Author: Christopher A. Morgan
Date: 13:04:44 01/19/03
Go up one level in this thread
A more accurate formula for setting hash table size is based on the speed of your particular engine in your GUI in your computer and is as follows: Megabytes of Hash = (MegaNodes per second) * (seconds per move) * 2. Round to a power of 2. If you want to be safe, raise to a power of four. Steve Lopez says that with too large a hash table your engine will be slowed down. I'm not sure there is any evidence for this. Anyone know? In some of the older versions of Fritz, I think it was, they showed how fast the hash table was filling up with a color bar. That way you could use the particular time for analysis in typical position and see whether or not your hash was filling up, on average, before analysis was complete. As you know there is a great variance in how fast different engines analyze. The above formula is more specific than the generalized formula Steve gives which does not look at the speed of the engine's analysis, but only at speed of processor. On January 19, 2003 at 13:27:35, Brian Katz wrote: >Can someone please tell me if this formula still applies to faster processors? >As an example a 350MHz and 3.0 gig processors at a game with an average >of 3 minutes (180 seconds p/move) >Also, in using the Athlon performance numbers such as the Athlon 2600+ which is >a 2.13 gig Processor. Which figure do you go by? The 2.13 for the Gig or the >2600+ for the performance? > >The formula is 2 X processor speed X avg. seconds per move / 1000. > >For the 1st example: > >2 x 350=700 >700x180=126000 >126000/1000=126 MB Hash Tables > > >For the 2nd example: > >2x3000=6000 >6000x180=1080000 >1080000/1000= 1080 MB Hash Tables > >Is this formula still a practical guide? If this is so, unless you have well >over a 1 gig of RAM with fast processors, this formula is useless for a normal >avg. 3 min per move tournament game. > >I am presently using an AMD Athlon 2600+ XP (2.13 gig) with 1 gig of RAM. > >It appears that trial and error is the best guide here. Not an exact science >however. > >Any opinions on this would be greatly appreciated. > >Brian Katz
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.