The investigation of the ARMv7 and Intel Haswell architectures suitability for performance and energy-aware computing

The investigation of the ARMv7 and Intel Haswell architectures suitability for performance and energy-aware computing

Stať ve sborníku v databázi WoS či Scopus

Reducing CPU frequency and voltage is a well-known approach to improve energy consumption of memory-bound applications. This is based on the conception that the performance of the main memory sees little or no degradation at reduced processor clock speeds, while power consumption decreases significantly and the overall energy efficiency improves. We study this effect on the Haswell generation of Intel Xeon processors as well as the ARMv7 generation of the 32-bit ARM big.LITTLE architecture. The goal is to analyze and compare performance, energy consumption and resulting energy efficiency on a series of tasks, each focusing on different parts of the system, provide analysis and generalization to other similar architectures. The benchmark suit consists of compute and memory intensive benchmarks as well as both single and multi-threaded scientific applications. The results show that frequency and voltage scaling can significantly improve algorithms' energy efficiency, up to 2.5x on ARM and 1.5x on Intel compared to the maximum frequency. ARM is up to 2x more efficient than Intel. The results are valuable for scientists in the areas of computer architecture, performance and power analysis as well as application developers seeking to optimize their Haswell x86_64 and ARMv7 hardware setup to improve energy efficiency.

Reducing CPU frequency and voltage is a well-known approach to improve energy consumption of memory-bound applications. This is based on the conception that the performance of the main memory sees little or no degradation at reduced processor clock speeds, while power consumption decreases significantly and the overall energy efficiency improves. We study this effect on the Haswell generation of Intel Xeon processors as well as the ARMv7 generation of the 32-bit ARM big.LITTLE architecture. The goal is to analyze and compare performance, energy consumption and resulting energy efficiency on a series of tasks, each focusing on different parts of the system, provide analysis and generalization to other similar architectures. The benchmark suit consists of compute and memory intensive benchmarks as well as both single and multi-threaded scientific applications. The results show that frequency and voltage scaling can significantly improve algorithms' energy efficiency, up to 2.5x on ARM and 1.5x on Intel compared to the maximum frequency. ARM is up to 2x more efficient than Intel. The results are valuable for scientists in the areas of computer architecture, performance and power analysis as well as application developers seeking to optimize their Haswell x86_64 and ARMv7 hardware setup to improve energy efficiency.

Haswell, ARMv7, Odroid XU4, k-Wave, LAMMPS, Energy Efficiency.

Haswell, ARMv7, Odroid XU4, k-Wave, LAMMPS, Energy Efficiency.

NIKL, V.; HRADECKÝ, M.; JAROŠ, J.; KELEČÉNI, J.

2018

20.06.2017

Springer International Publishing

Cham

978-3-319-58666-3

High Performance Computing

Lecture Notes in Computer Science

10266

377

393

16

https://www.fit.vut.cz/research/publication/11303/

ISC2017-inikl