New power budgeting and thermal management scheme for multi-core systems in dark silicon
Hai Wang, Ming Zhang, Sheldon X.-D. Tan, Chi Zhang, Yuan Yuan, Keheng Huang, Zhenghong Zhang · 2016
As integrated circuit technology advances, we are now facing the utilization wall of the chip, which prevents us from turning on all the cores at the same time and leads to dark silicon. In order to maximize performance without violating thermal threshold for dark silicon chips, we propose a new power budgeting and dynamic thermal management scheme. Utilizing model predictive control, the new method automatically determines power budget values as well as active core positions considering transient thermal effects. Then, a dynamic thermal management method with task migration and DVFS is used to assign appropriate task loads to active cores according to the previously determined power budget. Experiments on multi-core systems with different number of cores and different dark silicon ratios have shown the new method is able to calculate the power budget accurately, keep the chip in safe temperature range, and retain the system in high performance.