Скачать или смотреть Modern methods of energy consumption optimization in FPGA-based heterogeneous HPC systems

Modern methods of energy consumption optimization in FPGA-based heterogeneous HPC systems
O. Hryshchuk, S. Zagorodnyuk

High-Performance Computing (HPC) systems play a pivotal role in addressing complex computational challenges across various domains, but their escalating energy consumption has raised concerns regarding sustainability and operational costs. This paper presents a comprehensive investigation into the parametrization and modeling of energy consumption in heterogeneous HPC systems, aiming to provide valuable insights for optimizing energy efficiency while preserving performance.
We begin by characterizing the heterogeneity within modern HPC environments, which encompass diverse hardware components, such as CPUs, GPUs, FPGAs, and accelerators.
Our research delves into modeling techniques, leveraging heuristics methods and statistical approaches to construct accurate predictive models for energy consumption. Furthermore, we explore the integration of dynamic power management strategies, such as DVFS (Dynamic Voltage and Frequency Scaling) and task scheduling, to optimize energy usage without compromising performance.
This paper provides a vital foundation for sustainable HPC practices, enabling researchers and practitioners to make informed decisions for achieving enhanced energy efficiency without sacrificing computational performance.