Summary (2-4 sentences)
Radio Interferometers enable imaging the radio emission from celestial objects by combining signals from different antennas in a compute intensive process that may greatly benefit from GPU processing power and a high throughput computing model. We present an overview of this imaging process as well as current and future compute resource requirements. We discuss our HTCondor implementation for imaging in the PATh facility by distributing processing over multiple GPUs, as well as the development of a tool for data visualization that helps in high-level time profiling. We present testing results for measuring run time scalings of GPU-enabled imaging steps and successfully executing complete imaging procedures (imaging-to-convergence).
Primary author
Felipe Madsen
(National Radio Astronomy Observatory)