Author: Robert Hyatt
Date: 21:39:50 06/19/02
Go up one level in this thread
On June 19, 2002 at 13:33:19, Dann Corbit wrote: >On June 19, 2002 at 10:39:40, Robert Henry Durrett wrote: > >>On June 19, 2002 at 02:51:30, Dann Corbit wrote: >> >>>On June 18, 2002 at 21:38:49, Robert Henry Durrett wrote: >>> >>>>On June 18, 2002 at 21:08:00, Tom Kerrigan wrote: >>>> >>>>>On June 18, 2002 at 20:35:00, Robert Henry Durrett wrote: >>>>> >>>>>>Well, if people only resort to AMP when they absolutely have no choice, then >>>>>>they are unlikely to discover any of AMPs merits. It is only when they are >>>>>>willing to explore the unknown that they will see what is possible. >>>>> >>>>>The thing is, SMP is kind of a superset of AMP. You can do message passing on an >>>>>SMP computer if you want, but it's kind of a waste. (Extra work involved.) >>>> >>>>The big question in my mind is "Yes, but what ELSE can you do with AMP?" In a >>>>single processor, the interaction between parallel paths is fixed. The >>>>programmer can do little or nothing about it. But with separate processors, the >>>>programmer has more options, or more flexibility. [Or, at least, that's my >>>>perception of the situation.] >>> >>>In the words of Willy Wonka: >>>"Strike that, reverse it." >>> >>>AMP is more flexible than a single CPU, but give me SMP over AMP any day. >> >>Well Dann, I will defer to your experience and judgement in this matter . . . >>for now . . . but you may hear me say "I told you so!" sometime in the future. > >AMP has its place. It's a lot easier to coordinate 1000 CPU's that way. But >for a given number of CPU's (say 64) SMP is clearly better. Just so long as N is reasonable. And even then, most of the current SMP platforms are not really SMP per se, they are NUMA, which makes the memory system far cheaper to build. That is why Cray T90's are $60,000,000 machines. a memory system that can be accessed by 32 processors simultaneously is a horrendous switching problem.
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.