Someone floated an idea that sounded interesting ...
Has anyone heard of writing their calc functions in C, and burning them onto field programmable gate arrays? The calc models would have to be fairly static, but it might be able to give you magnitudes of speed improvements.
I confess that I know very little about this technology, but it bears some investigation.
Hopefully J.O'S can weigh in with something on his blog....
Here is an interesting article.
©2006 Marc Adler - All Rights Reserved
Subscribe to:
Post Comments (Atom)
1 comment:
Here is another one - video cards with N 'processors' in them can be used as mini-clusters. They are designed to do pixel shading but can be used to do any normal floating point math.... like this guy
http://www.newegg.com/Product/Product.asp?Item=N82E16814241006
has 48 pixel shader processors :) 48 parallel computations - with no MPI/ PVM, etc. It will be a while before we get that many cores in our CPUs - but we could use these light processors to do our math bidding.
Dmitiry and I had a chat about this the other day. It seems real and doable.
-Chris
Post a Comment