Experimental Analysis of GPU Acceleration for Linear Programming Applied in the Production Planning of a Slaughterhouse

This paper presents an experimental analysis of the use of GPUs in Linear Programming models, using a benchmark based on real instances found in the industry. A Linear Programming model was used for a real waterfall problem with two objective functions. The results were evaluated considering the execution time to obtain the optimal solution using the HiGHS solver, using 4 different algorithms: simplex, interior points, Primal-Dual Hybrid Gradient (PDLP), Primal-Dual Hybrid Gradient accelerated by GPU (cuPDLP). The results show that for large instances, the algorithm with GPU acceleration gets lower times than the other algorithms.