Computing hardware, from commodity processors to supercomputers to cloud computing, is being "parallelized" at various levels to achieve higher speed and larger scale. The programming model for parallel finite element methods is the SPMD type using MPI and OpenMP, which seems to be well established. However, under the rapidly changing circumstances such as memory and network hierarchy and linkage with GPGPU, it is important to conduct benchmark analysis to determine how much of the total performance of the computer can be effectively used, especially when performing large-scale calculations. Through benchmark analysis, knowledge can also be gained regarding the introduction of parallel computers and computing resources when submitting parallel computing jobs. An FEM model for linear elastic analysis of a cast shape model named "mold" with parametrically varying the number of nodes to 1 million, 11 million, 50 million, and 200 million nodes was created as a benchmark example problem. The element type is the second-order tetrahedral. In this presentation, Results of the analysis are presented using the open-source parallel FEM structural analysis program FrontISTR. FrontISTR can perform hybrid parallel computations combining MPI parallelism based on domain partitioning and OpenMP parallelism for loop operations within each subdomain on a variety of hardware environments (from FUGAKU to desktop PCs). In addition, FrontISTR can be easily added on to the FreeCAD system as one of its workbenches. This enables a series of CAE procedures such as geometry creation, pre-processing, parallel FEM structural analysis, and post-processing to be performed in a free environment without license fees. The impact of the combination of the number of nodes(=CPUs) and cores used, the speedup due to MPI parallelism among nodes, the speedup due to OpenMP parallelism within a node, the impact of the combination of linear equations and preprocessing, and the sustainable performance vs. peak is discussed.