Scalable Multi-node Fast Fourier Transform on GPUs

TL;DR

This paper introduces a scalable multi-node GPU-FFT library optimized for high-performance computing, demonstrating efficient scaling on large GPU clusters with impressive performance metrics.

Contribution

The paper presents a novel multi-node GPU-FFT library employing slab decomposition and MPI, achieving scalable performance on large GPU clusters with detailed benchmarking.

Findings

Good scaling observed for 4096^3 grid with 64 to 512 GPUs

GPU-FFT timings comparable to multicore CPU FFT on large cores

Efficient communication via NVlink enhances GPU-FFT performance

Abstract

In this paper, we present the details of our multi-node GPU-FFT library, as well its scaling on Selene HPC system. Our library employs slab decomposition for data division and MPI for communication among GPUs. We performed GPU-FFT on $1024^3$, $2048^3$, and $4096^3$ grids using a maximum of 512 A100 GPUs. We observed good scaling for $4096^3$ grid with 64 to 512 GPUs. We report that the timings of multicore FFT of $1536^3$ grid with 196608 cores of Cray XC40 is comparable to that of GPU-FFT of $2048^3$ grid with 128 GPUs. The efficiency of GPU-FFT is due to the fast computation capabilities of A100 card and efficient communication via NVlink.