OpenCL

OpenCL is an open standard for writing programs that execute parallel workloads across CPUs, GPUs, and other heterogeneous processors.

OpenCL kernel: An OpenCL function compiled for execution across many work-items on a device. Kernels contain the parallel computation that the host program dispatches.
OpenCL platform: An OpenCL implementation that exposes devices and associated runtime services. A platform identifies the vendor or software stack through which devices are accessed.
GPGPU: The use of graphics processing units for general-purpose computation beyond graphics rendering. OpenCL provides a way to express general workloads for GPU execution.
CUDA: NVIDIA’s parallel computing platform and programming model for its GPUs. CUDA offers a mature, NVIDIA-specific alternative to OpenCL’s cross-vendor standard.
Khronos Group: A member-funded consortium that develops open standards for graphics, media, and parallel computing. The consortium maintains OpenCL and coordinates its specification process.
NDRange: An OpenCL execution domain that organizes work-items into one-, two-, or three-dimensional index spaces. It defines the global and local shape of a kernel launch.
OpenCL device: A processor exposed by an OpenCL implementation as a target for computation. Programs select devices such as CPUs, GPUs, or accelerators for execution.
Image processing: The transformation or analysis of images using computational operations. Independent pixels and image regions often map naturally to OpenCL work-items.
SYCL: A C++ programming model for heterogeneous computing built around single-source code. SYCL offers a newer abstraction that can target OpenCL and other backends.
Apple Inc.: A technology company that develops consumer electronics, operating systems, and software. Apple initiated the OpenCL specification and contributed its early framework.

Linked from 6 pages