VDPred: Predicting Voltage Droop for Power-Effient 3D Multi-core Processor Design

Modern applications present rapid, time-varying demands to the power supply systems. Designing for worst-case voltage droop leads to inefficient operations. These inefficiencies are amplified in 3D multi-core processors. Minimizing this voltage variation is key to improving power efficiency. One widely accepted method is to apply droop compensation into the voltage regulators. However, as the design complexity rises, the response time of the droop compensation fails to keep pace with the speed of load current switching in the modern processors. In this paper, we propose VDPred, a droop prediction system integrated with a reactive control loop inside a typical on-chip buck regulator of the 3D processor. VDPred predicts and eliminates voltage droops that produce worst-case operating voltage margins. It constructs an offline learning model for power prediction based on the pipeline and cache microarchitectural events. When VDPred predicts a hazardous droop, the circuitry warms up the regulator and pulls up the supply voltage against a voltage loss before the droop happens. VDPred can deal with very fast droop events, on the order of nanoseconds, and thus the voltage guard-band and power consumption of the processor can be further reduced compared to existing state-of-the-art techniques. We evaluate VDPred in both single- and four-core configurations of a 3D processor with the cycle-accurate full-system simulator. Results show a significant reduction in voltage guard-band and up to 7% runtime power of the overall system.

[1]  Gaël Varoquaux,et al.  Scikit-learn: Machine Learning in Python , 2011, J. Mach. Learn. Res..

[2]  Arun Ravindran,et al.  A machine learning approach to modeling power and performance of chip multiprocessors , 2011, 2011 IEEE 29th International Conference on Computer Design (ICCD).

[3]  Pang-Jung Liu,et al.  Fast-transient DC–DC converter with hysteresis prediction voltage control , 2017 .

[4]  Yingying Kuai,et al.  Comprehensive and practical optimization of voltage regulator modules , 2011, 2011 IEEE Power and Energy Conference at Illinois.

[5]  J.R. Struharik,et al.  Implementing decision trees in hardware , 2011, 2011 IEEE 9th International Symposium on Intelligent Systems and Informatics.

[6]  Krzysztof Piotr Sozanski Modification of reactive power control algorithm , 2010, Signal Processing Algorithms, Architectures, Arrangements, and Applications SPA 2010.

[7]  Sanu Mathew,et al.  Exploiting Fully Integrated Inductive Voltage Regulators to Improve Side Channel Resistance of Encryption Engines , 2016, ISLPED.

[8]  H. Matsuo,et al.  A current compensation circuit for VRM in the sudden change of the load , 2005, 2005 European Conference on Power Electronics and Applications.

[9]  Bin Li,et al.  Accurate and efficient processor performance prediction via regression tree based modeling , 2009, J. Syst. Archit..

[10]  Monodeep Kar,et al.  An All-Digital Fully Integrated Inductive Buck Regulator With A 250-MHz Multi-Sampled Compensator and a Lightweight Auto-Tuner in 130-nm CMOS , 2017, IEEE Journal of Solid-State Circuits.

[11]  Sudhakar Yalamanchili,et al.  Eiger: A framework for the automated synthesis of statistical performance models , 2012, 2012 19th International Conference on High Performance Computing.

[12]  Meeta Sharma Gupta,et al.  Predicting Voltage Droops Using Recurring Program and Microarchitectural Event Activity , 2010, IEEE Micro.

[13]  Pradip Bose,et al.  Voltage Noise in Multi-Core Processors: Empirical Characterization and Optimization Opportunities , 2014, 2014 47th Annual IEEE/ACM International Symposium on Microarchitecture.

[14]  Kai Li,et al.  PARSEC vs. SPLASH-2: A quantitative comparison of two multithreaded benchmark suites on Chip-Multiprocessors , 2008, 2008 IEEE International Symposium on Workload Characterization.

[15]  André Carlos Ponce de Leon Ferreira de Carvalho,et al.  A study of cross-validation and bootstrap as objective functions for genetic algorithms , 2002, VII Brazilian Symposium on Neural Networks, 2002. SBRN 2002. Proceedings..

[16]  Michael D. Smith,et al.  Voltage Smoothing: Characterizing and Mitigating Voltage Noise in Production Processors via Software-Guided Thread Scheduling , 2010, 2010 43rd Annual IEEE/ACM International Symposium on Microarchitecture.

[17]  Jun Wang,et al.  Manifold: A parallel simulation framework for multicore systems , 2014, 2014 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS).