Author: Brian Richardson
Date: 17:56:28 06/22/01
Go up one level in this thread
On June 22, 2001 at 20:37:52, Dann Corbit wrote: >On June 22, 2001 at 12:40:55, Brian Richardson wrote: > >>On June 21, 2001 at 17:47:20, Rich Van Gaasbeck wrote: >> >>>On June 21, 2001 at 05:54:05, David Blackman wrote: >>> >>>[snip] >>> >>>>and most chess programs don't use floating >>>>point. >>>> >>> >>>But it does have population count and find first zero instructions which will >>>make bitboard programs like Crafty happy. >> >>I don't think it IA-64 includes find first non-zero _bit_ instruction--only >>bytes. > >Even at that, the offset to the byte gives a single 256 entry lookup to get the >exact answer. It will be a single CPU instruction, a table lookup and an >addition. Faster than any way to solve it right now, I think. > >The big monster benefit will come from all the native 64 bit operations on >bitboards. > >Look at how Crafty peels rubber on a single CPU 500 MHz 21264. It's faster than >my 950 MHz Athlon. When 64 bits hits the mainstream, the 0x88 programs will >suddenly cower in fear. The parallelism of the IA-64 architecture over time will help quite a bit I think, but not so much in the initial Itanium implementation. Wait until Intel publishes the SPEC results (which include Crafty, not tuned of course). During 2H01/1H02 I think then current IA-32 processors will do better, until about mid 2002 (McKinley) when IA-64 will start to outrun IA-32, a least for this specific type of workload.
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.