high performance computing on graphics processing units: hgpu.org

Parallel programming on GPUs is introduced in the context of simulating collision energy loss and bremsstrahlung for charged particles propagating in matter. The employed Monte Carlo methods and the involved physics are presented, followed by an introduction to the concepts of GPU parallel programming for the Nvidia CUDA architecture. The simulations implemented in C++ and CUDA are described. The physics results of the implemented simulations are presented, and the simulations are benchmarked for various configurations between the CPU and GPU implementations, comparing and discussing the results. The GPU implementation is shown to gain a significant advantage at high particle multiplicity for collision energy loss, while the CPU implementation performs and scales better with the amount of secondary particles produced by bremsstrahlung.