Computer Simulation Technology
Development of a GPU-based PIC solver
Trainee: Martina Sofranac
Supervisor: Frank Hamme
The final goal of this project is to develop a Multi-GPU based Particle in Cell (PIC) solver so that simulations can be executed much faster in comparison with a CPU or single-GPU solution. Such implementation is needed since optimization of accelerator structures generally requires PIC simulations with huge amounts of particles. These simulations can be quite time consuming, sometimes up to several weeks. The particular challenge is to develop interpolation and load balancing schemes which take full benefit from modern GPU architectures.
Besides a standard domain decomposition scheme, alternative load balancing schemes need to be implemented. The standard scheme is only suitable for special applications, but in the context of this project a general solution is essential. The computation of 3D field distributions is of high relevance for almost all accelerator components. The more realistic an initial (numerical) design can be, the faster the progress towards an operating facility can happen.
Image courtesy of CST
Video:
Scientist in charge: