Communications - July 2012

performance data for each benchmark and processor. Figure 2 shows an example of this data, plotting the power versus performance characteristics for one of the 45 processor configurations, the stock i7 ( 45).

4. PeRsPeCtiVe

We organize our analysis into eight findings, as summarized in Figure 1. The original paper contains additional analyses and findings. We begin with broad trends. We show that applications exhibit a large range of power and performance characteristics that are not well summarized by a single number. This section conducts a Pareto energy efficiency analysis for all of the 45 nm processor configurations. Even with this modest exploration of architectural features, the results indicate that each workload prefers a different hardware configuration for energy efficiency.

4. 1. Power is application dependent

The nominal thermal design power (TDP) for a processor is the amount of power the chip may dissipate without exceeding the maximum transistor junction temperature. Table 1 lists TDP for each processor. Because measuring real processor power is difficult and TDP is readily available, TDP is often substituted for real measured power. Figure 3 shows that this substitution is problematic. It plots measured power on a logarithmic scale for each benchmark on each stock processor as a function of TDP and indicates TDP with an “✗.” TDP is strictly higher than actual power. The gap between peak measured power and TDP varies from processor to processor and TDP is up to a factor of four higher than measured power. The variation among benchmarks is highest on the i7 ( 45) and i5 ( 32), likely reflecting their advanced power management. For example, on the i7 ( 45), measured power varies between 23 W for 471.omnetpp and 89 W for fluidanimate! The smallest

Figure 2. Power/performance distribution on the i7 ( 45). each point represents one of the 61 benchmarks. Power consumption is highly variable among the benchmarks, spanning from 23 W to 89 W. the wide spectrum of power responses from different applications points to power saving opportunities in software.

90 100

80

Power (W)

70

60

50

40

30

20

2.00

3.00 4.00 5.00 6.00 7.00

Performance/Reference

Native non-scale Native scale Java non-scale Java scale

8.00

108 CommuniCations oF the aCm | juLy2012 | voL. 55 | no. 7

variation between maximum and minimum is on the Atom ( 45) at 30%. This trend is not new. All the processors exhibit a range of benchmark-specific power variation. TDP loosely correlates with power consumption, but it does not provide a good estimate for ( 1) maximum power consumption of individual processors, ( 2) comparing among processors, or ( 3) approximating benchmark-specific power consumption.

finding: Power consumption is highly application dependent and is poorly correlated to TDP.

Figure 2 plots power versus relative performance for each benchmark on the i7 ( 45), which has eight hardware contexts and is the most recent of the 45 nm processors. Native (red) and managed (green) are differentiated by color, whereas scalable (triangle) and non-scalable (circle) are differentiated by shape. Unsurprisingly, the scalable benchmarks (triangles) tend to perform the best and consume the most power. More unexpected is the range of power and performance characteristics of the non-scalable benchmarks. Power is not strongly correlated with performance across workload or benchmarks. The points would form a straight line if the correlation were strong. For example, the point on the bottom right of the figure achieves almost the best relative performance and lowest power.

4. 2. historical overview Figure 4(a) plots the average power and performance for each processor in their stock configuration relative to the reference performance, using a log/log scale. For example, the i7 ( 45) points are the average of the workloads derived from the points in Figure 2. Both graphs use the same color for all of the experimental processors in the same family. The shapes encode release age: a square is the oldest, the diamond is next, and the triangle is the youngest, smallest technology in the family.

Figure 3. measured power for each processor running 61 benchmarks. each point represents measured power for one benchmark. the “✗”s are the reported tDP for each processor. Power is application dependent and does not strongly correlate with tDP.