Distributed Parallelization of xPU Stencil Computations in Julia

TL;DR

This paper introduces a simple method for distributed parallelization of stencil computations on GPUs using Julia, enabling efficient scaling and communication hiding for large-scale applications.

Contribution

It presents a new approach implemented in ImplicitGlobalGrid.jl for scalable distributed stencil computations on GPUs with communication hiding techniques.

Findings

Achieves close to ideal weak scaling on thousands of GPUs

Effectively hides communication costs behind computation

Demonstrates practical scalability for real-world applications

Abstract

We present a straightforward approach for distributed parallelization of stencil-based xPU applications on a regular staggered grid, which is instantiated in the package ImplicitGlobalGrid.jl. The approach allows to leverage remote direct memory access and enables close to ideal weak scaling of real-world applications on thousands of GPUs. The communication costs can be easily hidden behind computation.