Instruction-Level Parallelism and Its Exploitation - Computer Architecture: A Quantitative Approach - page 228

FIGURE 3.34 The parallel benchmarks used here to examine multithreading, as well as
in Chapter 5 to examine multiprocessing with an i7 . The top half of the chart consists of
PARSEC benchmarks collected by Biena et al. [2008] . The PARSEC benchmarks are meant
to be indicative of compute-intensive, parallel applications that would be appropriate for mul-
ticore processors. The lower half consists of multithreaded Java benchmarks from the
DaCapo collection (see Blackburn et al. [2006] ) and pjbb2005 from SPEC. All of these bench-
marks contain some parallelism; other Java benchmarks in the DaCapo and SPEC Java
workloads use multiple threads but have little or no true parallelism and, hence, are not used
here. See Esmaeilzadeh et al. [2011] for additional information on the characteristics of these
benchmarks, relative to the measurements here and in Chapter 5 .