Modeling, Simulation, and Measurement - Computer Architecture Techniques for Power-Efficiency - page 9

CHAPTER 2

Modeling, Simulation,
and Measurement

As the power problem has become prominent for computer architects, many ideas have been
proposed for managing power and energy issues through architectural techniques. In order to
compare these many ideas, quantitative techniques for architecture-level power modeling have
become very important. Thus, in this chapter, we discuss some of the key issues and techniques
for the field of architecture-level power modeling. This sets the foundation for later chapters
in which power-efficient ideas are discussed and compared.

2.1 METRICS
The metric of interest in power studies varies depending on the goals of the work and the type
of platform being studied. This subsection offers an overview of possible metrics, and discusses
the best practices regarding when to use them.
Energy : Energy, in joules, is often considered the most fundamental of the possible
metrics, and is of wide interest particularly in mobile platforms where energy usage relates closely
to battery lifetime. Even in non-mobile platforms, energy can be of significant importance. For
data centers and other “utility computing” scenarios, energy consumption ranks as one of the
leading operating costs and thus reducing energy usage is crucial [ 71 , 203 ]
Power : Power is the rate of energy dissipation. The unit of power is watts (W), which is
joules per second. Power's time component makes it the meaningful metric for understanding
current delivery and voltage regulation on-chip. Another related metric, areal power density,
is power per unit area. This metric is useful for thermal studies; 200 W spread over many
square centimeters may be quite easy to cool, while 200 W dissipated in the relatively small
( 4cm 2 ) areas of today's microprocessor dies becomes challenging or impossible to cool in a

cost-effective manner.
Energy-per-instruction : In some situations, however, focusing solely on energy is not
enough. For example, reducing energy at the expense of lower performance may often not
be acceptable. Thus, metrics combining energy and performance have been proposed. For

∼