Modeling, Simulation, and Measurement - Computer Architecture Techniques for Power-Efficiency - page 11

FIGURE 2.1: Performance/power tradoff zones. Reproduced from [ 84 ]. Copyright 2003 Intel.
design. In discussing design decisions in Intel's Pentium M processor, Gochman et al. used a
particularly compelling graphic to convey the different options [ 84 ]. Figure 2.1 reproduces it
here.
The figure depicts performance gains (or loss) on the x -axis, and power improvements
or degradations on the y -axis. Clearly, a technique that degrades both power and performance
is of little appeal; these lie in the upper-left quadrant of the graph. The other three quadrants
contain regions that represent possible power-performance options under different constraints.
The magenta region represents the realm in which both performance and energy are improved.
The orange region represents cases where performance improves (e.g. subject to a thermal
constraint) despite overall energy loss, and the green region shows where energy improves
despite a loss in performance.
Other metrics and issues : The remainder of this chapter and topic focuses on modeling
and research ideas primarily in the context of the above metrics. There are other metrics and
design goals that are also relevant, although space constraints preclude us from discussing them
in detail. In particular, we briefly discuss thermal modeling because of its relevance to leakage
energy modeling, but we cannot discuss thermal-aware design techniques in deep detail.
2.2 MODELING BASICS
This section gives an overview of modeling techniques for each of the primary power-related
issues: dynamic power, leakage power, and thermal behavior. Because issues of modeling
and simulation are particularly intertwined for leakage power and thermal behavior, we cover
simulators and simulation in these sections as well. Dynamic power simulation, however, is a
more mature and well-trodden area; as such, we cover it in Section 2.3 which follows.