Increasing predictability of GPU’s
Laboratoire DALI-LIRMM, 52 avenue Paul Alduy, 66860 Perpignan Cerdex – France
hal-00951920, (25 February 2014)
@techreport{defour:hal-00951920,
hal_id={hal-00951920},
url={http://hal.archives-ouvertes.fr/hal-00951920},
title={Increasing predictability of GPU’s},
author={Defour, David},
keywords={scheduling; GPU; predictability; WCET},
language={Anglais},
affiliation={Laboratoire d’Informatique de Robotique et de Micro{‘e}lectronique de Montpellier – LIRMM},
year={2014},
month={Feb},
pdf={http://hal.archives-ouvertes.fr/hal-00951920/PDF/predictabilite_europar.pdf}
}
GPU’s are massively multicore architectures managing several thousands of concurrent threads. This concurrence, maintained through several schedulers, is necessary to keep high performance but negatively impact predictability. In this work, we first propose measures of predictability as well as CUDA tests to estimate this measure regarding warp and block scheduler for architectures from G80 to GK104. Finally, we evaluate the impact of hardware reset, underclocking and heavy synchronization to increase predictability on those architectures.
March 3, 2014 by hgpu
