“An Effective POD based Thermal Simulation

and its Enabled Thermal-Aware Scheduling”

Guest Speaker; Dr. Yu Liu

Associate Professor, Department of Electrical and Computer Engineering

Clarkson University

MS Teams Meeting 

Time: 3 – 4 p.m., Friday, April 12, 2024

Abstract:

With advanced technical innovation, recent computing systems increasingly rely on multi-core CPUs and many-core GPUs, which inevitably results in more challenges to thermal management due to their advanced architectures and high-power consumption. Thermal-aware scheduling has been proven to be an effective proactive measure to address this challenge. However, traditional methods using RC thermal circuits in the thermal-aware are largely based on assumptions of steady-state uniform core temperature. These approaches have difficulty in complying with modern computing systems. A physics-informed proper orthogonal decomposition (POD), together with Galerkin projection (GP) of the heat transfer equation, is proposed to offer a fast and accurate thermal prediction with high spatial resolution for thermal-aware scheduling. The objective of this project is to develop an effective thermal-aware scheduling scheme for modern computing systems enabled by the proposed POD-GP methodology, which is able to capture all crucial hot spots, adapt to sudden workload change by utilizing accurate transient temperature, suppress hot spots, and control the peak temperature in CPUs/GPUs with a large number of cores. This could extend the reliability, improve the performance, and reduce the energy consumption. Our evaluation results demonstrate that our POD-GP thermal simulation outperforms the traditional method by thousands of times faster, while maintaining the high accuracy. Moreover, utilizing POD-based thermal-aware scheduling has proven to significantly reduce both the peak thermal variance and the peak chip temperature.

Bio:

Dr. Yu Liu is an Associate Professor in the Department of Electrical and Computer Engineering at Clarkson University. Prior to joining Clarkson University, he was a research scientist at the Canadian Nuclear Laboratories (CNL) from 2013 through 2017. In addition, he was employed at Motorola as a senior software engineer from 2003 through 2007, and IBM from 2011 through 2013. He received his B.S. and M.S. degrees from Sichuan University in 2000 and 2003, respectively, and his Ph.D. degree from Southern Illinois University Carbondale in 2011. His research interests include high-performance computing, computer architectures, real-time systems, and wireless sensor networks. He has published over 50 peer-reviewed research papers.

Event Details

See Who Is Interested

0 people are interested in this event

User Activity

No recent activity