Author: William Penn
Date: 07:36:14 03/01/04
Shredder 8 selectivity, tablebase access, hard drive churning Selectivity is one of the Shredder 8 engine parameters which can be changed. It ranges from -3 to +99. The default is -1. I have experimented with the -3 -2 -1 and 0 settings. I don't notice a big difference between the -3 -2 -1 settings except with tablebase access, however there are other significant differences with a setting of 0. Shredder 8 accesses tablebases much more with selectivity -1 (the default) than with Selectivity -3 (the minimum). In a typical test position in the early endgame (R+N+4P vs R+N+4P): 5r2/3n4/p1k1pp1R/2p3P1/2P5/1P1N1K2/P7/8 w - - 0 1) [D] 5r2/3n4/p1k1pp1R/2p3P1/2P5/1P1N1K2/P7/8 w - - 0 1) there was about 1 tablebase access per 5,000 nodes with selectivity -3, compared to 1 tablebase access per 1,000 nodes with selectivity -1. This rate of tablebase access is already beginning to slow the program considerably due to heavy hard drive access of the tablebase files. To wit, the speeds with those two selectivity settings were about 332 kN/s and 246 kN/s respectively. (If I turned off tablebase base access with selectivity -1 the speed went up to about 420kN/s). This is perhaps still an acceptable speed, but borderline because the hard drive is churning constantly now (paging??) and I fear for its health. Changing hash size or tablebase cache size seems to have no significant effect on the hard drive churning, nor does playing around with different windows pagefile sizes. Nevertheless I let it run for about 8 hours in infinite analysis mode in this position and somehow my poor overworked hard drive survived. I would expect to have to replace the hard drive frequently if I do this very often. Is that the price we must pay to seek the truth in endgame positions? This is with the native Shredder8.eng engine in the Chessbase GUI (of course), all 3-4-5 tablebases, running Windows XP Home here on an AMD Athlon XP 2400+/2.0Ghz, 1G RAM, 512MB hash, 32MB tablebase cache. I see no point to trying to add any 6-man tablebases unless this serious hard drive churning problem can be fixed. Perhaps the tablebase coding is outdated, designed for the old days when we only had 3-4 tablebases and a few 5 tablebases, and needs to be improved somehow(?). WP
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.