[geeks] Massively parallel... cheap?

[Joshua D Boyd]

On Sun, Mar 24, 2002 at 08:34:58AM -0800, David Passmore wrote:
> The entire set will have to be processed at once.
> 
> Basically what I'm looking for is some input on the most cost-effective
> multi-processor solutions that I can cluster together with the highest
> integer performance. I cannot afford to use a machine that has significant
> vector processing capabilities (although it would help). Are there 4-proc+
> Pentium Pro motherboards out there cheap? 

Neural networks, I believe, can be represented as systems of linear equations
to solve.  Although, I've only been told this, I've never worked out how to
do it, and my book on the topic doesn't specify.  I do have an idea of how it
might be worked out.

In the general case, my understanding is that neural networks don't cluster
well, but sometimes they are easy to partition into sub nets that can be
run on seperate machines.  I don't know if your problem can be easily 
partitioned or not.

On the upside, they are pretty simple to work with, except for the complexity
factor.

-- 
Joshua D. Boyd