COPPER: Compiler-Controlled On-Demand Approach to Power-Efficient Computing
Abstract
The goal of this research was to build and demonstrate a capability in hardware (processors-memory) and software for management of power resources, and explore its trade-off against speed, accuracy and throughput requirements. The emphasis of this project was as much on utilization of the most power efficient architectural and software techniques as on their coordinated management across hardware and software levels. Through a coordinated management of power-control knobs from compiler to architectural and micro-architectural strategies we were able to achieve a range of power adaptation versus performance needs. However, instead of only leveraging on low power design approaches used in the embedded world and building on our own previous research on adaptive system software, we were also able to investigate power adaptation at the hardware and software levels. Our techniques collectively achieved a factor of 2-4x power/energy reduction.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2003
- Accession Number
- ADA418787
Entities
People
- Alex Nicolau
- Alex Veidenbaum
- Nikil Dutt
- Rajesh Gupta
Organizations
- University of California, Irvine