A graphics card controls the graphics output in a computer . When a program is executed, the processor calculates the data, forwards it to the graphics card and the graphics card converts the data so that the monitor or projector ("beamer") can display everything as an image. Graphics cards are either as a PC expansion card (via the bus systems AGP or PCI , formerly PCI , ISA or VLB ) to the main board are connected or in the chip contained on the motherboard. In the meantime, the integration has progressed so far that components for graphics output are already available in the main processors (at Intel since the Core iX generation , at AMD in the Fusion program).
The most important components of modern graphics cards are: GPU , graphics memory , RAMDAC and the connections for external devices (e.g. for the monitor or graphics tablet). The GPU should not be confused with the graphics card as a whole, although “GPU” is often used as a reference for the graphics card.
The graphics card principle was used in series production for the first time in the Apple II microcomputer , the graphics capabilities of which were integrated on the motherboard and could be improved by purchasing additional plug-in cards. (" PAL color card", "80 character card"). This came on the market in 1977.
The first IBM PC came onto the market in 1981 with a card that only enabled text to be displayed in one color (MDA = Monochrome Display Adapter ). Hercules company offered a better card in 1982, the Hercules Graphics Card .
By 1989, the types of color graphics cards that IBM brought onto the market became the standard:
- 1982 the CGA card (popular from 1984)
- 1984 the EGA card
- 1987 with the PS / 2 models the MCGA and VGA cards
Even today, the VGA mode (640 × 480 dots in 16 colors) is the "emergency mode" for all PCs; Only up to this mode can the hardware of all current PC graphics cards be addressed by the software in a uniform manner.
However, VGA was not the last graphics card standard. The Video Electronics Standards Association (VESA) established a standard for video modes up to a resolution of 1280 × 1024 pixels in 16-bit color depth, which every PC graphics card can handle today.
Until around 1990, graphics cards were limited to converting the content of the graphics memory into output signals for the monitor using a so-called RAMDAC module. The programmer could essentially only use the text mode and set individual pixels to a certain color in the graphics mode . That was the first generation of graphics cards. Two more followed:
Acceleration through software
From 1991 the graphics cards were further developed into independent small computing units with their own GPU (Graphics Processing Unit), a so-called graphics or pixel engine or German graphics processor , with which you could not only set individual pixels , but also commands to draw lines and Filling of areas could send ( Windows accelerator ). Above all, these functions accelerated the movement of the windows (Windows) of the graphical user interface. The concept of additional functionality has been continued over time. For example, since 1995 functions to accelerate video playback (e.g. in AVI format ) and decoding of compressed video data (e.g. MPEG ) have been introduced ( video acceleration ). These functions were previously offered on separate plug-in cards .
After the mid-1990s with Doom the big boom of 3D games began, soon arrived from 3dfx the first practical 3D accelerator , the Voodoo Graphics - Chipset . A program can specify the geometric figures in the form of polygons and the textures with which the areas of the polygons are to be filled ( rendering ) to a 3D accelerator in a three-dimensional space . In the early 3D games, the CPU had to take over this rather simple but computationally intensive task ; Now it could be delegated to the graphics card, which led to a massive increase in the performance of 3D games (better image resolution, much more realistic images).
While the first-generation 3D accelerators were still installed on their own plug-in cards through which the graphics signal from the 2D graphics card installed in the system was looped through, solutions that combined 2D and 3D functionality on the same card soon established themselves.
In order to offer even more 3D performance, two or more 3D graphics cards or processors are connected in parallel with multi-GPU technology (see also SLI and AMD CrossFireX ) in order to be able to calculate even more graphics elements per period. However, this technology requires high costs and energy.
Structure and functionality
Hardware interfaces to the system
The best-known hardware interfaces for graphics cards are PCI , AGP and PCI Express ; ISA or VESA Local Bus were also common in the past . These interfaces are either bus systems or direct connections (AGP, PCI Express) that connect the bus controller to the graphics card. Since the specification of the interfaces is mostly carried out by interest groups in which both the controller and the graphics card or graphics chip manufacturers are members, all conforming graphics cards work (ideally) with all conforming controllers. In the past, however, there were various problems with some interfaces that restricted interoperability , for example "AGP Fast Writes" with AGP (on Intel platforms it could increase performance, on AMD platforms it could cause instability) or IRQ problems with PCI (possible crashes, freezes or performance drops, mostly causes bad or faulty implementation of the interface).
