Instruction-Level Parallelism and Its Exploitation - Computer Architecture: A Quantitative Approach - page 238

FIGURE 3.43 The CPI for the 19 SPECCPU2006 benchmarks shows an average CPI for
0.83 for both the FP and integer benchmarks, although the behavior is quite different . In
the integer case, the CPI values range from 0.44 to 2.66 with a standard deviation of 0.77,
while the variation in the FP case is from 0.62 to 1.38 with a standard deviation of 0.25. The
data in this section were collected by Professor Lu Peng and Ph.D. student Ying Zhang, both
of Louisiana State University.
The FP benchmarks achieve higher performance with a lower average CPI (0.89) and a lower
standard deviation (0.25). For the FP benchmarks, L1 and L2 are equally important in determ-
ining the CPI, while L3 plays a smaller but significant role. While the dynamic scheduling and
nonblocking capabilities of the i7 can hide some miss latency, cache memory behavior is still a
major contributor. This reinforces the role of multithreading as another way to hide memory
latency.
3.14 Fallacies and Pitfalls
Our few fallacies focus on the difficulty of predicting performance and energy efficiency and
extrapolating from single measures such as clock rate or CPI. We also show that different ar-
chitectural approaches can have radically different behaviors for different benchmarks.