2 energy efficiency ideals and the iron law

Energy efficiency is the work done per unit amount of energy consumed. Maximizing energy efficiency is important as it allows more work to be done for a given energy budget and also allows work to be done faster for a given power budget. This has economic and environmental benefits as it minimizes the energy needed to do a given computation. While compute capability, in terms of the number of transistors per chip, has steadily increased (Moore’s Law [158]), operating voltage has not reduced in proportion (limited Dennard scaling [63]). Borkar and Chien [35] observed that although Pollack’s rule [170] predicts a speedup potential (beyond speedups in transistor switching) in proportion to the square root of the number of transistors in a processor, energy-efficiency concerns discourage many microarchitectural techniques that can enable those performance gains. Thus, improving energy efficiency will also improve processor performance. Energy-proportional computing, that uses energy in proportion to the work done, is an important concept for energy-efficient systems since it seeks to eliminate energy waste by only using as much energy as the work done. However, modern computers are not energy proportional. For example, they use non-trivial power when they are powered on but not used. This is due to processor leakage power, DRAM refresh, and power draw