In other platforms than the IBM-compatible computers there were graphics cards for the Apple II slots , later in the first Macs for nuBus (later PCI and successor as with PC), for Amigas for their Zorro bus and also maps of Europe for systems based on the latter.
The graphics memory is used to store the data processed in the graphics processor (GPU) and as an image memory (" frame buffer "): These are digital images that are later displayed on the computer screen or with the projector .
The size of the graphics memory determined the maximum color depth and image resolution . The required memory for a desired resolution and color depth could easily be calculated by the user: If, for example, the resolution 1600 × 1200 with a color depth of 24 bits ( true color ) is desired, the number of image points ( pixels ) of this resolution ( 1600 horizontal × 1200 vertical = 1,920,000 pixels in total). The color depth “24 bit” means that 24 bit color information is available for each of these pixels. Thus, the number of pixels is multiplied by the color depth (1,920,000 × 24 bits = 46,080,000 bits). Now only the conversion into bytes is necessary. Since a byte consists of 8 bits, the number is divided by 8 (46,080,000 bits ÷ 8 = 5,760,000 bytes). Since graphics cards used to be usually delivered with 4 or 8 MB graphics memory, a graphics card with at least 8 MB graphics memory would have been required for the desired setting.
Today only graphics cards are built with much more memory than would be necessary for pure image storage. When rendering three-dimensional graphics, the data of the objects, such as size, shape and position, as well as the textures that are placed on the surface of the objects are saved here in addition to the framebuffer . In particular, the ever higher resolution textures have caused a sharp increase in memory size in current graphics cards. The memory size of current graphics cards is already in the three to five-digit megabyte range (512 MB , 1024 MB, 2048 MB, 3072 MB, 4096 MB, 6144 MB, 8192 MB, 12288 MB), 512 MB and less have become rare. The upper limit for gaming graphics cards was 12 GB at the beginning of 2016, whereas professional graphics cards can be equipped with up to 32 GB of graphics memory.
In onboard solutions , the main memory of the system is usually used as graphics memory, which is referred to as shared memory . Access takes place via the respective bus system and is therefore slower than directly connected storage.
Graphics processor (GPU)
The graphics processor is used to calculate the screen output. The first 3D accelerators hit the market in the mid-1990s . These graphics processors were able to independently perform some effects and triangle-based algorithms (such as Z-buffers , texture mapping ) and anti-aliasing . Such plug-in cards (e.g. 3dfx Voodoo Graphics ), which had to be installed in addition, gave the computer games sector a boost in development.
Today, GPUs are superior to CPUs in terms of computing power because of their specialization in graphics calculations. The number of transistors in the graphics processor from Nvidia (Geforce 8800GTS 512, 754 million) and that of an Intel model ( Core 2 Extreme QX9650 , 820 million) serve as a comparison . The difference becomes clear when you consider that over half of the CPU chip area is used for the 2 × 6 MB cache. The development of the integration density of graphics processors has even exceeded Moore's law with an annual factor of 2.4 .
As with the main processors of the computers in which the graphics cards are built, the GPUs on the graphics cards are often the subject of overclocking modifications to increase performance.
The computing capacity that is available on such graphics cards (see under GPGPU ) has already led to the fact that several graphics cards are built into one computer just to achieve maximum computing power. Such systems are sometimes combined in large numbers to form supercomputers .
Due to the high thermal power loss due to the increasing complexity of graphics processors and, in some cases, the graphics memory, complex cooling solutions similar to those for processor coolers are necessary. With a graphics processor (Nvidia Titan RTX), graphics cards consume up to 280 watts ( TDP ) of power, all of which has to be dissipated as thermal energy. There are several approaches to this:
- Active air cooling - the thermal energy is released via a heat sink to the ambient air, which is circulated by fans . This is the simplest and cheapest way to dissipate large amounts of heat, but it also causes noise.
- Passive air cooling - the thermal energy is released into the ambient air by convection through a heat sink . This is only possible with low power or with very large heat sinks. Often, large heat sinks are attached to both sides of the graphics card, which are connected to a heat pipe . The advantages are the lower power consumption (see also: Green IT ) and the elimination of fan noise, the disadvantages are the difficult installation and the higher mechanical stress on the slot due to the sometimes heavy, very large heat sinks.
