The CPU (Consuming Power Unlimited)

The CPU forms the core of any modern computing machine. To be able to perform operations at clocks as high as 4.5 GHz, CPUs need enormous power. In the electronics world, the (dynamic) power consumed by a component is estimated as:

Power = Capacitance * Voltage2 * frequency

As we can observe, an increase in clock speed directly translates to increased power draw. An Ivy Bridge i7 die having 177 mm2 area, dissipating 100 W implies a power density of 565 kW/m2, while power density of the sun’s radiation on the surface of the earth is approximately 1.4 kW/m2(which is why overclocking burns the chip). Modern CPUs consume more than 200 W of power at peak performance, and often represent the biggest drain on your laptop’s battery, followed by the GPU and then display screens.

To save power, manufacturers developed the concept of P-states and C-states, which basically shut down some parts of the CPU according to the available load. Any modern CPU (assuming recent manufacturing process <45 nm) should consume less than 10 W at idle state. An interesting fact to note here would be the power consumption of fans running to keep the system cool. A single fan can eat up to 5 W, and a group (common use case) together may consume as much as 20 Watts, which can overshadow the motherboard+RAM+CPU combined.

The users have indirect influence over the CPU’s power conditions and battery use through their selection of power settings. The TDP numbers associated with a chip imply the maximum power according to which the cooling systems need to be designed, not the actual power in day-to-day scenarios. Any reports on power consumption which give statements like “the processor’s power consumption is 35 watts” are both false and imprecise, because the actual number is far more dynamic and dependent on multiple factors.

One might assume that a CPU heavy application should increase only CPU power usage, but the motherboard also has to supply the data at higher rates, which translates to increased disk I/O. This implies that the motherboard, buses, RAM and data disks, all consume more power to deliver the higher data throughput. Since these subsystems are intertwined and hardware manufacturers rarely provide actual power consumption numbers, our best bet to estimate the individual power draw would be to develop regression models to predict the numbers. This is precisely what is done by PowerTop.

 

Advertisements