Author: Robert Hyatt
Date: 09:08:23 04/28/03
Go up one level in this thread
On April 27, 2003 at 16:35:20, Jorge Pichard wrote: >On April 27, 2003 at 11:01:18, Robert Hyatt wrote: > >>On April 27, 2003 at 05:34:04, Jorge Pichard wrote: >> >>>On April 26, 2003 at 22:49:57, enrico carrisco wrote: >>> >>>>On April 26, 2003 at 14:16:14, Jorge Pichard wrote: >>>> >>>>>On April 26, 2003 at 12:27:28, Anson T J wrote: >>>>> >>>>>>I'm going to run a few games of blitz @ the time control 3mins + 2sec (inc) >>>>>> >>>>>>On a dual AMD XP2400+ is 128 MB hash sufficient for Deep Fritz 7? >>>>> >>>>>For Blitz of less than 30 Minutes 64 MB hash would be more efficient. >>>> >>>>Not necessarily true for these faster systems. Has someone tested performance >>>>regarding hash size with 2GHz+ machines? >>>> >>>>In the case of HIARCS, more hash does not inhibit performance. >>>> >>>>-elc. >>> >>>I have three computers a Celeron 450 Mhz, an Athlon 1200 GHz and a XP2800+ with >>>512 MB, but for Blitz of less than 30 Minutes, 64 MB hash run faster than using >>>128 MB. >>> >>>Jorge >> >> >>What metric are you using to measure speed? >> >>NPS or time to depth? >> >>NPS is not what you want to maximize. > > >I measure depth within certain seconds, for instance in 10 seconds what depth >does it reach with 64 MB versus using 128 MB. > >Jorge Then you are mis-interpreting the results. Bigger hash does two things: 1. Time to depth decreases in the general case due to catching more transpositions and having better move ordering since less gets overwritten. 2. However, bigger hash table _also_ makes the search more accurate. To see this, run fine 70 with a tiny hash and with a big hash. Big hash will let the program find the solution _much_ earlier (in terms of search depth required). Ignoring #2 is a bad idea. I go for NPS and so long as increasing the hash does not significantly lower the NPS, I go for the biggest I can get. Too big and the NPS drops as paging starts and that's a bad trade-off of course..
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.