Dong Hyuk Woo and Hsien-Hsin S. Lee conclude in “Extending Amdahl’s Law for Energy-Efficient Computing in Many-Core Era.”, Computer, Vol. 41, No. 12, pp.24-31, December, 2008.
Extending Amdahl’s law to take power and energy into account, our analysis clearly demonstrates that a symmetric many-core processor can easily lose its energy efficiency as the number of cores increases. To achieve the best possible energy efficiency, our work suggests a many-core alternative, featuring many small, energy-efficient cores integrated with a full-blown processor. Our analytical models also show that by knowing the amount of parallelism available in an application prior to execution, we can find the optimal number of active cores for maximizing performance for a given cooling capacity and energy in a system. To further optimally control the number of active cores adaptively, future many-core runtime must be capable of dynamic per-core power profiling and have a feedback mechanism to manage thread dispatch.
See also their (and Intel’s) POD (parallel-on-demand), as in “POD: A 3D-integrated broad-purpose acceleration layer” and
For another recent attempt at extending Amdahl’s Law see “Amdahl’s Law in the Multicore Era” by Mark D. Hill.
And for a bottomless cup of multicore coffee, see MultiCoreInfo.com.