As shown in Figure 3, the execution time of the CPU is much
longer than that of the GPU. In general, the GPU processes a large
amount of operations more efficiently than the CPU, because the
many-core based GPU is optimized for parallel processing.
Especially, mathematical or scientific applications can get high
performance by utilizing the many-core GPU. Since the executed
benchmark applications (SAD, CP, TPACF, MM) require parallel
operations for high performance, the GPU can process the
operations more efficiently than the CPU. As shown in Figure 4,
energy consumption is also reduced significantly when the
application is executed on the GPU compared to the CPU. In this
evaluation, energy efficiency is mainly determined by the execution time. Figure 5 shows the temperature of two processing
units. In the graph, CPU temperature goes higher when the
application is executed on the CPU than the GPU. GPU
temperature also goes higher when the application is executed on
the GPU than the CPU, since the temperature goes up as more
resources are utilized. Temperature is a very important factor in
determining the system efficiency, because high temperature in the
processor causes negative impact on reliability, lifetime, and
performance[4][5][6]. The error rates of electronic devices
increase exponentially as the temperature goes up[18][19].
Moreover, the lifetime of the processor is reduced by half if the
temperature increases by about 10~15