A GPU-based Correlator X-engine Implemented on the CHIME Pathfinder

TL;DR

This paper details a cost-effective, power-efficient GPU-based correlator system for the CHIME Pathfinder radio telescope, utilizing consumer hardware and custom software to handle large-scale radio data processing.

Contribution

It introduces a novel GPU-based correlator built with consumer-grade hardware and OpenCL, achieving high performance at low cost and power consumption.

Findings

Correlates 32,896 baselines over 400MHz bandwidth

Uses AMD GPUs with packed integer operations for efficiency

Achieves 105 TOPS within 10kW power budget

Abstract

We present the design and implementation of a custom GPU-based compute cluster that provides the correlation X-engine of the CHIME Pathfinder radio telescope. It is among the largest such systems in operation, correlating 32,896 baselines (256 inputs) over 400MHz of radio bandwidth. Making heavy use of consumer-grade parts and a custom software stack, the system was developed at a small fraction of the cost of comparable installations. Unlike existing GPU backends, this system is built around OpenCL kernels running on consumer-level AMD GPUs, taking advantage of low-cost hardware and leveraging packed integer operations to double algorithmic efficiency. The system achieves the required 105TOPS in a 10kW power envelope, making it among the most power-efficient X-engines in use today.