[rescue] Re: [geeks] THIS. MAKES. ME. SICK.

[David Cantrell]

On Wed, Jun 13, 2001 at 03:31:58PM -0500, Bill Bradford wrote:
> On Wed, Jun 13, 2001 at 03:27:45PM -0400, David Michaud wrote:
> > What about the Xilinx FPGA reconfigurable supercomputers? Ability to do
> > a single function _really_ fast, and reconfigure itself to do something
> > else... _really_ fast. Of course, this is scalar, not vector. Not sure
> > about the feasibility, but it's certainly an idea.
> 
> Haven't companies been announcing vaporware "infinitely faster" machines
> based on this for the past 2-3 years now?  I dont think anyone
>  has delivered PRODUCT yet..

It's not strictly true that FPGAs can reconfigure *really fast* either.
Sure, they can re-configure quickly (well, quickly when you bear in mind
how complex the job is), but to me, really fast means that it takes maybe
an order of magnitude or two more time - at the most - than a context
switch on a traditional processor.

Also, I believe (but could be wrong) that most FPGAs can only be
re-programmed a certain number of times (of the order of tens or hundreds
of thousands, not millions) before they begin to fail.  Not good!

They are being used in some applications - a friend of mine is working
with one embedded in a mobile phone, and yes, he does re-config it on
the fly to switch between two different systems.  But that is a very
specialised market, and most definitely not in the same league as
general-purpose computing.  It doesn't particularly matter if it takes
a second or two to reconfigure in his case, as it happens so rarely.

-- 
David Cantrell | david at cantrell.org.uk | http://www.cantrell.org.uk/david/

      Good advice is always certain to be ignored,
      but that's no reason not to give it            -- Agatha Christie