Technical Papers and Presentations

Besides the accuracy of the results, the required computation time is an important factor for simulations in general. Thus, the question arises, that is also discussed in many blog posts, which computer properties influence the computation time. Here, we present findings from a systematic benchmark test for different models which can be found in the Application Libraries in COMSOL Multiphysics®. The models are chosen such that they cover a wide range of ambitious simulation tasks. First, the “electronbeamdivergence” model is used in our benchmark test, where the AC/DC Module is coupled with the Particle Tracing Module to model the propagation of charged particles. The next model used for the benchmark tests is the “displacementventilation” model, which combines the CFD Module and the Heat Transfer Module via the Nonisothermal Flow multiphysics interface. The third model under study is located in the Application Libraries within the Fuel Cell & Electrolyzer Module. In the model “htpem”, the transport of reactants and water in a high temperature PEMFC is investigated. In addition to the Fuel Cell & Electrolyzer Module, the CFD Module is used to model the gas flow for both, the anode and the cathode side of the PEMFC. The last model, that is used in our benchmark test, is the “seismicwavesearth” module in the Acoustics Module, where the propagation of seismic waves through Earths’ internal structure is modelled. In our benchmarking, the performance of different physical and virtual machines is studied. On the one hand, three physical setups are used; an AMD Threadripper (CPU: Threadripper 5965 - 24 cores @ 3.8 GHz; RAM: 256 GB), a typical gaming PC (CPU: Intel Core i7-8086K - 6 cores @ 5.0 GHz; RAM: 32 GB) and a desktop workstation (CPU: Intel Core i9-13900 - 16 cores @ 5.6 GHz; RAM: 64 GB). On the other hand, we have access to different types of computer clusters for virtual machines. The faculty Applied Mathematics, Physics and Humanities of the university owns a HPC computer cluster where multiple virtual machine configurations are used for the benchmarking (CPU: Intel Xeon Gold 6154 @ 3.00 GHz – 12/16/32/64 cores; RAM 64/128 GB). In addition, three HPC clusters of the Erlangen National High Performance Computing Center (NHR@FAU) are included in our study; Woody (CPU: Intel Xeon Gold 6326 -32 cores @ 2.90 GHz; RAM: 256 GB), fritz (CPU: Intel Xeon Platinum 8360Y -72 cores @ 2.40 GHz; RAM: 256 GB) and meggie (CPU: Intel Xeon E5-2630 v4 – 20 cores @ 2.20 GHz; RAM: 64 GB). All studies are performed with COMSOL Multiphysics® version 6.1.0.346.

From the findings of our study, we expect that we can derive recommendations for optimal hardware properties for a given set of COMSOL Application Library models. In addition, the results can be used to optimize virtual machines for specific tasks. This benchmark can help to choose the most suitable, cost efficient hardware for specific problems while decreasing the required computation time.