Author: Robert Hyatt
Date: 20:07:47 06/19/04
Go up one level in this thread
On June 18, 2004 at 06:22:04, Vincent Diepeveen wrote: >On June 17, 2004 at 21:06:51, Jeremiah Penery wrote: > >chessprograms in general do not profit going from 512KB to 1MB. > >Crafty doesn't either. Actually it _does_ profit. I posted those results quite a while back. I actually ran on the original xeon 400 with 512K, 1M and 2M of L2. Each increase sped the program up using just one CPU. With the parallel stuff it helped even more... > >www.aceshardware.com proves it doesn't matter for diep either. > >>On June 15, 2004 at 06:33:57, David Mitchell wrote: >> >>>Crafty is so cache-stingy, it might not make a big difference, but I'd be amazed >>>if a program like TSCP, wouldn't see a sizeable difference if the L2 cache size >>>was doubled. >> >>TSCP probably fits in any processor's L2 cache, and even the L1 cache of >>Athlons. So increasing L2 cache size won't help it at all. Crafty should enjoy >>at least some benefit from increasing the cache size (up to 1MB, or perhaps >>more). >> >>>Of course, I've been amazed before, and nothing beats a little testing - all >>>theories aside, you see the darndest results! >> >>The reason that increasing cache size doesn't increase the speed of chess >>programs is that almost all the memory accesses in a chess program are hash >>probes, which will cause cache miss and must go to main memory, which is very >>much slower. >>If you make the hash table small enough to fit inside the cache, then maybe you >>will see some benefit in terms of NPS from making cache bigger (because you can >>increase hash table size and still be in cache), but your time to solution will >>go up also with the smaller hash table.
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.