Radio-Astronomical Imaging: FPGAs vs GPUs

Abstract

FPGAs excel in performing simple operations on high-speed streaming data, at high (energy) efficiency. However, so far, their difficult programming model and poor floating-point support prevented a wide adoption for typical HPC applications. This is changing, due to recent FPGA technology developments: support for the high-level OpenCL programming language, hard floating-point units, and tight integration with CPU cores. Combined, these are game changers: they dramatically reduce development times and allow using FPGAs for applications that were previously deemed too complex. In this paper, we show how we implemented and optimized a radio-astronomical imaging application on an Arria 10 FPGA. We compare architectures, programming models, optimizations, performance, energy efficiency, and programming effort to highly optimized GPU and CPU implementations. We show that we can efficiently optimize for FPGA resource usage, but also that optimizing for a high clock speed is difficult. All together, we demonstrate that OpenCL support for FPGAs is a leap forward in programmability and it enabled us to use an FPGA as a viable accelerator platform for a complex HPC application.