Soft GPGPU versus IP cores: Quantifying and Reducing the Performance Gap

TL;DR

This paper evaluates the performance gap between soft GPGPUs and specialized IP cores on FFT tasks, proposing architectural improvements that significantly enhance soft GPU efficiency, making them competitive for FPGA-based digital signal processing.

Contribution

The paper quantifies the performance gap and introduces two novel architectural features for eGPU that improve FFT efficiency by 50%, reducing the performance-area gap with specialized IP cores.

Findings

eGPU achieves high clock frequencies (>750 MHz) and small footprint.

Modified eGPU reduces the performance-area gap to 3x that of specialized IP cores.

eGPU demonstrates superior efficiency compared to Nvidia A100 GPGPUs for FFTs.

Abstract

eGPU, a recently-reported soft GPGPU for FPGAs, has demonstrated very high clock frequencies (more than 750 MHz) and small footprint. This means that for the first time, commercial soft processors may be competitive for the kind of heavy numerical computations common in FPGA-based digital signal processing. In this paper we take a deep dive into the performance of the eGPU family on FFT computation, in order to quantify the performance gap between state-of-the-art soft processors and commercial IP cores specialized for this task. In the process, we propose two novel architectural features for the eGPU that improve the efficiency of the design by 50\% when executing the FFTs. The end-result is that our modified GPGPU takes only 3 times the performance-area product of a specialized IP core, yet as a programmable processor is able to execute arbitrary software-defined algorithms. Further comparison to Nvidia A100 GPGPUs demonstrates the superior efficiency of eGPU on FFTs of the size studied (256 to 4096-point).