Thanks to the appearance of the General-Purpose computing on Graphics Processing Units (GPGPU), researchers have benefited from the spectacular High Performance Computing (HPC) provided by GPUs. Different research fields, such as combinatorial optimization, have taken advantages from the GPUs HPC. In this context, our paper introduces some different Particle Swarm Optimization (PSO) implementations for solving Maximal-Constraint Satisfaction Problems (Max CSPs) using GPU, based on different parallelism levels. These implementations are then compared. The experimental results, presented at the end, show the effectiveness and the efficiency of using GPU to optimize Max-CSPs by PSO.
