The potential in energy efficiency of a speculative chip-multiprocessor

ABSTRACT

While lower supply voltage is effective for energy reduction, it suffers performance loss. To mitigate the loss, we propose to execute only the part, which does not have any influence on execution speed, with low-speed and low-voltage. We are investigating a multithreaded execution, named Contrail Architecture, which divides an instruction stream into two streams using trace-level value prediction. One is the speculation stream, which is the main part of a program and is applied value predictions. The other is the verification stream, which verifies the predictions. The energy consumption is reduced by the decrease in the execution time in the speculation stream and by the low-speed execution in the verification stream. This paper evaluates its potential in energy efficiency.

REFERENCES

• 1: Takenori Koushiro , Toshinori Sato , Itsujiro Arita, A trace-level value predictor for Contrail processors, ACM SIGARCH Computer Architecture News, v.31 n.3, June 2003 [doi>10.1145/882105.882112]
• 2: M. Pilla et al.: Predicting trace inputs with dynamic trace memorization: determining speedup upper bounds, PACT, 2001.
• 3: Manoj Franklin, Multiscalar Processors, Kluwer Academic Publishers, Norwell, MA, 2002
• 4: Karthik Sundaramoorthy , Zach Purser , Eric Rotenburg, Slipstream processors: improving both performance and fault tolerance, Proceedings of the ninth international conference on Architectural support for programming languages and operating systems, p.257-268, November 2000, Cambridge, Massachusetts, United States
• 5: E. Larson et al.: MASE: A novel infrastructure for detailed microarchitectural modeling, ISPASS, 2001.
• 6: Lucian Codrescu , D. Scott Wills, On Dynamic Speculative Thread Partitioning and the MEM-Slicing Algorithm, Proceedings of the 1999 International Conference on Parallel Architectures and Compilation Techniques, p.40, October 12-16, 1999
• 7: Pedro Marcuello , Antonio González , Jordi Tubella, Thread Partitioning and Value Prediction for Exploiting Speculative Thread-Level Parallelism, IEEE Transactions on Computers, v.53 n.2, p.114-125, February 2004 [doi>10.1109/TC.2004.1261823]
• 8: M. Guthaus et al.: MiBench: a free, commercially representative embedded benchmark suite, WWC, 2001.

INDEX TERMS

General Terms:
Design, Experimentation, Performance

Keywords:
dual pipeline, energy efficiency, speculative multithreading