The basic graphics processor unit—GPU is recognized as one of the most significant developments in computer processing, exceeding the DSP in importance, and challenging the ubiquity of the CPU.
GPUs can be stand-alone or integrated or embedded. As such, we have assigned lower case prefixes to help distinguish them. The more popular ones are listed here:
dGPU — discrete or stand-alone GPU, typically found on an AIB, but can also be found on system or motherboards in mobile PCs and various industrial and scientific systems.
iGPU — integrated or embedded, a GPU that shares the die with another processor, typically a RISC or CISC CPU such as an Arm or x86.
eGPU — external GPU, typically an AIB in an external cabinet or chassis connected to a PC via a high-speed cable (e.g., Thunderbolt 3 or USB 3c). This is a controversial label because eGPU can also stand for embedded GPU, which is one that is tightly coupled and packaged with a CPU, but not on the same die. Intel’s multi-chip Kaby Lake-G with AMD dGPU is an example.
vGPU — virtual GPU is when the GPU (on an AIB) is in a remote location and the local or client machine uses it. This has also been called a rGPU, for remote.
There are other examples of where the term, or more accurately acronym GPU is used. For example:
GP-GPU — when a GPU is used for compute (and not graphics) is often called GP-GPU. The more correct nomenclature is GPU-compute, but the industry has gravitated to the easier (and more fun) to say GP-GPU term.
OpenGPU — an AMD initiative designed to provides access to a comprehensive collection of visual effects, productivity tools, and other content at no cost.
DT-GPU — a desktop GPU, typically an AIB.
NB-GPU — a GPU used in a notebook (mobile PC) typically on the system or motherboard.
|The various GPUs relationship to one another|
The diagram is both inclusive and exclusive. Typically, one would have a lite machine with maybe an iGP and be connected to a server for a vGPU. Or an iGPU system connected to an eGPU. That would be exclusive of all other GPUs.
However, there’s nothing to prevent one from having a heavy-weight desktop system with an AIB (with a dGPU on it), and a CPU with an iGPU in it, and connected to a server for a vGPU and connected to an external chassis for an eGPU. That would be an inclusive example.
And all the combinations between the two are possible.