Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GPU Based Differential Evolution: New Insights and Comparative Study

Dylan Janssen, Wayne Pullan, Alan Wee-Chung Liew

TL;DR

This paper analyzes how Differential Evolution can be efficiently mapped to GPU architectures, detailing architectural choices, parallel models, and kernel designs that shape performance. It surveys prior GPU-DE work, introduces a GPU-based numerical optimisation benchmark inspired by CEC competitions, and presents two case studies comparing All-GPU DE variants and open-source implementations. The findings demonstrate substantial wall-clock speedups from All-GPU and kernel-consolidation strategies, while also highlighting convergence challenges on complex benchmarks and the need for standardized benchmarking. The work has practical impact by providing a baseline GPU-DE benchmark and concrete guidance for developing faster, more scalable GPU-based DE algorithms, with future directions including history-based adaptive DE and multi-population GPU models.

Abstract

Differential Evolution (DE) is a highly successful population based global optimisation algorithm, commonly used for solving numerical optimisation problems. However, as the complexity of the objective function increases, the wall-clock run-time of the algorithm suffers as many fitness function evaluations must take place to effectively explore the search space. Due to the inherently parallel nature of the DE algorithm, graphics processing units (GPU) have been used to effectively accelerate both the fitness evaluation and DE algorithm. This work reviews the main architectural choices made in the literature for GPU based DE algorithms and introduces a new GPU based numerical optimisation benchmark to evaluate and compare GPU based DE algorithms.

GPU Based Differential Evolution: New Insights and Comparative Study

TL;DR

This paper analyzes how Differential Evolution can be efficiently mapped to GPU architectures, detailing architectural choices, parallel models, and kernel designs that shape performance. It surveys prior GPU-DE work, introduces a GPU-based numerical optimisation benchmark inspired by CEC competitions, and presents two case studies comparing All-GPU DE variants and open-source implementations. The findings demonstrate substantial wall-clock speedups from All-GPU and kernel-consolidation strategies, while also highlighting convergence challenges on complex benchmarks and the need for standardized benchmarking. The work has practical impact by providing a baseline GPU-DE benchmark and concrete guidance for developing faster, more scalable GPU-based DE algorithms, with future directions including history-based adaptive DE and multi-population GPU models.

Abstract

Differential Evolution (DE) is a highly successful population based global optimisation algorithm, commonly used for solving numerical optimisation problems. However, as the complexity of the objective function increases, the wall-clock run-time of the algorithm suffers as many fitness function evaluations must take place to effectively explore the search space. Due to the inherently parallel nature of the DE algorithm, graphics processing units (GPU) have been used to effectively accelerate both the fitness evaluation and DE algorithm. This work reviews the main architectural choices made in the literature for GPU based DE algorithms and introduces a new GPU based numerical optimisation benchmark to evaluate and compare GPU based DE algorithms.
Paper Structure (45 sections, 8 equations, 7 figures, 4 tables)

This paper contains 45 sections, 8 equations, 7 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Differential Evolution Algorithms
  3. Parallel models for Evolutionary Algorithms
  4. Master Slave model
  5. Island Model
  6. Cellular Model
  7. Nvidia GPU Hardware and Software
  8. Hardware Architecture
  9. Thread Hierarchy
  10. Memory Hierarchy
  11. Global memory
  12. Local memory
  13. Texture memory
  14. Shared memory
  15. Constant memory
  16. ...and 30 more sections

Figures (7)

  • Figure 1: Parallel models for Evolutionary Algorithms
  • Figure 2: Comparison of the main differences in hardware architecture between CPUs (left) and GPUs (right).
  • Figure 3: Memory hierarchy in Nvidia GPUs.
  • Figure 4: DE using All-GPU parallel model where each DE operation is implemented as a separate kernel.
  • Figure 5: Improved DE using All-GPU parallel model that combined several logically-related kernels to increase shared memory use, reduce global memory access, and utilised CUDA streams to enable concurrent kernel execution.
  • ...and 2 more figures