SIMD (Single Instruction Stream, Multiple Data Stream) Architecture
Main Feature of the architecture:
Large number of small or simple processors.
Each processor has its own memory.
Typically only connected to its neighbors (NEWS), and usually
customized (not off-the-shelf) processors.
A single "control processor" issues each instruction, in turn,
every processor executes the same instruction on its local data at the
same time.
Some processors may be turned off on any instruction. This
allows certain logical operations to be performed.
A schematic view of the architecture is shown below.
Main advantage of these machines is the ease of programming. The
disadvantage is that they are specialized machines and not suited to
all applications.
These architectures support the interesting model of
data-parallelism.
In data parallel computing, the parallelism lies in the data, not
in the control. Furthermore, data parallel computing relies on the
vector model of parallelism. This model assumes multiple data
and serial control.
The following is a diagram of a vector architecture (vector RAM or
VRAM). Notice that the program resides in the scalar memory and that
there is one program and one program counter.
This model is called SIMD, a term that was coined by Flynn in the late
sixties.
