Back to glossary

Data Level Parallelism

A form of parallelism where the same operation is performed on multiple data points simultaneously.

AbbreviationDLP
Categoryhardware

Overview

DLP is highly efficient for workloads that involve large arrays or vectors of data, such as image processing, scientific simulations, and AI inference.

Implementation

  • SIMD (Single Instruction, Multiple Data): CPU extensions like AVX-512 or ARM NEON.
  • Vector Processors: Specialized hardware designed for vector math.
  • GPUs: Massive arrays of simple cores designed specifically for DLP.

Benefit

DLP provides much higher throughput and energy efficiency than ILP for suitable workloads because it reduces the overhead of fetching and decoding instructions.