Graphics processing unit

The graphics processor (English Graphics Processing Unit - GPU, rare Visual Processing Unit - VPU ) is used to calculate the screen output to computers, game consoles and smartphones. This is either in the die of the CPU with integrated graphics unit ( as so-called APU ) on the motherboard (onboard, as an integrated graphics processor, now less common ) or on an expansion card ( slot card ). In the latter case, multiple GPUs on a graphics card, or even multiple graphics cards per PC are possible.

Almost all currently produced graphics processors for personal computers come from AMD, Intel and Nvidia. The integration of plug-in cards, however, lies down to special and reference models for some time with other manufacturers.

• 4.1 Current
• 4.2 Past

History

Precursor of GPUs since there were about in the early 1980s. At the time, this only served as a link between the CPU and the screen output and video display controllers were therefore called. Neither had the functionality, nor were they designed for independent computations. First, they were primarily responsible for an independent text and graphics output, thus sparing the system. Some were later able to represent at least Sprites independently.

That changed in the mid-1980s with computers such as the Commodore Amiga and the Atari ST. These already had blitting functions. In the x86 -PC sector GPUs came with such additional functions on the transition from text-based DOS programs to graphics-based Windows programs. Building blocks such as the ET -4000 were simple commands (eg, " draw rectangle " ) execute independently. Due principally for use under Windows they were also called "Windows Accelerator ".

Mid-1990s, the first 3D accelerators came on the market. This graphic processors were capable of some effects and triangle -based algorithms ( such as Z- buffering, texture mapping ) and perform anti-aliasing. Especially the field of computer games helped those plug-in cards (such as the 3dfx Voodoo Graphics ) to a surge in development. At the time, such applications were mainly limited by the processor.

The term was first used by Nvidia GPU intensively to market the Nvidia GeForce 1999 published -256 series. This graphics card was equipped ( in the retail business) as the first with a T & L unit.

GPUs were superior because of their specialization in graphics calculations and focus on massively parallelizable tasks to CPUs in their computing power always. A CPU is universally designed, the individual CPU cores are also usually optimized for fast execution of sequential tasks. The GPU on the other hand is characterized by a high degree of parallelization, because 3D calculations can be parallelized very well and is also specialized for 3D calculations so remain for certain tasks (eg for texture filtering) specialized units ( "Fixed Function Units" ) contained in the GPU. If you were piloting on a GPU to run a program that only one or two threads with relatively little data parallelism ( currently standard program ) makes available the same time, the GPU would be underutilized. The relatively small caches on the GPU would lead to larger latencies in program execution, which could be compensated due to lack of parallelism of the program, not by simultaneous execution of many tasks. In sequential tasks, the CPU is therefore faster.

The performance advantage over CPUs in highly parallelizable tasks and existing SIMD features make current GPUs for scientific applications as a stream processor interesting. This purpose is called GPGPU. The inclusion of the GPU has led, for example, in the distributed computing project Folding @ home to an enormous increase in computing power. She was initially limited to the chip manufacturer ATI / AMD, came in 2008, but also Nvidia GPUs from the GeForce 8 series added. For graphics cards from Nvidia CUDA exists as an API to use the GPU for calculations. This is now also used to perform physics calculations in games using PhysX. Meanwhile, there are the open programming platform OpenCL can be developed with the programs for the CPU and GPU alike. In addition, can not count only single-precision, double-precision but today's GPUs.

Tasks

The graphics processor handles compute-intensive tasks of 2D and 3D computer graphics and freeing up the main processor (CPU). The functions are controlled via software libraries such as DirectX or OpenGL. The vacant processor time can thus be used for other tasks.

Properties

• Support graphics interface DirectX and OpenGL
• Antialiasing - some angle-independent anti-aliasing
• Anisotropic Filtering - illustration / screening of textures
• Multi-GPU technologies - collaboration of multiple graphics processors
• Free programmability almost any GPU component ( shader, includes T & L)
• Texture - pattern illustration, using at least one Texture Mapping Unit ( TMU )
• Image synthesis, with the help of at least one raster operation Processor ( ROP), also known as Render Output Unit, or Grid Operations Pipeline

Power consumption

After the major manufacturers of CPUs have started since about the beginning of 2005 to reduce the power consumption of their products, especially at low workload part very clearly, was in this respect a pressure on the manufacturers of graphics processors that previously but rather doing the opposite: high-end graphics cards walk not rare to even without load more than 50 W in heat loss, although there is virtually no difference in performance to much simpler models or onboard graphics in this state. The end of 2007 added AMD with the ATI Radeon HD 3000 series for the first time efficient power saving features in its desktop graphics cards. Nvidia developed the method Hybrid Power, which made ​​it possible to eliminate a high-end graphics card in 2D mode and switch to the frugal onboard graphics chip, for which, however, a Hybrid SLI -enabled motherboard was a prerequisite. After a relatively short time, Nvidia, approved by this concept. In the meantime (2009) dominate the GPUs of both manufacturers relatively efficient power saving mechanisms.

Manufacturer

Current

AMD (formerly ATI), ARM Limited, Qualcomm, Intel, Matrox, Nvidia, PowerVR, S3 Graphics

Former

3dfx, 3Dlabs, Cyrix, Diamond Multimedia, NeoMagic, Oak Technology, Inc. S3, SiS, Trident Microsystems, Tseng Labs, XGI Technology Inc.

Due to the strong competition and high development costs associated with most manufacturers were bought ( 3dfx, 3Dlabs ) or focus on a niche market ( Matrox, XGI ).

For years, Intel is the clear market leader in graphics processors for PCs. The main reason is the high number of office computers that are equipped almost exclusively with integrated on the motherboard GPUs, which are a part of the chipsets mainly supplied by Intel. In for PC gamers more appropriate area of ​​the plug-in card-based graphics solutions, AMD / ATI and Nvidia share the market.