Increased computational performance benefits simulations run on high-performance computing systems. While Moore’s law on integrated circuits weakens, specialized processors promise to accelerate certain mathematical operations in many common software stacks. However, emerging processing elements must significantly reduce the total time to solution, allow for an energy reduction, and come with readily available APIs to justify financial and development efforts. We focus on comparing mesh decomposition implementations of finite element analysis, a task for which we use a readily available biomechanical real-world application called Direct Tensor Computation. The data of the mesh decomposition step and the software’s downstream behavior about the METIS serial graph partitioning for defining mesh parts is the ground truth of the study. We will compare this to the mesh decomposition’s quadratic unconstrained binary optimization (QUBO) formulation run on the laser-based processing unit (LPU) from LightSolver to determine its characteristics in an exemplary and exploratory manner. First, we investigate offline and online comparisons of the METIS and the LPU implementation. Once these results are promising, we apply the QUBO LPU implementation to OpenFoam, a widely used computational fluid dynamics software for large-scale simulations in HPC. This paper suggests an approach to investigating recurring tasks in engineering applications sent to devices in heterogeneous computing platforms. Succeeding publications will show in-depth results, while this paper concentrates on the concept and discussion of our proposed methodology.