SIMD Parallelism Quiz

Parallelism in SIMD

• SIMD (Single Instruction, Multiple Data) achieves data parallelism, where a single instruction is executed on multiple data items simultaneously.

ALUs in SIMD

• If there are fewer ALUs (Arithmetic Logic Units) than data items in SIMD, the processing is done in batches, with each batch requiring a separate clock cycle.
• In classic design, ALUs in SIMD are required to be identical, with each ALU performing the same operation on different data items.

Efficiency of SIMD

• SIMD is efficient for problems that involve the same operation on a large dataset, such as scientific simulations, image processing, and data compression.

Alternative to SIMD

• Vector processors are mentioned as an alternative to SIMD, which also perform the same operation on multiple data items, but with a different architecture.

Distributed Memory System Cluster

• The main characteristic of a Distributed Memory System cluster is that each node has its own local memory, and data is exchanged between nodes using interconnection networks.

Interconnection Networks

• The number of devices connected to a bus interconnect affects performance, as it increases the latency and reduces the bandwidth of the network.
• There are two categories of interconnection networks: static networks (e.g., mesh, torus) and dynamic networks (e.g., bus, switch).

Computation Units in a Cluster

• The individual computation units in a cluster of a Distributed Memory System are called nodes.

Alternative Term for Distributed Memory System

• A Distributed Memory System cluster is also referred to as a cluster computer or a massively parallel processor (MPP).