PIConGPU - A scalable GPGPU implementation of the particle-in-cell algorithm

We present PIConGPU, an efficient and scalable implementation of the particle-in-cell algorithm for GPGPUs. We discuss the main building blocks of PIConGPU, the data access patterns used for both particle and field data and the communication model that allows to hide the large latency of network communication between GPGPU nodes on a cluster. PIConGPU provides a general framework which can be used to study both relativistic and nonrelativistic plasmas. We show first results on relativistic laser wakeeld acceleration of electrons in underdense plasmas and on the progress of integrating new physics models. The fast response time of the code makes it possible to receive results in hours compared to weeks with particle-in-cell codes running on mid-size commodity clusters. With this increase in computational
speed extensive parameter scans become possible even for large physical systems.