[Beowulf] Quasi-Non-Von-Neumann hardware in a Beowulf cluster.

David Mathog mathog at mendel.bio.caltech.edu
Thu Mar 10 14:13:18 EST 2005


"Omri Schwarz" <ocschwar at MIT.EDU> wrote:

> Would specialty  (albeit commodity) coprocessors hanging off a
> PCI slot be suitable for your applications?

Well, a couple of points:

1.  There may very well be no PCI slot available for such a 
coprocessor.  Forget about it for a 1U, and for a 2U the
available slots may already be used up by a small graphics card,
myrinet, hardware monitor etc.  An open slot may not be sufficient,
there are almost certainly going to be space and cooling problems.
(I can't count the number of times an empty slot had to be
left between cards on PCs I've worked on.)

2.  Assuming that you can get the card in there and not fry anything
you're going to have to at the very least rewrite the code to make
use of the specialized hardware.  Unless this board comes with a
very, very clever compiler to automagically detect the bits it can
do best you're looking at some serious time and/or money spent on
programming.

3.  Where's the market to pay for this?  Perhaps some sort of
specialized rendering engine might be able to sell enough units to
the folks who make CGI movies.  Similarly, a specialized FFT
engine might find a home in many places.  The physics engine that
started this thread might be of some use to, well, physicists. 
Or maybe not, I'm going to guess that the "physics" it
implements may not work so well when scaled up to multi-lightyear
distances or down to the point where quantum mechanics is important.

Regards,

David Mathog
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list