Author: Albert Bertilsson
Date: 23:23:11 12/10/03
Go up one level in this thread
On December 10, 2003 at 17:00:05, Mike Siler wrote: >On December 09, 2003 at 10:29:00, Albert Bertilsson wrote: > >>Hi! >> >>I've finally got around verifying perft(10) from the starting position and it >>is: 69352859712417, that is 69*10^12 nodes. >> >>It took 85 hours for my four computers to calculate it. If some people has a >>couple of spare cpu cycles perft(11) would be no mayor problem to calculate. >> >>Hardware used: >>XP2100+ (256MB hash) >>XP1700+ (128MB hash) (not working on the problem 24/7) >>Celeron 800 (128MB hash) (not contributing very much =)) >>P4 2GHz (256MB hash) (not working on the problem 24/7) >> >>/Regards Albert > >Maybe I'm missing something here, but aren't hash tables really risky here? You >are bound to get lots of collisions and even one collision would mean that your >final count is wrong. > No, hash tables in them selves are not risky. What is risky is using a short hash key that makes false hits probable. Most chess programs uses 64bit zobrist keys because they are quite compact, cheap to calculate and give very few false hits. Ideal would be the complete board stored, preventing any false hits. 64 bit zobrist keys would probably be ok, but just to be safe I use 128 bit zobrist keys when calculating perft. The risk of getting false hits is low with 64bit, with 128 bit keys it's extremly low even for huge problems like perft(11) or perft(12). /Regards Albert >Michael
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.