Of the various current GPGPU implementations, OpenCL is by far the most interesting GPGPU API due to the absence of limitations. It is available for virtually all mainstream GPUs and platforms, even enjoying support on select mobile platforms.

Another distinguishing feature of OpenCL is that it's not limited to just GPGPU either. As part of its name (Open Computing Language), it abstracts a system into the so-called compute devices, each with their own capabilities. GPGPU is the most common application, but this feature makes it fairly easy to test implementations on a CPU first, for easy debugging.

One possible disadvantage of OpenCL is that it employs a high level of abstraction for memory and hardware details, which can negatively affect performance, even as it increases the portability of the code.

In the rest of this chapter, we will focus on OpenCL.