaurung can now use up to 4 CPUs (or so I hope; I haven't tested it with more than 2).

On single-CPU computers, there is no significant difference between Glaurung 2 - ε/2 and Glaurung 2 - ε/3. There are a few very minor bug fixes and optimizations, but nothing which is likely to have a measurable impact on playing strength. I therefore advice those of you who spend lots of effort on making rating lists and have already started testing 2 - ε/2 to skip the new version, or to let it replace the old version and inherit its rating.

In more detail, the changes in this new version are (taken from the README file):

• Parallel search, with support for 1-4 CPUs. The program currently always allocates a separate pawn hash table and material hash table for four threads, which is a pure waste of RAM if your computer has just a single CPU. This will be fixed in a future version.
• Fixed a bug in book randomization. When using Polyglot books, the previous version would always select exactly the same move in the same position after a restart of the program. Thanks to Pavel Háse for pointing this out.
• Fixed a UCI pondering bug: Glaurung no longer instantly prints its best move when the maximum depth is reached during a ponder search, as the previous version did. According to the UCI protocol, it is not allowed to print the best move before the engine has received the "stop" or "quit" command.
• Additional search information: The new version displays hash saturation and the current line(s) of search.
• Several minor bug fixes and optimizations in the search and evaluation.

Glaurung 2 - ε/3 is probably the version that will play in the OPCCC in the end of the month.

Tord

Hello Tord!

Thanks for this nice message.

I was little worry after Epsilon 2, then I do not see a window for settings. Why have Epsilon 2 no Window to tuned a own setting?

Best!

[Tord Romstad]

I was little worry after Epsilon 2, then I do not see a window for settings. Why have Epsilon 2 no Window to tuned a own setting?

Hello Eduard,

You are right, there are no configurable parameters in the current development version of Glaurung 2. This is simply because the program is still very far from finished; many features are still missing. Later versions will add some configurable parameters, and many new features.

Tord

[Edward German]

I was little worry after Epsilon 2, then I do not see a window for settings. Why have Epsilon 2 no Window to tuned a own setting?

Hello Eduard,

You are right, there are no configurable parameters in the current development version of Glaurung 2. This is simply because the program is still very far from finished; many features are still missing. Later versions will add some configurable parameters, and many new features.

Tord

Hello Tord!

Thank You for future configurable parameters.

I have installed now epsilon4 W32 on my P3 600 MHz PC with Win 98.

Unfortunately epsilon4 runs under Fritz 8 very slow (slower than epsilon 2) and I can not see kn/s.

Perhaps a BUG?

Best!

[Dirt]

Later versions will add some configurable parameters, and many new features.

Tord

Later versions? You're already up to 2 - ε/4, is there no limit?




Ok, it's a lame joke, but I couldn't not do it.

[PauloSoare]

Thanks Tord, your engine is very strong in endgames. I think
that the new Glaurung are a top endgame engine.

Paulo Soares

[Tord Romstad]

I have installed now epsilon4 W32 on my P3 600 MHz PC with Win 98.

Unfortunately epsilon4 runs under Fritz 8 very slow (slower than epsilon 2) and I can not see kn/s.

Perhaps a BUG?

Hard to say - I haven't seen anybody else report this problem. But on your P3 600, I don't think there is any reason to use version 2 - ε/4 at all. The only real improvment in the new version is the addition of parallel search, which is unusable on your single-CPU computer. You should probably just continue using 2 - ε/2.

Tord

[Tord Romstad]

Hello Paulo!

Thanks Tord, your engine is very strong in endgames. I think
that the new Glaurung are a top endgame engine.

That's somewhat surprising, because the program does not use bitbases or tablebases (apart from a single small bitbase for KP vs K endgames), and has very little actual endgame knowledge. But perhaps you are still right; in the games I have seen I have noticed that my new program is able to squeeze out surprisingly many wins in the endgame. I guess the search must somehow be very efficient in the endgame.

Tord

[PauloSoare]

Hello Paulo!

That's somewhat surprising, because the program does not use bitbases or tablebases (apart from a single small bitbase for KP vs K endgames), and has very little actual endgame knowledge. But perhaps you are still right; in the games I have seen I have noticed that my new program is able to squeeze out surprisingly many wins in the endgame. I guess the search must somehow be very efficient in the endgame.

Tord

Tord, in these tests I did not use any aid of
tablebases or bitbases.
And all programs with only 1 thread.

Paulo Soares

[Tord Romstad]

Tord, in these tests I did not use any aid of
tablebases or bitbases.

In that case, you should be aware that Glaurung, and possibly other programs, are cheating. Glaurung does not read any bitbases from disk, but it computes a KP vs K bitbase during program initialization. There is no way to disable this feature. Other programs might do the same, possibly also with other bitbases than KP vs K.

Tord