Page 1 of 1

Suggestions please for working with the full sub-7-man EGTs

Posted: Sat Jan 23, 2010 11:08 am
by guyhaw
A colleague is experiencing difficulties with FRITZ taking a very long time, and oftern 'timing out', when contemplating the fact that it has all the sub-7-man Nalimov EGTs to work with.

I have suggested only making available those EGTs that are likely to be relevant but the real question is:

What are the choices of GUI, engine etc - and settings that will make working with the full set of EGTs feasible and quasi-optimal?

Thanks in advance for your suggestions.

g

Re: Suggestions please for working with the full sub-7-man EGTs

Posted: Sun Jan 24, 2010 3:02 pm
by mbourzut
The default behavior for the engines I know is to pre-initialize all the Nalimov endings in the EGTB paths provided. This is rather costly, since for each ending it will read the header of the compressed file and initialize an index file with 4 bytes for each block of 8k positions. This is well over 1G if all the 6-man endings are initialized.

In my own utility I only initialize endings on first use, which is much more efficient if only a small subset of the endings is going to probed, which is usually the case. Maybe there are engines which do that, but I'm not aware of any. Perhaps in very fast bullet games this can be an issue if during the game a sizeable fraction of time is spent initializing a tablebase when it is first encountered, although I would personally not be terribly worried about that.

Other than that, your solution of putting the most important endings on a separate path and only providing that path for "normal" use makes the most sense.

Ok - thanks, mb

Posted: Sat Jan 30, 2010 11:14 pm
by guyhaw
g