Clarkson University has recently been awarded a three-year National Science Foundation (NSF) grant for a research project entitled “CDS&E: An Effective Thermal Simulation Methodology for GPGPUs Enabled by Data-Driven Model Reduction.” This new award will fund talented graduate students to work with Clarkson faculty to investigate a novel thermal simulation methodology for general purpose GPUs (GPGPUs). Professors Ming-Cheng Cheng and Yu Liu of Electrical and Computer Engineering Department will co-lead this project.
Demands for GPGPUs (Graphics Processing Units) in recent years have increased rapidly due to the needs for scientific, engineering and statistical computing. With hundreds or thousands of cores running in each GPGPU, severe heating is a serious challenge that can significantly degrade GPGPU performance, reliability and energy efficiency. To ease all these problems, effective thermal management and thermal-aware task scheduling for GPGPU operation are needed, which however requires an accurate simulation tool that is able to offer efficient dynamic thermal prediction with a reasonable spatial resolution. Currently, there is a lack of thermal simulation tools that offer high efficiency and accuracy with a reasonable resolution.
The goal of this project is to develop a multi-block simulation methodology for efficient, accurate prediction of dynamic thermal profiles of GPGPUs derived from a reduced learning algorithm. To reduce simulation space and thus the computational time while maintaining accurate thermal solution, the domain structure of a GPGPU is projected onto a functional space described by a set of basis functions obtained from the reduced learning method. This methodology offers a reduction in the computational time by several orders of magnitude for thermal simulation of semiconductor chips, compared with the direct numerical simulation. It is expected that the developed thermal simulation model will be even more efficient than the widely used compact resistance-capacitance (RC) model. Also, the multi-block approach possesses a natural advantage of effective parallel computing, which will further speed up the thermal simulation of GPGPUs.
Read more about the award at https://www.nsf.gov/awardsearch/showAward?AWD_ID=2003307