Instruction-Level Parallelism and Its Exploitation - Computer Architecture: A Quantitative Approach - page 243

Fallacy Processors With Lower CPIs Will Always Be Faster.: Processors With
Faster Clock Rates Will Always Be Faster
The key is that it is the product of CPI and clock rate that determines performance. A high
clock rate obtained by deeply pipelining the CPU must maintain a low CPI to get the full be-
neit of the faster clock. Similarly, a simple processor with a high clock rate but a low CPI may
be slower.
As we saw in the previous fallacy, performance and energy efficiency can diverge signiic-
antly among processors designed for different environments even when they have the same
ISA. In fact, large differences in performance can show up even within a family of processors
from the same company all designed for high-end applications. Figure 3.46 shows the integer
and FP performance of two different implementations of the x86 architecture from Intel, as
well as a version of the Itanium architecture, also by Intel.
FIGURE 3.46 Three different Intel processors vary widely . Although the Itanium processor
has two cores and the i7 four, only one core is used in the benchmarks.
The Pentium 4 was the most aggressively pipelined processor ever built by Intel. It used
a pipeline with over 20 stages, had seven functional units, and cached micro-ops rather than
x86 instructions. Its relatively inferior performance given the aggressive implementation, was
a clear indication that the atempt to exploit more ILP (there could easily be 50 instructions in
light) had failed. The Pentium's power consumption was similar to the i7, although its tran-
sistor count was lower, as its primary caches were half as large as the i7, and it included only
a 2 MB secondary cache with no tertiary cache.
The Intel Itanium is a VLIW-style architecture, which despite the potential decrease in com-
plexity compared to dynamically scheduled superscalars, never atained competitive clock
rates with the mainline x86 processors (although it appears to achieve an overall CPI similar
to that of the i7). In examining these results, the reader should be aware that they use diferent
implementation technologies, giving the i7 an advantage in terms of transistor speed and
hence clock rate for an equivalently pipelined processor. Nonetheless, the wide variation in
performance—more than three times between the Pentium and i7—is astonishing. The next
pitfall explains where a significant amount of this advantage comes from.