You are on page 1of 18

Computer Architecture

SIMD
Ola Flygt
Växjö University
http://w3.msi.vxu.se/users/ofl/
Ola.Flygt@msi.vxu.se
+46 470 70 86 49
Outline
 Design space
Granularity
Connectivity
Processor complexity
Local Anotomy
 Fine grained architectures
 Course grained architectures

CH01
A typical SIMD system
The development of SIMD systems
Design space
 There are several factors to consider
when designing a SIMD. The most
important ones are:
Granularity
Connectivity
Complexity of processing element
Local autonomy
Granularity
mapping data to PEs
Granularity
Fine to coarse grain
Connectivity
Not all the
types of
connectivity
for data
parallel
arch. are
used in SIMD
arch.
Performance of
interconnection networks
 There are several important
parameters in an interconnection
network
Diameter, the shortest path between the
two most distant nodes
Bandwidth, how many data exchanges
simultaneous in the network
Blocking, how many exchanges are
blocked
Performance of
interconnection networks
N=128*128 N=128*128

1 1
1 16384
256 16384
28 28
14 14
14 16384
1 16384
Processor complexity
 The design space for processor
precision
Single bit – fine grained, mainly for image
processing
Integer – general computing
Floating point – Coarse grain, scientific
computing
 Fine grained -> more PEs
Local
autonomy
Fine grained SIMD
The MPP system
Fine grained SIMD
The MPP processing element
Fine grained SIMD
The MPP integrated circuit
Coarse grained SIMD
The CM5 processing element
Coarse grained SIMD
The CM5 fat tree connectivity
Coarse grained SIMD
The CM5 system parameters

You might also like