Copyright by William Lloyd Bircher 2010 - The Laboratory for ...

Table 3.4 Instruction Linear Regression Model Comparison
Coefficients
Retired
instructions/cyc
29
Completed
instructions /cyc
α0 α1 α0 α1
36.8 5.28 36.3 5.52
Avg Error 5.45% 4.92%
Coefficient of
Determination
0.679 0.745
Of the µop-based models fetched µops is the most representative metric for power
consumption. This suggests that µops that do not update the architected state of the
machine still consume a significant amount of power. For the case of cancelled µops,
this is not surprising since these µops did complete execution but were not retired. So,
they would have traversed nearly the entire processor pipeline consuming a similar power
level as retired µops. More surprising is the effect of fetched µops on the power model.
Fetched µops includes retired and cancelled operations. It also includes the remaining
µops that were cancelled before completing execution. Since fetched µops provides the
most accurate model, cancelled µops must be consuming a significant amount of power.
These models generate minimum and maximum power values (36W – 47W) similar to
what was found on a Pentium 3 (31W-48W) [Be00] with similar µop/cycle ranges (0 –
2.6). The stated average error values are found using the validation set described in
Table 3.1.