Increasing Performance of the NCAR Command Language via Parallel Processing

07/26/2011 - 1:45pm to 2:05pm
Main Seminar Room - ML
Christopher Kruse

Christopher Kruse, University of North Dakota

Abstract:  This study focused on increasing the performance of the NCAR Command Language (NCL) through parallel processing.  OpenCL 1.0 was used to parallelize codes for commodity graphics processing units (GPUs) and multi-core central processing units (CPUs) that an average NCL user might have access to.  Simple operators (“+”,”*”, etc), transcendental functions, matrix multiplication, and two built in functions with high arithmetic intensity were parallelized.  The parallelized interpreter was tested on a variety of commodity hardware.  Results vary with quality of hardware, but significant performance increases were achieved by parallelizing matrix multiplication and the built-in NCL function "gc_inout", on every system.  On systems with a multi-core CPU and a lower performance GPU, the parallelized code performed better on the multi-core CPU compared to the GPU and native NCL.

Video replay of the presentation

Please install the Flash Plug-in or click here for non-flash supported devices.

Download the entire presentation (right-click/save link as)

Presented on July 26, 2011 at NCAR