15 Oct
2005
15 Oct
'05
6:18 p.m.
William Donzelli declared on Saturday 15 October 2005 02:07 pm:
A couple of years ago, Cray was working on a proposal for a DARPA grant,
one of the ideas they were playing with in the design of the machine
included memory-local processor, sort of like NUMA, except the memory
modules themselves had processors that you could send 'tasks' to, sort
of a vector of processors (as opposed to a vector processor:).
Pat
--
Purdue University ITAP/RCAC --- http://www.rcac.purdue.edu/
The Computer Refuge --- http://computer-refuge.org
Hmm,
interesting. In what ways other than the processor-memory model
could things have evolved in the early years? It'd be intersting to
speculate where computing might be today if it had evolved down some
totally different path very early on...
How about the model with memory and processor being one? The "closest"
production machines like this were probably the CM-1 and CM-2. Maybe
some neural network machines in the labs will finally hit the point
when they get useful. Or to really get wild, perhaps quantum machines
or those that work on probablities?
Processor-memory and analog were the easiest (by far) to implement
years ago, so that is why they were there, and the above weird
technologies were not. 30 years from now, some of those weird
technologies may be quite normal.