This thesis presents the development of an Octree pattern-based parallel solver for simulation of elastic wave propagation in large regions, utilizing Graphics Processor Units (GPUs) with NVIDIA’s CUDA technology. The study adopts an automatic preprocessing method based on octree-based mesh generation. This approach efficiently manages neighbouring elements of varying sizes. Hanging nodes are eliminated by incorporating polyhedral elements, which are effectively handled within the scaled boundary finite element method (SBFEM). The solver supports both implicit and explicit time integration schemes. This approach significantly reduces the computational cost and memory requirement by exploiting the limited number of master cells in a balanced octree grid and is advantageous for GPU computation. Parallelization is achieved using mesh-partitioning techniques and message-passing-interface (MPI) directives, supplemented by the NVIDIA Collective Communication Library (NCCL) for optimal performance in high-performance computing (HPC) environments. To demonstrate the solver’s capabilities, several real-world scenarios are modelled. Notable examples include an image-based 3D model of a portion of the Moon’s surface, featuring a layered structure and approximately 440 million degrees of freedom, simulated using the explicit solver. This example achieved a remarkable speed-up of 865 on a single computational node equipped with eight NVIDIA A100 GPUs operating in parallel. Additionally, the GPU-based HHT-alpha method (implicit solver) is applied to simulate wave propagation in a virtual city with underground tunnels, involving approximately 61 million degrees of freedom. To further enhance the framework’s robustness for higher frequency excitations and greater accuracy, a high-order time integration scheme based on partial fractions is implemented within the HPC framework. This high-order scheme enables more accurate modelling of complex dynamic responses compared to the second order implicit method such as HHT-alpha. To validate its effectiveness, the framework is applied to analyze a virtual city subjected to excitations with higher frequency contents, demonstrating its capability to handle challenging seismic scenarios with superior accuracy.