Author: Eugene Nalimov
Date: 16:43:05 01/25/99
Go up one level in this thread
On January 25, 1999 at 18:36:45, Peter Fendrich wrote: >On January 25, 1999 at 17:45:31, Eugene Nalimov wrote: > >>On January 25, 1999 at 17:11:48, Peter Fendrich wrote: >> >>>In another post Bob mentioned "Eugene's new endgamedatabase code". >>>What's thatand are there any improvements compared to the code from Steven J. >>>Edwards? >>>//Peter >> >>You can download source of the generator and Win32 executables >>from ftp.cis.uab.edu/pub/hyatt/TB. >> >>Main advantages of my generator over Edwards' are: >>- better indexing schema, so my TBs are approx. 2 times smaller; >> also, better data locality speedups lookups, >>- probing code hash chunks of TBs in memory, so you can trade >> RAM for probing speed, >>- it supports SMP, >>- en-passant captures are correctly handled, >>- I generated full set of 5-man (3+2) TBs; you can find it at >> Bobs' site, too. Total size is 22.5Gb. >> >>If I remember it correctly, in the positions with a lot of TB >>probes, Crafty with my probing code is ~6 times faster than the >>old one (3 times faster because of hashing, plus additional 2x >>speedup because of SMP). >> >>Next improvement is pending - on-the-fly TB decompression. >>TBs must become 3-5 times smaller with only a small impact on >>probing speed. The code is ready and waits for Bob to test it >>and incorporate it into Crafty. After that decompression code >>as well as compression utility will be available, too. >> >>Eugene > >Thank you, it looks great! >I haven't used any *real* table bases yet and plan to incorporate that now in my >program. >I saw that they require a lot of RAM to be generated. But did I understand it >right that using them doesn't? > >//Peter You can live with ~300Kb of RAM for probing code, but I'd recommend to use several megabytes for hashing. It *really* speeds things up. Also, to probe compressed TBs, you'll have to use more RAM - indices and decompression tables for *all* TBs are read into memory at program startup time. If you'll use my probing code (and there is no way to use my TBs withoutusing it - indexing schema is very complicated), I'd like to have 2 copies of your chess program in my collection. Eugene
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.