Author: Matt Taylor
Date: 15:44:01 02/11/03
Go up one level in this thread
On February 11, 2003 at 13:49:01, David Hanley wrote: > >>Well, actually he made a pretty good guess. In tests I've found that faster cpus >>actually aren't affected much at all by L2 cache size. > >I'd call that a massively unfounded assumption. It all depends on your memory >access patterns--a fast CPU hitting main RAM non-stop is a waste of electricity. > >dave He gave results for Crafty, and Crafty has obvious access patterns. The assumption is that Crafty's access patterns are representitive of all Chess engines. I do not think this assumption is completely unfounded. How many Chess engines completely peg main memory? Chess engines still have an I/O burst cycle just like every other piece of software. I do think his conclusion was slightly incorrect. As desktop CPUs have become faster, they have included more & more cache memory -- necessary as memory becomes slower and slower to access (relative to cache & registers). He probably observed a cut-off and assumed it was due to clock speed rather than cache size. Above a certain size, the most important code & data in the program will be completely contained in the cache. Large hash tables blow the cache frequently, and increasing the cache (unless it's huge) won't really help such data structures. -Matt
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.