Journal's Facebook pageComparison and Analysis of Software and Hardware Energy Measurement Methods for a CPU+GPU System and Selected Parallel Applications
- Faculty of Electronics, Telecommunications and Informatics, Gdańsk University of Technology
Narutowicza 11/12, 80-233 Gdańsk, Poland
pczarnul@eti.pg.edu.pl
Abstract
In this paper authors extend upon their previous research on powercapped optimization of performance-energy metrics of deep neural networks training workloads. A professional power meter Yokogawa WT-310E is used, as well as Intel RAPL and Nvidia NVML interfaces, to examine power consumption of a much more comprehensive set of multi-GPU and multi-CPU workloads, including: selected kernels from NAS Parallel Benchmarks for CPUs and GPUs as well as Horovod-Python Xception deep neural network training using several GPUs. A comparison and discussion of results obtained by both power measurement methods has been performed using 2 systems, one with 2 Intel Xeon CPUs and 8 Nvidia Quadro RTX 6000 GPUs and the second 2 Intel Xeon CPUs and 4 Nvidia Quadro RTX 5000 GPUs. We compared power consumption between hardware and software interfaces for CPU, GPU and mixed CPU+GPU workload configurations, using 1-40 threads for the CPUs and 1-8 GPUs.
Key words
high performance computing, performance-energy optimization, energy measurements, measurement accuracy, parallel benchmarks
Full text
Available in PDF
Portable Document Format
How to cite
Koszczał, G., Matuszek, M., Czarnul, P.: Comparison and Analysis of Software and Hardware Energy Measurement Methods for a CPU+GPU System and Selected Parallel Applications. Computer Science and Information Systems
