GPU-Powered Coherent Beamforming

TL;DR

This paper presents a GPU-optimized coherent beamforming implementation for radio astronomy that significantly outperforms CPU versions and is integrated into real-time pipelines at the BEST-2 array.

Contribution

The authors developed and optimized a CUDA-based GPU beamformer for radio astronomy, enabling real-time processing and integration into existing pipelines.

Findings

Achieves ~1.3 TFLOPs on NVIDIA Tesla K20

Approximately 10x faster than CPU implementation

Successfully integrated into real-time pipelines at BEST-2

Abstract

GPU-based beamforming is a relatively unexplored area in radio astronomy, possibly due to the assumption that any such system will be severely limited by the PCIe bandwidth required to transfer data to the GPU. We have developed a CUDA-based GPU implementation of a coherent beamformer, specifically designed and optimised for deployment at the BEST-2 array which can generate an arbitrary number of synthesized beams for a wide range of parameters. It achieves $\sim$1.3 TFLOPs on an NVIDIA Tesla K20, approximately 10x faster than an optimised, multithreaded CPU implementation. This kernel has been integrated into two real-time, GPU-based time-domain software pipelines deployed at the BEST-2 array in Medicina: a standalone beamforming pipeline and a transient detection pipeline. We present performance benchmarks for the beamforming kernel as well as the transient detection pipeline with beamforming capabilities as well as results of test observation.