Author: Uri Blass
Date: 19:50:22 12/19/03
Go up one level in this thread
On December 19, 2003 at 21:59:07, Ricardo Gibert wrote: >Neither. The number of positions a program examines in the course of a game is >finite. If the TT is large enough to accomodate all of them, then further >increasing the size the TT will not be productive. This is not just a case of >diminishing returns. It's more like a brick wall. I asked about a case when the TT is not large enough to have all of them. I am interested in the question if I can get significant bigger improvement with more time. Of course it is clear that for very short time control there is no improvement and the question is what happens later and if the % of the improvement get a wall or not. If we use big hash then later may happen only after some minutes and this is the reason that I suggested 1 mbyte and 2 mbytes to make testing easier. If we have t(1) time of solutions with 1 mbytes and t(2) the expected time of solution with 2 mbytes then the question is what is the behaviour of t(2)/t(1) as a function of t(1) I understood that for gothmog t(2)/t(1) may behave similiar to a^t(1) when a>1 from Tord's post in the winboard forum see http://www.f11.parsimony.net/forum16635/messages/59181.htm He was talking about t(48)/t(24) but the idea is the same and I think that it is easier to check the behaviour of t(2x)/t(x) for small x. Uri
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.