Speaker
Description
In high-intensity ion accelerators, particle dynamics is significantly influenced by nonlinear collective effects, particularly the space charge effects. The Particle-in-Cell (PIC) method is widely used to study these effects, but the computational burden restricts the applications in accelerator design, failure compensation, and machine learning dataset generation. In this work, a high-performance beam dynamcs code, named AVASX (Advanced Virtual Accelerator Software X), was developed with CUDA (Compute Unified Device Architecture) and performed on NVIDIA GPUs (Graphics Processing Units). The AVASX framework implements an adaptive particle tracking method that alternates between time and position as the independent variable. This switching mechanism optimizes the trade-off between the simulation accuracy and the computational efficiency. For the purpose of improving computing performance in charge deposition with atomic operations performed on GPU global memory, three optimization schemes, such as the thread aggregation method, the strategy of scattering data processing scopes, and the use of duplicate memory instances, were employed in this work and profiled using NVIDIA Nsight Compute to prove their effectiveness. According to the test results of simulating different beamlines, the GPU-based code is reliable and achieves 320 ~ 450 times faster than the CPU-based code performed on 56 cores.
| I have read and accept the Privacy Policy Statement | Yes |
|---|