- Water cooling - just as water cooling is sometimes used for CPUs , the graphics card can also be integrated into one (or the same) circuit. The thermal energy is then transferred to the water in the circuit and from there to the ambient air via a radiator. This enables large amounts of heat to be transported, but it is also the most complex and expensive cooling solution. There are also some cards with pre-installed water cooling with the advantage that the guarantee is retained. Some dealers also offer already assembled water coolers and then offer a guarantee of their own accord, although the manufacturer's guarantee would end with the installation of a different cooler. The advantage of water cooling in GPUs is that they are larger than CPUs and there are also fewer thermal hotspots in the chip, so the large amount of heat can be released more evenly.
Due to the required surface area of the heat sink, the designs of the air cooling in particular are often significantly larger than the specifications of the slot allow (see Fig. Right). For this reason, the adjacent slots on the mainboard often cannot be used.
The RAMDAC (Random Access Memory Digital / Analog Converter) is a chip that is responsible for converting digital (video memory) into analog image signals (monitor). It controls the signal outputs. It can also be integrated in the graphics processor.
External signal outputs
- DisplayPort is a relatively new connection standard for video and audio signals. It is compatible with VGA, DVI and HDMI 1.4 and supports the copy protection methods HDCP and DPCP (DisplayPort Content Protection). In 2014, the spread was already advanced. DisplayPort has already established itself in the professional environment, especially for Apple hardware.
- DVI-Out / Mini-DVI
- The DVI output delivers a digital signal and thus the best possible image quality on screens with a DVI input. Most of today's graphics cards are equipped with a DVI-I connection ( i for integrated ) and thus also provide an analog RGB image signal. Thus, with a passive adapter (usually included), screens with an analog D-Sub input can be connected, but the picture quality then largely corresponds to that of the D-Sub output. There are also the variants DVI-D with exclusively digital signal lines and DVI-A with exclusively analog signal lines. With DVI-D, a distinction must be made between (single-link) DVI and dual-link DVI; the latter contains twice as many data lines and can therefore deliver a larger bandwidth. This is necessary for resolutions greater than WUXGA (1920 × 1200) in order to guarantee a refresh rate of at least 60 Hertz despite the larger amount of data per image. As with VGA, there is also a mini-DVI variant for notebooks without space for a full socket.
- HDMI out
- Graphic cards with HDMI (High Definition Multimedia Interface) output have also been available since 2007. Here the video signal is also output digitally and, if necessary, encrypted with HDCP . DVI-D signals can also be transmitted via HDMI, making DVI devices compatible with HDMI. Support for HDCP is optional with DVI, which is why not all devices can reproduce signals protected in this way. The transmission of audio signals is only possible via HDMI connections.
- Low Force Helix 60 (LFH60) is a 60-pin connector. The contacts are arranged in four rows of 15 pins. LFH60 is used in network technology and PC technology. The socket is called LFH60-F (female) and the plug is called LFH60-M (male). The connector is used by some manufacturers (e.g. Matrox ) to connect a breakout cable for two DVI connectors.
- VGA-Out / Mini-VGA
- An analog RGB signal is provided at a 15-pin D-Sub socket . In tight spaces, the output is also designed as a mini-VGA (e.g. on the Apple iBook ). CRT monitors (tube monitors), projectors or flat screens are connected via a VGA cable with a corresponding connector .
Additional signal outputs and inputs are implemented differently depending on the card. Corresponding sockets ( Cinch , S-Video , LFH60 ) are sometimes available directly on the slot bracket. Mainly for reasons of space, manufacturers also provide an indirect connection via adapter cables or cable whips. Then there is a socket directly on the graphics card, e.g. B. from the Mini-DIN family, whose wiring is not standardized and which often has the general name VIVO (for video-in-video-out). A manufacturer-specific breakout cable is connected here, which then provides further connection options.
- HDTV video data is output in analog YPbPr color-coded via the component output (three cinch sockets) . As described above, the output is usually not implemented directly on the graphics card.
- (also video-in ) In addition to an output, some cards also have a TV input for digitizing external analog video sources. Since there are hardly any situations in which TV-out and TV-in are needed at the same time, and because of space and cost reasons, TV-in and TV-out are often implemented in one socket (video-in-video-out). In this case, it is not possible to use TV-in and TV-out at the same time.
