Heterogeneous computing

Article By:

Wang, Kevin G. Department of Aerospace and Ocean Engineering, Virginia Polytechnic Institute and State University, Blacksburg, Virginia.

Last reviewed:2014

DOI:https://doi.org/10.1036/1097-8542.YB150584

In the context of scientific computing, heterogeneous computing refers to the use of different types of processors, such as the central processing unit (CPU) and the graphics processing unit (GPU), in a given scientific computing task. The objective is to improve performance by assigning different parts of the task to specialized processors. Heterogeneous computing is an emerging technology with great potential in many scientific and engineering fields where large-scale computation plays an important role. Examples include, but are not limited to, fluid dynamics, molecular dynamics, and quantum mechanics. Today, a large percentage of computers used for scientific computing, including most workstations and many clusters, are equipped with both CPUs and GPUs, which makes heterogeneous computing possible. On the other hand, the increase of computing power in CPUs has significantly slowed down in recent years due to both architecture limitations and physical constraints (for example, the “power wall”), which motivates the use of other available resources. Originally designed for image rendering, GPUs specialize in parallel vector operations on very large data sets. In the past decade, a number of languages, compilers, and libraries have been developed for programming GPUs for scientific computing, and significant speedup (up to 100-fold) has been achieved in various applications by assigning part of the computing task to GPUs.

The content above is only an excerpt.