Improved Multi-GPU parallelization of a Lagrangian Transport Model

TL;DR

This paper explores enhancing GPU parallelization for a Lagrangian Transport Model by enabling a single process to utilize multiple GPUs, aiming to improve resource efficiency and performance.

Contribution

It introduces a method allowing one MPI process to access multiple GPUs, overcoming OpenACC limitations and reducing resource contention.

Findings

Single-process multi-GPU support improves resource utilization

Reducing process count decreases contention and overhead

Enhanced parallelization benefits model performance

Abstract

This report highlights our work on improving GPU parallelization by supporting compute nodes with multiple GPUs. However, since the default support for multi-GPUs in OpenACC is limited[6], the current implementation allows each MPI process to access only a single GPU. Thus, the only way to take full advantage of multi-GPU nodes in the current version is to launch multiple processes, which increases resource contention. We investigated the benefits of having only one process offload to all available GPU devices.