Author: Russell Reagan
Date: 23:48:22 12/31/03
Go up one level in this thread
On January 01, 2004 at 02:19:28, Ed Trice wrote: >Well, not exactly. When I was being asked about the Gothic Chess program's >speed, it was not completed. This did not satisfy the person who still wanted >some metrics, so one had to be devised. > >Since that time, I have completed the program, and just recently incorporated >some of Dr. Hyatt's insights in the Crafty program. > >On a 2.8 GHz Pentium IV, Gothic Vortex will average about 445,000 nodes per >second from the stating position. This includes conducting a full search. > >Please understand the original request for data was made prior to even having a >primitive alpha-beta search hooked up. > >As it was worded, the request was "how fast is your move generator", which I >tried to answer. Okay, so now that you have things working, I would suggest implementing perft: http://homepages.caverock.net.nz/~peter/perft.htm First let's define two kinds of perft. "Honest perft" and "fast perft". "Honest perft" is when you just count nodes and don't do anything out of the ordinary. You just take your move generator, make move, and undo move routines and test them as is. You can make them as fast as you want, but no "tricks" to artificially speed up perft calculation. "Fast perft" is when someone calculates perft, but they use some tricks to speedup the calculation, such as using hash tables to store previously computed results (just like how the transposition table works). "Fast perft" has a purpose, but not for this discussion. It would be useful if someone is calculating perft to a high depth, and they wanted a fast perft calculator. "Fast perft" doesn't really compare the speed of the movegen/make/undo routines, and it doesn't help you to debug your search. You could probably write an array based perft calculator for gothic chess without much work, and then you could compare the perft results for various positions and verify that they match up. If they give the same results, then you probably don't have any glaring bugs in movegen/make/undo. If your program is still blowing away programs like Crafty in calculating perft, and you implemented "honest perft", then I'm sure there are a lot of people who would be interested in seeing how you are accomplishing that.
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.