|
17.
12.
2013
Performance Portability and Energy Issues in Computing Architectures
Guilherme Calandrini
TU Berlin
The usage of different computing architectures such as CPU and GPUs, has
appeared as a novel alternative to increase computing performance as well
energy efficiency in computing applications, in what it is known
heterogeneous system. The OpenCL standard already promises a vender
neutral programming solution for CPUs, GPUs and DSPs, although the OpenCL
standard ensures a functional portability through different
architectures, the performance and energy of code executed on different
architectures are still a further issue for solve.
This presentation shows the results of an analysis made about performance
and power efficiency of distinct computing architectures (CPUs and GPUs)
using a suite of OpenCL micro benchmarks. The results are a key for
understanding the performance and power implication of optimization
strategies for the different architectures as well a guide for selecting
appropriate compute architecture according to performance or energy
efficiency requirements.
|