Glaurung 1.2
Posted: 06 Jul 2006, 01:01Hi all,
I am pleased to announce that Glaurung 1.2 is finally available. So far, I only have the source code and a Mac OS X executable. Windows binaries will follow later. This is probably the Glaurung version that will play in Mainz next month.
Here are the main changes compared to Glaurung 1.1 (I hate when authors release a new version and don't tell us anything about what they have done):
Tord
I am pleased to announce that Glaurung 1.2 is finally available. So far, I only have the source code and a Mac OS X executable. Windows binaries will follow later. This is probably the Glaurung version that will play in Mainz next month.
Here are the main changes compared to Glaurung 1.1 (I hate when authors release a new version and don't tell us anything about what they have done):
- Support for up to 4 threads. In fact, it is even possible to compile an executable with support for an even higher number of threads, but at the moment Glaurung isn't very efficient with more than 4 threads. Bob Hyatt kindly ran some tests on the 8-CPU Opteron he used in the WCCC, but it turned out that Glaurung was only very slightly faster with 6 threads than with 4, and actually a bit slower with 8 threads. If I ever get access to such a monster machine myself, I'll try to improve this.
- Lots of new UCI options. It is now possible to switch late move reductions and futility pruning on and off, to change the futility pruning margins, and to choose the criteria used for late move reductions (knowledge, history, or both). For full details about these and all the other new options, have a look at the Readme.txt file.
- Somewhat more selectivity in the search. This seems to help a bit in blitz games, but I am less sure about slower games.
- Several evaluation improvements, the most important is perhaps the new "piece in hole" bonus (thanks, Ryan!). This has made Glaurung slower than ever. It is now almost as slow as Gothmog (my previous engine).
- Some new endgame knowledge. Glaurung now understands some fortress draws in KQ vs KRP endgames, and draws in KB and multiple rook pawns against K. I also have some new code for KRPPP vs KRPP with no passed pawns which I intended to include, but it ended up being a bit too complicated and messy. Perhaps I will be able to clean it up and include it in the next version.
- Static null move pruning is now used in the last 3 plies, rather than the last 2 plies. This looks very risky, but in my tests it seems to work well.
- Many little tweaks to search and evaluation parameters.
- Several minor bug fixes.
Tord