- (also video-out ) The TV output (TV-out), designed as a cinch or S-video socket, can be connected to a television or projector. You can work with several screens (PC screen + television). However, the signal quality of the connection is usually not very high, since it is an analog composite or S-video signal and most cards do not have the necessary circuitry to get the best possible out of these signal types. Some older Matrox cards (such as the G400 MAX) are positive exceptions.
Designs and areas of application
There are basically four types of graphics solutions:
Solutions on the motherboard ("onboard")
With these Integrated Graphics Processor , IGP for short , the functionality of the graphics card is integrated in the chipset of the motherboard or in the processor (e.g. Intel Core i5). IGPs offer all 2D functions, but mostly only slow or outdated 3D functionality and are therefore mainly used in areas with fewer graphics requirements. In the meantime, however, the graphics solutions from AMD and Intel are already powerful enough to play current games with at least low details. A lot has also happened in terms of the supported features. Both manufacturers currently support DirectX in version 11 as well as current standards such as OpenCL or OpenGL . Because of their low power consumption , they are also often used in notebooks . The low power consumption is also a motive for use in embedded PCs ; In the case of particularly critical applications, such as in medicine, there is the additional advantage that the failure source of the plug-in contacts between the main board and graphics card is eliminated. Particularly compact or inexpensive devices do not have their own graphics memory and instead use the computer's main memory (see Unified Memory Architecture , Shared Memory ), which, however, has a negative effect on performance .
These are full-fledged graphics cards that pay little attention to the 3D functions, but rather that should primarily deliver a sharp and high-contrast image. There are also variants with additional 3D functions, especially for CAD applications.
Gaming graphics cards
These graphics cards are available in various price classes from around € 25 to € 2700, with the expensive cards reflecting what is technically feasible in the area of 3D display. When it comes to game cards, it is mainly AMD ( AMD Radeon series) and Nvidia ( Geforce series) that compete with each other. There are also providers such as S3 Graphics , Matrox (was one of the pioneers of 3D game graphics cards, but was pushed back into the professional market by powerful competition) and XGI Technology , but these only play a subordinate role and are mostly used in office PCs Find.
Since most games are developed for Microsoft's Direct3D interface (part of the Windows system component DirectX ), game graphics cards are optimized for maximum performance with this system. Graphics cards that offer full hardware support for the current DirectX version can calculate practically all technically feasible 3D rendering functions in real time. Some game manufacturers rely on OpenGL , especially id Software .
Version 10 of DirectX has been on the market since 2006, but only works in conjunction with the Microsoft Windows Vista and Windows 7 operating systems. DirectX 10 is supported by Nvidia from the Nvidia GeForce 8 series and upwards, and by AMD by the cards of the ATI Radeon HD 2000 series and upwards. Cards from the ATI Radeon HD 3000 series even support the successor version DirectX 10.1, which is supplied with Service Pack 1 for Windows Vista and brings only minor innovations (used in only a few games, such as the aerial combat game HAWX or Bethesda's Skyrim ). DirectX 10 increases many restrictions in shader programming and is said to have a lower overhead than DirectX 9, which is said to make the execution of Direct3D commands faster. The disadvantage is that since the introduction of DirectX 10 only a few games have been optimized for DirectX 10 (most prominent example: Crysis ), since the commercial sale of Windows Vista did not begin until January 30, 2007 and the use of the new effects of DirectX 10 was one require enormous computing power and consequently only work satisfactorily on high-end graphics cards. Many new games often only support DirectX 9 and sometimes DirectX 11 in parallel, so DirectX 10 is of little importance.
Version 11 of DirectX has been available since the end of 2009. This version is supported by cards from ATI (or AMD) from the “HD5000” series and from the “GTX 400” series from Nvidia. The start of DirectX 11 went better than that of DirectX 10, since with the introduction of Windows 7 and thus DirectX 11 there was already a game with DirectX 11 ( BattleForge ) and others quickly followed. However, the games all still support DirectX 9, which means that they can also run on Windows XP .
In 2013, AMD also introduced the AMD Mantle API , which so far can only be used with AMD graphics cards from the ATI Radeon HD 7000 series . Mantle offers both better utilization of performance for multi-core processors and less development effort than Direct3D. Support for Mantle is currently limited to a few games. The Frostbite 3 engine (see Frostbite Engine ) fully supports Mantle, so games like Battlefield 4 , Dragon Age: Inquisition or Star Wars: Battlefront can be played with Mantle. In the future, support for Nvidia graphics cards and Linux operating systems is quite conceivable, but not yet announced by AMD.
