[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-smp
Subject:    Re: Beowulf vs Dual
From:       "Robert M. Hyatt" <hyatt () cis ! uab ! edu>
Date:       2001-04-20 16:51:21
[Download RAW message or body]


don't forget the T932...  it went to 32, although it was (is) a
very expensive solution...


Robert Hyatt                    Computer and Information Sciences
hyatt@cis.uab.edu               University of Alabama at Birmingham
(205) 934-2213                  115A Campbell Hall, UAB Station 
(205) 934-5473 FAX              Birmingham, AL 35294-1170

On Fri, 20 Apr 2001, Matti Aarnio wrote:

> On Fri, Apr 20, 2001 at 07:41:11AM -0500, Robert M. Hyatt wrote:
> > On Fri, 20 Apr 2001, Rik van Riel wrote:
> > > On Fri, 20 Apr 2001, Matti Aarnio wrote:
> > > > 	It certainly is possible to build machine which has extremely
> > > > 	fast memory access crossbar in between scores of processors,
> > > > 	and their memories, but such is hideously expensive.
> > > 
> > > Not to mention that it would violate several laws of
> > > nature, such as the speed of light (or it would make
> > > so much heat in so little space the silicon would
> > > melt away).
> > 
> > Cray Research has been violating such laws for years.  Of course,
> > the economics is something else entirely.  :)
> 
> 	Nope, Cray does not violate laws of physics, they just
> 	do things very close to the extreme edge, and generate
> 	terrible amounts of heat in small volume (and remove it
> 	successfully).   And their beasts are hideously expensive..
> 
> 	Nevertheless, where some Cray X-MP4** (4-way SMP CRAY)
> 	did have memory crossbar of 6 ns per access, it didn't
> 	scale beyond about 16 processors.
> 
> 	Modern multiprocessor machines (100+ CPUs) are normally
> 	network-interconnected (price of the network depends on
> 	the vendor -- Cray has very high performance, and price)
> 	two to four CPU nodes, where CPUs are some commodityish
> 	microprocessors.    Package may be exotic, but the CPU
> 	chip itself is some generally available one.
> 
> 
> 	Like I said initially, doing SETI-like work is nearly
> 	infinitely scalable, while solving large groups of partial
> 	differential equations (like atmospheric models for weather
> 	prediction) needs lots of, and frequent communication all
> 	over the place.
> 
> 
> 	These days a dual-CPU board (for pentiums) cost very little
> 	more than similarly equipped board for single processor.
> 	The boards are coming with one or two Fast Ethernet ports too.
> 	For that matter...   Making beowulf systems with passive
> 	motherboard cases and boards with two CPUs might create
> 	interesting packageing density..  Perhaps more than 8 CPUs
> 	in 4U height (2 CPUs in 1U is normal thing currently.)
> 
> 
> > > cheers,
> > > Rik
> 
> /Matti Aarnio
> 

-
To unsubscribe from this list: send the line "unsubscribe linux-smp" in
the body of a message to majordomo@vger.kernel.org

[prev in list] [next in list] [prev in thread] [next in thread]