Simulation has become an essential tool for understanding and optimizing additive manufacturing (AM) processes. Nevertheless, conventional simulation techniques are often too slow and computationally demanding to be deployed in real-time applications or embedded design pipelines to serve as digital twins. In this study, we propose a high-performance, GPU-accelerated simulation framework tailored for AM. This framework integrates particle-based meshfree methods with strategies from computer graphics and massively parallel computing. The developed framework is capable of modeling a wide range of physical phenomena involved in AM, including fluid flow, heat transfer, solidification, and mechanical deformation, all within a unified solver architecture. By discretizing the domain into particles and leveraging Smoothed Particle Hydrodynamics (SPH), we circumvent the limitations of mesh-based approaches and naturally handle large deformations, moving boundaries, and material transitions. The simulations are implemented entirely on the GPU, enabling real-time performance on high-end desktop hardware. This allows for fast prototyping and interactive feedback, which is beneficial for AM researchers and engineers. The system's modular and extensible nature allows for the integration of multiple AM processes, materials, and control strategies, providing a foundation for training machine learning-based controllers. The capabilities of this approach are demonstrated through several benchmark scenarios relevant to AM, including material deposition, thermal evolution, and multi-material interface dynamics. The findings indicate that high-fidelity, real-time AM simulation is attainable, thereby unveiling novel prospects for design, optimization, and autonomous control within digital manufacturing workflows. This work establishes the foundation for future simulation-based control and learning in AM systems, drawing inspiration from the role of fast simulation in robotics and reinforcement learning.