Advanced Computing in the Age of AI | Friday, April 26, 2024

Notes from the European COMSOL Multiphysics Conference 

Wolfgang Gentzsch surveys conference participants and finds that many, despite the use of workstations, PCs and a few HPC clusters, are experiencing severe resource limitations.

The purpose of my visit to the European COMSOL Multiphysics conference in Ludwigsburg, Germany, October 26 – 28, (along with 420 participants from industry and research) was twofold: to learn more about the participating engineers’ use of computers for a relatively complex application software tool such as COMSOL Multiphysics; and the penetration of High Performance Computing (HPC) in this community.

The COMSOL Multiphysics simulation software environment facilitates all steps in the modeling process – defining your geometry, meshing, specifying your physics, solving, and then visualizing your results.

More than 95 percent of COMSOL licenses are on PCs and workstations, a situation, which can be observed with other CAE codes. They offer single user basic licenses, plus additional modules for many specific applications, and floating (network) basic licenses, which are independent of the number of jobs a single user is submitting, and the number of cores the jobs are using, as long as they come from one and the same user and are submitted to the same system. This makes COMSOL especially attractive for multi-core, multi-node systems. The price of a software license is still several times that of a workstation, quite common in CAE.

I conducted two types of interviews. First, I asked 52 participants from industry what computers they are using for their simulations. The result was: 17 are mainly using PCs, 33 are mainly using workstations, and two are mainly using an HPC cluster – i.e. 33 percent are using mainly PCs, 63 percent are using mainly workstations, and four percent HPC clusters.

Second, to learn more, I interviewed 20 conference participants mainly from industry, asking them five questions about their specific use of hardware:

1.    What are your applications that you are using COMSOL for?
2.    For the simulation of your application, are you mainly using PCs, workstations, or HPC clusters?
3.    Are you satisfied with the performance of your system, or do you regularly experience certain limits (e.g. runtime, memory size, number of cores)?
4.    Do you have HPC expertise in your group?
5.    Are you considering using cloud computing for your simulations in the foreseeable future?

Summary of Interviews

I interviewed representatives from the following organizations: Bosch Rexroth, Fresenius, Oechsler, PTB, Wolf, Wenger, Microvisk, IFM-Geomar, ABB, Mercedes, KIT, Zeiss, Continental Powertrain, Teknova, Esaote, Bosch, Philips, Nestle, ZBT, and Continental Teves. Below are short summaries of the interview answers:

1.    What are your engineering applications that you are using COMSOL for? – CFD for pumps and engines, blood transfusion, structural mechanics, CFD for plastics, acoustics, medical devices for endoscopy, thermodynamics, micro-mechanical structures, for medical diagnostics and sensors, CO2 storage in porous media, electro-thermal heating in power transmission systems, metal corrosion, materials, thermal and electromagnetic effects in electron microscopy, mass air flow in turbo loaders, ultrasound, energy storage and conversion, electronic readers and ink, heat and mass viscous flows in food production, fuel cells, and electronic brake systems.

2.    For your application, are you mainly using PCs, workstations, and/or HPC clusters? – 16 participants (out of a sample of 20) are using (mainly) workstations, seven are using PCs, and five are making some use of clusters (sometimes remotely, from another department, computer center, or neighboring university).

3.    Are you satisfied with the performance of your system, or do you regularly experience certain limits (e.g. runtime, memory size, number of cores)? – From the 15 participants who do not use HPC clusters, 10 report experiencing severe resource limitations. Some of the more common limitations are: jobs are running far too slow/long; memory isn’t sufficient; can only run 2D (or axisymmetric) geometries; parametric studies have to be performed serially; can’t do full physics; etc.

4.    Do you have HPC expertise in your group? - From the 20 participants asked, 12 of them explicitly said that they don’t have any experience in HPC (clusters, MPI, etc.).

5.    Are you considering Cloud Computing for your simulations in the foreseeable future? - Nobody uses or is considering cloud computing so far, and almost all were uncertain about the benefits of cloud computing especially for their simulations.

The HPC PRIMERGY Division of Fujitsu in Germany sponsored the study.  

EnterpriseAI