Author: Robert Hyatt
Date: 07:05:19 08/09/99
Go up one level in this thread
On August 09, 1999 at 02:41:45, david burrell wrote: >There is a huge variation between the NPS reported by a variety of >chess software. I realize that there are many factors that might >effect NPS like move-ordering, complexity of static eval, hashing etc. >but perhaps not all chess programs count the same thing. > >So what do you define NPS as? I would count calls to make-move. This is >what my program does. I have heard that some people count total nodes >generated by movgen and others count only leaf nodes. most of us increment a counter (nodes_searched++;) right at the top of Search() and at the top of Quiesce(). Since we normally do a MakeMove() followed by a call to Search() or Quiesce() this gives a sane number. If you count calls to MakeMove() this will be the same most of the time, unless you do something cute like call MakeMove() and then notice if you are in check or not before calling Search(). This screens out illegal moves, but also misses the call to Search() which increments the node count. There is no "rule" to follow here. I use NPS to measure "hard work" and for me, MakeMove() is not nearly so hard as a recursive call to Search() becauase of hash probes, repetition checks, and the null-move search that is always done... > >I am concerned that my program is very slow, about 15K NPS in opening >positions on a P200, (crafy reports about 50K NPS). How does crafty >count NPS? as above... > >If the LG2000 winboard engine has peaked at over 1M NPS on a K6-2/450 >then surely LG2000 is not counting calls to make-move 1M is very high!! > >Any comments would be greatly appreciated. Thanks. > >David.
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.