Molecular Dynamics (MD) simulations are computationally intensive, requiring efficient force calculation methods to study large-scale physical systems. While the AutoPas library provides an framework for dynamic algorithm selection on CPUs, the increasing power of GPUs presents an opportunity for acceleration. This thesis implements a GPU-suitable particle simulation algorithm using Kokkos and evaluates its performance within the AutoPas ecosystem. The results show that the GPU implementation outperforms CPU variants in high- density scenarios, achieving speedups of up to 7×. Since increasing the cutoff distance increases the number of particles within interaction range, the GPU implementation is also particularly effective for high-accuracy simulations requiring larger cutoffs. Conversely, in low-density systems, the overhead of data transfer and cluster rebuilding causes the GPU implementation to be outperformed by optimized CPU variants. However, the GPU variant scales more effectively as computational intensity increases.     «

Molecular Dynamics (MD) simulations are computationally intensive, requiring efficient force calculation methods to study large-scale physical systems. While the AutoPas library provides an framework for dynamic algorithm selection on CPUs, the increasing power of GPUs presents an opportunity for acceleration. This thesis implements a GPU-suitable particle simulation algorithm using Kokkos and evaluates its performance within the AutoPas ecosystem. The results show that the GPU implementat...     »