Author: Robert Hyatt
Date: 08:43:24 10/18/05
Go up one level in this thread
On October 18, 2005 at 00:02:10, K. Burcham wrote: > > >How does Crafty manage and use its hash? > >If you are playing long time control, >If you set the hash at 432 for Crafty, and lets say this will fill in 4 minutes >with strong hardware. If the program does not move and continues to think---does >the program just write over the top of the existing hash? Yes. But to help with this, the "adaptive hash" algorithm attempts to properly size the hash table based on the average time per move for the time control selected... >Exactly where on each memory stick does the program choose to overwrite? Hashing, by its definition, is a random memory probe. So it is impossible to predict where a particular position will probe in the table, until the probe is done using the updated hash signature. >Is the data collected, that is stored in the ram (hash), rated as "worst move, >best move"? How is the data that is collected in four minutes of thinking stored one entry per position, period. >in the ram? Is this in some kind of an order so as to only write over less >valuable data? > Primarily it is an issue of depth. I'd prefer to save an entry that represents a deeper search over one that represents a shallower search. >If we say that a program on strong hardware can fill 432 megs hash in four >minutes, >If we say that a program is making every move in four minutes, >Can we say that this program writes over its hash every move? Take Crafty. You would be using 384M for hash, with each entry being 16 bytes. so you have 24M positions. On any decent hardware, it won't take very long to start overwriting just 24M positions, when the program can easily search 1-2M nodes every second. On my quad dual-core opteron, I could just about fill that hash table in one second. > >kburcham
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.