Author: Vincent Diepeveen
Date: 07:57:11 04/09/02
Go up one level in this thread
On April 09, 2002 at 10:03:49, Robert Hyatt wrote: >On April 09, 2002 at 00:57:33, Tom Kerrigan wrote: > >>On April 08, 2002 at 13:16:01, Robert Hyatt wrote: >> >>>Real programs don't do 500-1000 instructions per node... >> >>Heh, that number is dropping pretty quick, huh? >> >>-Tom > > >Nope... that was Vincent's number, if you would simply read things >with a bit of care... My number hasn't changed one iota... Still well >over 3000 at least... > >Not quite so easy to calculate for Crafty since the number is different for >32 bit and 64 bit architectures. But for Intel, it seems to be in the 6000+ >range... I don't find it that interesting actually, because many nodes which Schach saw were nothing more than a compare with a register and a return after that. Same is true for when diep's king is in check. i don't evaluate these nodes at all of course. Idem for transpositiontable hits. I don't evaluate these nodes at all, i get a value from the transpositiontable there. In short number of clocks a node is always so much depending upon how many 'cheap' nodes you are doing, because these take the average number down a lot. way more interesting to see is what actually gets seen in the evaluation at a leaf position! All these 500 clocks a node programs share that they are nothing more than beancounters based upon a piece square table. Other methods of getting the number of nodes a second higher are things like lazy evaluation. They produce loads of extra nodes which are simply inaccurate. Like Bob i don't see these as real nodes at all.
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.