Comparison of different parallel implementations of the 2+1-dimensional KPZ model and the 3-dimensional KMC model

Abstract

We show that efficient simulations of the Kardar-Parisi-Zhang interface growth in 2 + 1 dimensions and of the 3-dimensional Kinetic Monte Carlo of thermally activated diffusion can be realized both on GPUs and modern CPUs. In this article we present results of different implementations on GPUs using CUDA and OpenCL and also on CPUs using OpenCL and MPI. We investigate the runtime and scaling behavior on different architectures to find optimal solutions for solving current simulation problems in the field of statistical physics and materials science.

Aging of the (2+1)-dimensional Kardar-Parisi-Zhang model

“…For this work we added the capability to perform simulations with arbitrary probabilities p and q. Benchmarks, comparing our GPU implementation on a Tesla C2070 to the optimized sequential CPU implementation running on an Intel Xeon X5650 at 2.67 GHz, have shown a speedup factor of about 230 for the raw simulation. The basic version from [47], which contains less computational effort per update, reaches a raw simulation speedup of about 100, in the same setup.…”

“…A more detailed description of our CUDA implementation can be found in [46,47]. For this work we added the capability to perform simulations with arbitrary probabilities p and q. Benchmarks, comparing our GPU implementation on a Tesla C2070 to the optimized sequential CPU implementation running on an Intel Xeon X5650 at 2.67 GHz, have shown a speedup factor of about 230 for the raw simulation.…”

Universality of (2+1)-dimensional restricted solid-on-solid models

“…A double-tiling scheme was applied by splitting up the simulation cells into tiles, split further into two subtiles along each spatial direction [34]. In the present two-dimensional problem this yields 2 d = 4 sets of subtiles, each of which can be updated by multiple independent workers.…”

MaterialVis: Material visualization tool using direct volume and surface rendering techniques

“…We tested the tool with various datasets. In the sponge dataset [2], which was already mentioned in the introduction section, we tackled the volumetric imaging of a highly complicated structure. In the dataset we used, the stoichiometry of SiO x was fixed to x = 1, i.e., SiO by setting the silicon excess to 30 vol.%.…”

“…For instance, the sponge dataset [2] is a material produced from silicate, which has interesting nano-technological properties. Very recently, it has been experimentally shown that a silicon-rich oxide film can decay into a silicon nanowire network embedded in SiO 2 by spinodal decomposition during rapid thermal treatment [3], which has also been confirmed by accompanying kinetic Monte Carlo simulations [4].…”