Accelerated Processing Unit

from Wikipedia, the free encyclopedia

Accelerated Processing Unit (short APU ; English for "accelerated processing unit"), also Advanced Processing Unit (English "extended processing unit") is a term based on the abbreviations "CPU" and "GPU" for a main processor with integrated coprocessors , which is used almost exclusively by AMD . As the name suggests, the main processor is to be accelerated by the coprocessor (s) in certain situations. A graphics processor (GPU), vector processor , stream processor or any other processor that is superior to the main processor in certain situations can serveas coprocessor. Only a mathematical coprocessor in the form of a floating point unit (also abbreviated as "FPU"), which - among others - was integrated in the main processor long before the definition of the APU term and thus became part of the main processor, is excluded from this.


The term Accelerated Processing Unit was created and coined by the microprocessor manufacturer AMD , who uses this term to identify main processors with an integrated graphics unit . AMD Fusion is the code name of the processor concept that CPU, GPU, video and other hardware accelerators on a The united.

The first announcements for the integration of a graphics processor core in a main processor were made in 2006, after the takeover of ATI by AMD. From May 2010, prototypes were delivered to the first customers and at Computex in spring 2010 it was announced that the APU models Llano and Ontario will go on sale in the first half of 2011. In addition, in the run-up to CES 2011 it was announced that the first computers in the form of tablet PCs and computers based on the APU series Zacate and Ontario will be delivered in the first quarter of 2011.

Intel has also been delivering main processors with integrated graphics units with its Core I series since 2012, but does not use the term 'APU'. For example, the GPU in the Ivy Bridge processors can also be used for universal computer calculations thanks to the support of OpenCL and Direct Compute . Other competitors do not use the "APU" term either.

Technical details

To be able to call a chip 'APU', the processor (CPU) should contain one or more main processor cores and at least one additional coprocessor for special tasks. The additional coprocessor is usually a graphics processor (GPU). The main aim of this concept is to combine the advantages of different processor types / processor architectures. The concept itself is therefore also based on multi-core processor systems as heterogeneous computing (English: heterogeneous computing ) because now heterogeneous processor types occur in a system.

Since graphics processors gave up part of their specialization in graphics in the course of their development history and increasingly became freely programmable processors that can, in particular, process data-parallel tasks very quickly, the manufacturers saw the possibility here of using main processors optimized for sequential tasks with this type of coprocessor to complete. Since every PC needs a graphics processor anyway, the transition in the time when the corresponding software for the coprocessor is missing should be bridged by the pure advantages of a graphics processor.

Decisive for the designation APU is accordingly not only the integration of a graphics processor in the main processor, but also the suitability of this graphics processor for the calculation of universal tasks apart from pure graphic calculations.

Graphics processors were used for GPGPU purposes very early on , but the programming was done very close to the hardware and could therefore not easily be transferred to other graphics processors, which also meant that the distribution of corresponding programs and tools remained very limited. Another important step from a pure main processor with IGP to the so-called APU is the support of manufacturer-independent programming standards such as OpenCL and quasi-standards such as Direct Compute in the DirectX -11 API from the graphics processor.

In addition to the programmable GPU parts, the GPU also has hard-wired units (Fixed Function Units) that can be used apart from the graphics. The video decoding units such as UVD or Nvidia's video processor have been used for a long time to decode certain codecs and thus also relieve the main processor. In addition to such decoding units, Intel's integrated GPU in Sandy Bridge CPUs also have hard-wired encoding units with which the encoding of videos can be accelerated with certain codecs. Since both encoding and decoding are only limited to certain codecs and are not freely programmable, corresponding CPUs are still referred to as "CPUs with IGP and additional acceleration functions" and not as APUs, given the concept here, if the APU definition is narrowly interpreted the connection of several processor architectures with different advantages is missing and only one processor architecture is specialized by hard wired units in some areas.

Individual evidence

  1. Fusion: AMD relies on processors with integrated graphics - Article at , from October 25, 2006
  2. AMD has the first samples of the Fusion APUs - article at , from May 19, 2010
  3. Fusion: AMD shows demo and names date - article at Heise online , from June 2, 2010
  4. AMD shows for the first time Fusion processor APU with DirectX-11 - article at , from June 2, 2010
  5. Zacate and Ontario: AMD presented Fusion CPUs - Article at , January 4, 2010
  6. IDF: Intel presents the next processor generation Ivy Bridge , news from September 14, 2011.
  7. Processors 2010: The merger begins - Article at , from January 2, 2010