Author: Dann Corbit
Date: 15:07:16 09/07/01
Go up one level in this thread
On September 07, 2001 at 17:52:06, Mike S. wrote: >What is the typical, or average ply depth Deep Fritz (6) calculates per move on >your dual machine (MHz?) in a game at 40/2h, or 3 min/move? > >I can only make rough estimations of the speed increase from dual to 8 CPUs >(maybe one of you can make a more precise assumption), but if we can expect it >to be *3 times* as fast as a dual - then I think we can also expect Deep Fritz 7 >to calculate approximately *1 ply deeper* than on a dual. > >Is this ok? > >With that, we would have a better comparison with Deep(er) Blue than by the node >rates. From the D.B. logs of 1997, I have the impression that it's typical >searching depth was 11(6), or 17 plies, except in the endgame where it was more >of course. > >I think if we base the DF7 estimation on duals from 2*800 MHz to 2*1 GHz, it >will be ok because I think the Bahrain machine should have clock rates in this >range. Plus one ply and we'll see roughly, what depth gap to expect to D.B.'s >depth. If there is one :o) I don't think you can use the plies in a comparison. The Deeper Blue search was ultra-fat, in that they did not use null move. Hence, this position (from another post): rnb2bk1/1pqnr2p/p2pp1P1/6PB/4P3/2N1B3/PPP4P/R2Q1RK1 w - - will blast most computers, but Deeper Blue would smash it in a flash. Same depth != same depth, unless the search trees are identical. They are not. Consider if DF gets 1 M NPS per CPU, that would be 8M. Maybe we can safely double it to 16M. We can only search a tree with 1/10 of the nodes that Deeper Blue did in the same time. What a ply means from program to program is (unfortunately) incredibly non-standard. You can make meaningful comparisons of ply depth only for a given program against itself (and even then, only if the search tree retains its shape for both tests!). None of this says anything about strength, and we would have to play the programs against each other a huge number of times to determine ability. But as far as assumptions about equal ply depth -- they don't hold any water. That's too bad (I wished for some uniform measure like that myself) but that's just the way it is.
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.