Author: Robert Hyatt
Date: 11:20:09 02/06/06
Go up one level in this thread
On February 06, 2006 at 05:52:14, Mike S. wrote: >I recommend to make a compromise (don't bother to continue reading if you do not >accept compromises :-)) > >1. In GAMES*, do NOT use all giant tbs. you can get. I recommend a subset, all >3- and 4-piece tbs. and all KR?-KR (incl. KQR-KR) 5-piece tbs. Rook versus rook >endings are the most important in practice. So, you would get most of the >practical benefit and at the same time, escape from the slowdown effect which >would be caused by the other 5-piece tbs. you don't use then. > >(Pure pawn endings are of course very important too, but easier to calculate in >conjunction with the 4-piece support.) > >*) Analysis of late endgame positions is of course a different topic; I use >online ressources which offer up to 6-piece tbs. (which I do not have installed) >to check such positions. > >In addition to all 70 3- and 4-piece tbs. files (~30 MB) these are the files > >01.06.2001 01:00 14.061.624 kqrkr.nbw.emd >01.06.2001 01:00 28.475.167 kqrkr.nbb.emd >01.06.2001 05:00 11.267.500 krrkr.nbw.emd >01.06.2001 05:00 17.169.632 krrkr.nbb.emd >01.06.2001 05:00 16.970.339 krnkr.nbw.emd >01.06.2001 05:00 5.584.441 krnkr.nbb.emd >01.06.2001 05:00 17.189.887 krbkr.nbw.emd >01.06.2001 05:00 4.338.256 krbkr.nbb.emd >01.06.2001 06:00 68.946.403 krpkr.nbb.emd >01.06.2001 06:00 81.247.218 krpkr.nbw.emd > >(~265 MB) > >2. Use a larger tbs. cache. I recommend 32 MB. Some believe this is not >necessary and the Windows cache handles everything much better. Make your own >choice. > >3. I assume that evals from tbs. hits are stored in the hash tables just like >any evals are. That would mean, large hash tables help to reduce the tbs. >slowdown effect too, again assuming that hash is always probed first no matter >if it is a tbs. position or not. (Of course, take care to not make them too >large, IOW avoid memory swapping.) > >I'm not 100% sure about point (3.); maybe programmers can check if my >assumptions are correct? > >Regards, >M.Scheidl I can only reply about my program. I do, in fact, do exactly that. Flagging such positions so that they do not get overwritten if at all possible as well.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.