A Preliminary Study on Accelerating Simulation Optimization with GPU Implementation

TL;DR

The study investigates accelerating simulation optimization via GPU implementation for three problems including the mean-variance objective f(w) = 0.5 Var(w^T R) - E(w^T R) and a logistic-based binary classification objective c(omega; x) = 1/(1+exp(-x^T omega)). It employs a JAX-based GPU workflow to implement Frank-Wolfe on the first two tasks and a stochastic quasi-Newton method on the third, benchmarking against CPU across expanding problem sizes. Results show GPU implementations achieve about 3 to 6x speedups while preserving convergence and accuracy, with larger gains for high dimensional problems, indicating practical impact for simulation optimization workflows. Limitations include reliance on third-party GPU libraries and a focus on gradient-based methods.

Abstract

We provide a preliminary study on utilizing GPU (Graphics Processing Unit) to accelerate computation for three simulation optimization tasks with either first-order or second-order algorithms. Compared to the implementation using only CPU (Central Processing Unit), the GPU implementation benefits from computational advantages of parallel processing for large-scale matrices and vectors operations. Numerical experiments demonstrate computational advantages of utilizing GPU implementation in simulation optimization problems, and show that such advantage comparatively further increase as the problem scale increases.

Figures (2)

• Figure 1: GPU Architecture
• Figure 2: Computation Time for Three tasks with Different Size