Title:

Performance impact of DVFS for molecular dynamics simulations on Tesla K40 GPU

Author:

Astsatryan Hrachya

Type:

Conference

Co-author(s) :

Narsisian Wahi ; Poghosyan Aghasi ; Shahinyan Artashes

Uncontrolled Keywords:

Gromacs ; GPU benchmarking ; energy-efficient computing ; dynamic frequency scaling ; HPC

Abstract:

Nowadays the most crucial challenge in High-Performance Computing (HPC) is to optimize the power consumption of resources and to couple the performance with energy efficiency. The graphics processing units (GPUs) have been used intensively due to their computational performance, which are significantly accelerate the execution of many HPC applications. A series of studies have been carried out based on GROMACS (GROningen MAchine for Chemical Simulations) package to reveal the nature of NVIDIA Tesla K40 graphical card with different frequencies and to estimate the energy efficiency from frequency viewpoint. The analyzes show that the performance of the Tesla K40 is equivalent to the performance of the 1024 cores of IBM BlueGene/P supercomputer or the 64 CPUs of HP CP4000BL nodes. In the meantime, the usage of low frequencies leads to the energy conservation up to 20-30%. It is stated that the optimal power efficiency is achieved with the low-frequency GPU.

Publisher:

IEEE

DOI:

10.23919/MIPRO.2018.8400158

ISBN:

978-953-233-095-3 ; 978-1-5386-3777-7

Language:

English

Volume:

41

URL:


Affiliation:

Institute for Informatics and Automation Problems ; International Scientific-Educational Center

Country:

Croatia

Year:

2018

Time period:

21-25. April

Conference title:

2018 41st International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO)

Place:

Opatija