Author: Steven Edwards
Date: 05:56:53 08/12/03
Go up one level in this thread
On August 12, 2003 at 05:04:57, Uri Blass wrote: >On August 12, 2003 at 03:28:10, Steven Edwards wrote: > >>Here are the first two subtree counts for depth ten: >> >>Na3: 2440848135252 >>Nc3: 3096505857746 > >I believe that with hash tables it is possible to be significantly faster than >40 years even for depth 11. > >Even without them I believe that depth 9 takes less than 24 hours with fast >hardware for my perft so depth 11 probably takes less than 3 years. CT has a variety of position representation classes from simple to complex along with a variety of move generation classes and a variety of move execute/retract plus position save/restore routines. The CTEnumeration class can work with or without transposition tables/files. So there is a whole family of movepath enumeration configurations that can be tried. Their purpose is to help verify the lower level routines and so these configurations try to exercise as much of the underlying code as possible. It's really not intended to be any kind of a performance test. Anyway, the current version of CT's enumeration, configured to use the most extensive classes for moves, positions, environments, and everything else runs at about 300 KHz (from the CTSIHertz class) on my 800 MHz notebook without transposition assisitance. The CTMachine class tells me that's about 2700 CPU cycles per node. Simpler representations and transposition support, as expected, greatly speed up the rate. Note that the foundation only has to be fast enough to meet the needs of the workbench. My anticipated typical workbench coded search need only process nodes in the sub 100 Hz rate (yes, less than a hundred nodes per second).
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.