Author: Tom Kerrigan
Date: 18:05:41 09/22/01
Go up one level in this thread
On September 22, 2001 at 18:29:46, Andreas De Troy wrote: >When I see results for the so called "Fritzmarks", I notice that the actual >numbers decrease with increasing hashtables. Is this an artefact of the >measurement? In other words, are larger hashtables always better? I suppose it >depends of the speed of the processor. If you have, for instance an Athlon 1 Ghz >(or a 1.5 Ghz or...), does it -in general- make sense to increase the size of >the hashtables to 256 Mb, 512 Mb or more? > >Thanks in advance for any help! The only disadvantage to having huge hash tables is when the program is designed to clear the hash tables at the beginning of every search. With a huge amount of memory, clearing it can take even a couple of seconds, which could be most of your move time during a blitz game and performance will obviously be severly decreased. I don't know how Fritzmark works, but there could be a few explanations for the odd numbers you're seeing. If people report their own scores, then maybe people with huge amounts of memory tend to get slower & cheaper (CAS3) memory and that's why it's slower. Or if they have Rambus memory, the latency of RDRAMs increases as modules are added, so a system with a ton of Rambus memory will also be slower. A certain Pentium chipset did not cache any memory over 64MB, so upgrading past 64MB often hurt performance badly, but I don't think a newer chipset would have this sort of design inadequacy. The other thing about big hash tables is that changing the hash table size will change the locations of where each hash table entry is written. It's possible that increasing the size would cause the mapping to change such that a critical hash entry for a test position is overwritten before it becomes useful and it takes longer to get a solution. -Tom
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.