This repository contains the code for experiments comparing simulated annealing algorithm runtimes with parallelization algorithms. We compared multi-start simulated annealing and coupled simulated annealing on optimization of the Schwefel function (10, 100 dimensional). We experimented with runtime optimization with function inlining, OpenMP, and CUDA kernalization. Please find a more comprehensive overview of our experiments in the Technical Report or at this Presentation.
We assume a SLURM environment for CUDA experiments.
| MSA Single Thread | MSA | CSA Single Thread | CSA | |
|---|---|---|---|---|
| CPU | 3493635 | 876549 | 3884980 | 995846 |
| GPU + CPU | 1358433 | 344000 | 1563125 | 440082 |
| GPU (Warping) + CPU | 1208433 | 321389 | 1413214 | 400981 |