Can GPU Performance Increase Faster Than the Code Error Rate? - Archive ouverte HAL Accéder directement au contenu
Article Dans Une Revue Journal of Supercomputing Année : 2024

Can GPU Performance Increase Faster Than the Code Error Rate?

Résumé

Graphics Processing Units (GPUs) are the reference architecture to accelerate High Performance Computing (HPC) applications and the training/interference of Convolutional Neural Networks (CNNs). For both these domains, performance and reliability are two of the main constraints. It is believed that the only way to increase reliability is to sacrifice performance, e.g., using redundancies. We show in this paper that this is not always the case. As a very promising result, we found that most GPUs performance improvements also bring the benefit of increasing the number of executions correctly completed before experiencing a Silent Data Corruption (SDC). We consider four different common GPUs' performance optimizations: architectural solutions, software implementations, compiler optimizations, and threads degree of parallelism. We compare different implementations of a variety of parallel codes and, through beam experiments and applications profiling, we show that the performance improvement typically (but not necessarily) increases the GPU SDC rate. Nevertheless, for the vast majority of the configurations the performance gain is much higher than the SDC rate increase, allowing to process a higher amount of correct data. As we show, the programmer choices can increase up to 25x the number of correctly completed executions without redesigning the algorithm nor including specific hardening solutions.
Fichier principal
Vignette du fichier
sn-article.pdf (2.69 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04528798 , version 1 (02-04-2024)

Licence

Paternité

Identifiants

Citer

Fernando Fernandes dos Santos, Paolo Rech. Can GPU Performance Increase Faster Than the Code Error Rate?. Journal of Supercomputing, In press, pp.1-32. ⟨10.1007/s11227-024-06119-4⟩. ⟨hal-04528798⟩
17 Consultations
5 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More