This conference paper was submitted for presentation at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

Finite Element Method/Analysis (FEM/FEA) is a cornerstone of engineering simulation, enabling precise modeling of complex physical systems (and obtaining accurate results for complex engineering tasks). The recent developments in FEM software and the supporting hardware technologies have set the industrial product development on the path of a new paradigm shift, from simulation supported design to simulation driven design, where FEA is involved from the early stage and supports all stages of product development: from component to system level, from single to multi-physics and scales, and from pre-design to testing and certification (virtual testing); allowing thus for reduced number of design iterations and physical testing, which translates into reduced development costs and time to market. However, as the size and complexity of simulations grow, so does the computational demand, and this is why all FEM commercial software development companies have included robust multi-processor parallel computing capabilities in their portfolio, allowing thus the user to significantly accelerate analyses and tackle larger and more complex problems, closer to the real industrial designs, manufacturing conditions and operating environments. This article presents a benchmark study related to maximizing the computation speed and scalability using parallel processing in one of the most known FEM commercial software packages, namely Abaqus 2024 edition. The selected problem is representative to current simulation challenges in industry companies: multi-body and multi-material assembly, multiple contact interfaces, large deformations and non-linear material behavior. The assessment is run on a hardware configuration commonly used by industry companies, namely multi-CPU individual work station equipped with GPGPU. Quantitative metrics in terms of both hardware (number of CPU, GPU) and licensing (tokens) costs are presented, as applicable to both the Implicit and Explicit solvers in Abaqus. Similar problems setups of different sizes (number of degrees of freedom) are evaluated, in order to assess the scalability of the multi-processor parallel computing capabilities. The study offers guidelines for hardware selection, configurations, and best practices for multi-processor parallel computing in industrial environments.