首页|AVASX: A GPU-accelerated High-performance Beam Dynamics Simulation Code Using Hybrid Tracking Methods for Ion Linear Accelerators
AVASX: A GPU-accelerated High-performance Beam Dynamics Simulation Code Using Hybrid Tracking Methods for Ion Linear Accelerators
AVASX: A GPU-accelerated High-performance Beam Dynamics Simulation Code Using Hybrid Tracking Methods for Ion Linear Accelerators
In the study of beam dynamics for high-intensity linear accelerators, 3D multi-particle simulations based on the particle-in-cell (PIC) algorithm are necessary due to numerous nonlinear effects caused by strong space charge effects. However, since such simulations require a large amount of computations, the simulations running on traditional CPUs (Central Processing Units) are always inefficient and time-consuming, which restricts their simulation scales and applications. In this work, a high-performance beam dynamics simulation code, AVASX (Advanced Virtual Accelerator Software X), designed based on a CPU-based code, AVAS (Advanced Virtual Accelerator Software), is developed with CUDA (Compute Unified Device Architecture) and runs on NVIDIA GPUs (Graphics Processing Units). AVASX enables the adaptive and dynamic switching between two particle tracking methods, one using time as the independent variable (t-code) and the other using position (z-code). This switching mechanism optimizes the trade-off between the simulation accuracy and the computational efficiency. To improve the computing performance of charge deposition when performing atomic operations on GPU global memory, three optimization schemes, which are the thread aggregation method, the strategy of scattering data processing scopes, and the approach of using duplicate memory instances, are proposed in this work and profiled using NVIDIA Nsight Compute to verify their effectiveness. According to the test results of simulating different beamlines, AVASX running on an A100 GPU is reliable and achieves a 174 ~ 550x speedup over AVAS running on 56 physical CPU cores, leading to the simulation durations being reduced from days or hours to a few minutes or seconds. Based on AVASX, the simulation tasks such as accelerator design, fault compensation, and machine learning datasets generation could be performed in a high-efficiency way.
加速器计算技术、计算机技术
Beam dynamics simulation Ion linear accelerators Particle-in-Cell algorithm GPU-accelerated code
Yuan Tian,Zhongyi Li,Changwei Hao,Chao Jin,Xin Qi,Yaxin Hu,Zhijun Wang,Yuan He.AVASX: A GPU-accelerated High-performance Beam Dynamics Simulation Code Using Hybrid Tracking Methods for Ion Linear Accelerators[EB/OL].(2025-11-03)[2025-11-07].https://chinaxiv.org/abs/202511.00004.点此复制
In the study of beam dynamics for high-intensity linear accelerators, 3D multi-particle simulations based on the particle-in-cell (PIC) algorithm are necessary due to numerous nonlinear effects caused by strong space charge effects. However, since such simulations require a large amount of computations, the simulations running on traditional CPUs (Central Processing Units) are always inefficient and time-consuming, which restricts their simulation scales and applications. In this work, a high-performance beam dynamics simulation code, AVASX (Advanced Virtual Accelerator Software X), designed based on a CPU-based code, AVAS (Advanced Virtual Accelerator Software), is developed with CUDA (Compute Unified Device Architecture) and runs on NVIDIA GPUs (Graphics Processing Units). AVASX enables the adaptive and dynamic switching between two particle tracking methods, one using time as the independent variable (t-code) and the other using position (z-code). This switching mechanism optimizes the trade-off between the simulation accuracy and the computational efficiency. To improve the computing performance of charge deposition when performing atomic operations on GPU global memory, three optimization schemes, which are the thread aggregation method, the strategy of scattering data processing scopes, and the approach of using duplicate memory instances, are proposed in this work and profiled using NVIDIA Nsight Compute to verify their effectiveness. According to the test results of simulating different beamlines, AVASX running on an A100 GPU is reliable and achieves a 174 ~ 550x speedup over AVAS running on 56 physical CPU cores, leading to the simulation durations being reduced from days or hours to a few minutes or seconds. Based on AVASX, the simulation tasks such as accelerator design, fault compensation, and machine learning datasets generation could be performed in a high-efficiency way.
Beam dynamics simulation Ion linear accelerators Particle-in-Cell algorithm GPU-accelerated code
展开英文信息
评论