In this thesis, a high-performance GPU-based simulation- and analysis-environment for dynamical systems is designed, implemented, and tested. Computation is optimized on the basis of the GPU hardware architecture. Minimal task-specific simulation programs are created using runtime compilation. Scan execution and analysis methods are designed and optimized for massive parallelization and integrated into a modular, efficient, scalable, and streamlined architecture. The new program is tested for performance and result accuracy. The impact of floating-point inaccuracy is discussed and shows how the current implementation of AnTGPU can mitigate this problem.