)First, a link to basic information: http://chessprogramming.wikispaces.com/Syzygy+Bases
It says that the Syzygy bases of Ronald de Man, take the 50 move rule into account - what the Nalimovs lack - and are very compact: Less than a gigabyte for 3/4/5-men.
These tables you can either download or create yourself. Who has a flat rate and a fast internet connection will probably like to use the following source:
http://chess.cygnitec.com/tablebases/syzygy/ (3/4/5-men)
Still no 7-Zip on board? http://7-zip.org/
Although not a single file, but a few only. Just unzip everything into the same directory - as always best to SSD or flash memory - and that is the Syzygy-Tablebase path. If I am not mistaken, it should be 290 files (except for any text files you do not need).
Alternatively, you can also generate them yourself. It took me here overnight on dual core i5 CPU just under 6 hours. Total RAM 4 GB (I do not know what is the minimum for five piece). The TCEC chatter Trym has kindly made available a download that contains the software for the generation that is under the GNU license and a PERL environment:
http://www.solheimsvollen.net/TBGen/SZGen.zip (approx. 27 MB)
PERL is only necessary to use a perl script that can greatly simplify the generation. For operation of the tables is of course not necessary and could be uninstalled after a successful preparation of tables.
If you have PERL (*) installed and unpacked the rest in any directory - ideally in the planned future Syzygy directory - you open a command line window and enter:
run.pl --generate --max 5
Optionally, you can also number the CPU threads to be used, for example --threads 4 (whichever you have).
Warning: 6-men generation requires 16 GB of RAM ... and probably a lot of time.
First experience:
Coincidentally, the Syzyzgy-enabled Stockfish version I found (the download link I unfortunately forgot) is of the same date as the development version I have currently installed, 23.09.2013. Thus, it could almost be a 1:1 comparison with / without Syz. Bases. This I have done with ultra-short time control and using 10 endgame starting positions. Not all symmetrical, as is also queen against two rooks or minor piece against three pawns, all of course with more pawns.
Not surprisingly, all 20 test games were drawn. It was, however, interesting to see: 1 Technically flawless course, 2 in-depth observation no computing differences detected (Tables on flash memory card). The engines typically started with 18 depts, then followed with less material soon 20+ depths.
But that was on a single core, ponder off. That it may come on very large computing depths and / or with many cores to "phenomena", about that I cannot make a statement. But basically I think that first impression is quite confidence-inspiring.
Current sources of Syzygy-enabled Stockfish is available at https://github.com/syzygy1/stockfish
But so far, unfortunately no executables.
Seriously, I am very interested in implementing this in Komodo. Hopefully it will be soon.
