Author: Robert Hyatt
Date: 06:27:24 10/15/99
Go up one level in this thread
On October 15, 1999 at 04:56:02, Francesco Di Tolla wrote: >>Most duals don't do this, and memory becomes a bottleneck. Even with two >>cpus, it is not uncommon to see applications run only 1.5 times faster because >>of the memory bandwidth (or lack thereof). > >If parallel computing was just adding new cpu's then a pc with 4 cpu would cost >like a normal pc+the cost of the extra 3 cpu. Of course the key point in >parallel computing is the bandwidth to the memory, a number most of the people >don't pay enough attention to the first time they look at a multiprocessor >compute. >What makes Cray computer so different respect to other, cheaper, parallel >computer is the huge bandwith. Of course there are problems for which shared >memory is not required, so that a distributet memory parallel machine, with >local buses, si enough. > >By the way Bob, do you know which of the parallel programs around work on >distributed memory machines? and how do they handle a position that shows up in >two different heads? > >regards >Franz cilkchess, p.conners, sun phoenix, zugzwang, and probably others I forgot to name come to mind. All run on distributed machines. The most common thing to do for your second question is a global transposition (hash) table. But then sharing this information consumes a lot of bandwidth and introduces lots of overhead. I'm going to start looking at this later this year, myself...
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.