GPU Intensive Modular Cluster Rendering for Hardware In-the-Loop Sensor Simulation

To address the U.S. Army need for a real-time physics-based near-infrared laser sensor scene generation system based on graphics processing unit (GPU) rendering algorithms, Physical Optics Corporation (POC) proposes to develop a new cluster accelerated simulation modular infrared (CASMIR) system based on GPU-intensive physics-based simulation algorithms exploiting both cluster- and node-level parallelism. The system takes full advantage of newly developed GPU architectures that combine highly programmable shading processors with hardware optimized to perform vector algebra, rapid table lookups, and other rendering operations. POC will also develop a novel form of cluster, combining a conventional Beowulf cluster with multiple PCI express (PCIe)-coupled GPU-based video cards in each node. Inter-node parallelism coupled with intra-node parallelism and the flexibility of GPU-based simulation enables CASMIR to perform real-time physics-based generation of high fidelity IR scenes. In Phase I POC will demonstrate the feasibility of the CASMIR system by means of a prototype cluster with intra- and inter-node parallelism. In Phase II POC plans to extend the CASMIR system to a high performance cluster capable of arbitrary IR simulation.