Author: Sune Fischer
Date: 16:15:33 09/24/03
Go up one level in this thread
On September 24, 2003 at 18:52:10, Uri Blass wrote: >On September 24, 2003 at 18:00:09, Sune Fischer wrote: > >>On September 24, 2003 at 17:37:06, Uri Blass wrote: >> >>>>Of course it matters most when doing long analysis, not very important for blitz >>>>games. >>> >>>How much it matters? >> >>Every tiny bit matters. >> >>>I read claims in the past that doubling the hash only gives 6-7% speed >>>improvement and if you lose 1-2% from division then it may be important. >> >>Yes it is in that ball park probably. >> >>I don't think it's constant though, those 6-7% could be less in very fast bullet >>games and more in long over-night analysis sessions. > >I am talking about long analysis. >The question is if bigger hash tables mean always better branching factor or >only in some plies when the behaviour of better branching factor stops after >them. The larger table will always contain more history, whether this gives a linear or exponential speedup I don't know. >>>The only way to know is testing and from my experience 1-2% speed reduction >>>because of division is possible. >> >>It's crazy that they take so long. > >Other people may say that it is crazy to want to use always more hash tables. Is there any reason not to be greedy here, what else did you buy all that memory for? Crazy is not taking what you are offered, if you want my opinion. The optimal solution would perhaps be to choose at runtime between % and & based on how much memory is set for the game. If it's close enough to the power two we take the 1% speedup, course that presents other problems. -S. >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.