Advanced Programming Platform for efficient use of Data Parallel Hardware

TL;DR

This paper presents a programming platform with a visual editor designed to facilitate efficient use of GPU hardware for data-parallel tasks in distributed HPC clusters, demonstrated through FFT and image compression applications.

Contribution

It introduces a novel visual programming platform that simplifies deploying GPU-accelerated data-parallel algorithms in distributed environments.

Findings

Enhanced performance in FFT and image compression tasks.

User-friendly visual editor for parallel data flow programming.

Effective deployment of GPU hardware in HPC clusters.

Abstract

Graphics processing units (GPU) had evolved from a specialized hardware capable to render high quality graphics in games to a commodity hardware for effective processing blocks of data in a parallel schema. This evolution is particularly interesting for scientific groups, which traditionally use mainly CPU as a work horse, and now can profit of the arrival of GPU hardware to HPC clusters. This new GPU hardware promises a boost in peak performance, but it is not trivial to use. In this article a programming platform designed to promote a direct use of this specialized hardware is presented. This platform includes a visual editor of parallel data flows and it is oriented to the execution in distributed clusters with GPUs. Examples of application in two characteristic problems, Fast Fourier Transform and Image Compression, are also shown.