Author: Kris Jordan
Date: 12:15:12 03/09/02
Go up one level in this thread
The bottom line is Steve Lopez is wrong. He is the first person I have ever heard say that large hash tables hurt performance. KJ On March 09, 2002 at 11:24:14, karen Dall Lynn wrote: >Greetings [[all]] > >Steve Lopez has recently released a T-note advancing a formula to deploy >optimized hash table sizes. As known, the formula is a function of processor >size and average thinking time per move. What is characteristic of this formula >is that it settles much smaller hashtable sizes than I have been using or than >the most commonly referred sizes in ordinary games. > >But if hashtables should be set to something between 16-64Mb for ordinary blitz >games on processors 1500(+-)500 Ghz, **why the call of the optimize-strength >funcion in Fritz 7 makes the hashtable size grow to the top size under available >RAM?*** In my case, it increases to 608Mb from my 768Mb, leaving sometimes no >more than 7-30Mb to the rest? Again, if Steve Lopez' formula is the right way to >do it, why Fritz 7 optimizes strength entering bulk sizes to fat hashtables? Why >the program does not apply some algorithm akin to the thinner proposed formula? > >I have also tested my F7 Fritzmark (P4 1500Ghz/768Rambus)against distinct >hashtable sizes, getting the following results: > >16Mb 763kN/s (+-)7 >32Mb 785kN/s (+-)14 >64Mb 801kN/s (+-)0 >128Mb 778kN/s (+-)0 >256Mb 763kN/s (+-)5 >512Mb 723kN/s (+-)18 > >It suggests that neither top nor bottom sizes will mirror optimal hardware speed >- so, if hardware speed is a hallmark of efficience, maybe optimizing >automatically strength will boil down to a slower and therefore less efficient >engine ability. > > >Karen
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.