Accelerating Geospatial Modeling in ArcGIS with Graphical Processor Units

Geospatial data can be enormous in size and tedious to process efficiently on standard computational workstations. Distributing the processing tasks through highly parallelized processing reduces the burden on the primary processor and processing times can drastically shorten as a result. ERSI's ArcGIS, while widely used in the military, does not natively support multi-core processing or utilization of graphic processor units (GPUs). However, the ArcPy Python library included in ArcGIS 10 provides geospatial developers with the means to process geospatial data in a flexible environment that can be linked with GPU application programming interfaces (APIs). This research extends a custom desktop geospatial model of spatial similarity for remote soil classification which takes advantage of both standard ArcPy/ArcGIS geoprocessing functions and custom GPU kernels, operating on an NVIDIA Tesla S2050 equipped with potential access to 1792 cores. The author will present their results which describe hardware and software configurations, processing efficiency gains, and lessons learned.