Evaluating graphical processing units for clutter model calculation in a simulation facility

摘要

Graphics processing units (GPUs) are being fused together with the traditional central processing units (CPUs) to form accelerated processing units (APUs). These hybrid computing machines provide a viable alternative to traditional processors for high-performance computing systems thanks to substantial computational performance and advanced programmability. A team at Oak Ridge National Laboratory experienced with radar clutter modeling and simulation, and with computation using a variety of processors, has explored the viability of GPUs for accelerating clutter simulation for the Joint Research Analysis and Assessment Center at the Missile and Space Intelligence Center. The results show substantial performance gains of approximately 20× relative to a CPU-only algorithm. We show a number of ways these accelerated clutter calculations provide real-time results starting from readily extended simulation models.