State of the art
As of 2019, the current top models are Nvidia with the Geforce RTX 2080 Ti based on the "TU102-300" chip of the "Turing" architecture manufactured in 12nm . In December, the "Titan V" based on the "GV100" of the current "Volta" architecture manufactured in 12 nm was replaced by the Nvidia Titan RTX with the current "Turing" architecture. This entails a radical change in the VRAM, since Nvidia uses 24 GB of "GDDR6" for the "Titan RTX" and therefore no longer uses "GDDR5" or "GDDR5X" .
In May 2016 Nvidia presented the 16-nm models Geforce GTX 1080 and Geforce GTX 1070 with the "Pascal" architecture. The use of the new memory variant GDDR5X should give the cards a higher bandwidth.
AMD , on the other hand, competes with the Radeon R9 Fury X with the Fiji XT chip, which is also manufactured using the 28 nm manufacturing process. The Fury X is equipped with 4 GB of VRAM, but in contrast to the GDDR5-VRAM of Nvidia cards, it uses so-called high bandwidth memory , which is clearly achieved by stacking the memory modules and connecting each stack via two 128-bit channels larger bandwidth is connected than normal GDDR5 memory.
On June 29, 2016, AMD released the mid-range graphics card Radeon RX 480 with the code name "Polaris". The GPU has a boost clock of 1266 MHz and a memory clock of 4000 MHz (effectively 8000 MHz). The Radeon RX 480 has an 8 GB GDDR5 graphics memory. The direct competitor of the RX 480 is the GeForce GTX 1060.
In the games with 4K resolution, the Radeon R9 Fury X performed better than the GeForce cards in the first benchmarks. However, the standard water cooler of the Fury X often produced annoying noises in the first series, which resulted in numerous complaints. AMD promised a new revision of the card and delivered an improved version of the water cooler.
These are primarily graphics cards for CAD and GIS applications. The cards offer special functions required for CAD / GIS that can only be emulated on "normal" graphics cards and can therefore be used much more slowly. Since the last special chip supplier, 3DLabs, ceased business in 2006, only AMD (under the brand name ATI ) and Nvidia have offered solutions for the OpenGL workstation segment. The two companies use derivatives of their gaming graphics card chips. These are then optimized with a modified ROM and driver for the 2D representation of OpenGL and no longer for the 3D representation of DirectX and OpenGL. The drivers of these graphics cards support the drawing of several million smooth lines and user clip plans . Although the hardware between 3D gaming chips and OpenGL chips is only marginally different, professional cards cost considerably more. The reason for this is the optimization of the drivers, the extensive customer service that workstation customers have to offer, and the very expensive SRAM with which some graphics cards are equipped. Furthermore, additional capabilities are often available such as DisplayPort connections for using a higher color gamut or the projection of a large area with several image sources. The product lines are called ATI FireGL at AMD and now AMD FirePro and at Nvidia Quadro .
In addition to the DirectX graphics cards described above, there are special cards that only support OpenGL . These are often used in animation and are completely of no interest to gamers today, as most PC games only support DirectX (but differently on the Macintosh platform). Every DirectX graphics card today also supports OpenGL as standard, but the reverse is not the case.
Since the beginning of the programmable graphics pipeline in 2000, it has been possible to use the computing power of the graphics processors to compute parallelizable computing operations such as B. occur in technical and economic simulations to use. This application is known as GPGPU (General Purpose Computation on Graphics Processing Unit), see also CUDA .
Software graphic interfaces
In order to be able to use graphics cards without having to develop hardware and software for each individually, there are various software graphics interfaces.
The BIOS , which provides important text and graphic output functions, is of particular interest at the basic functional level . a. can be used by text consoles under DOS or Linux . These functions are relatively slow, but work reliably on any graphics card.
In most of today's operating systems there is an abstraction layer between programs and hardware, the so-called device drivers . Without this, programs would have to address the hardware directly, which, due to the differences between graphics cards, would lead to a high degree of specialization and thus high programming effort for supporting many graphics cards. However, since graphics card drivers can also offer very different functions, various graphics APIs have been developed over time to facilitate access to these functions. The best known of these are OpenGL , DirectX (more precisely: DirectDraw, Direct3D) and Quartz , which enable the programmer to display 2D and 3D graphics easily and independently of the graphics card. For DirectX and Quartz, the interfaces do not necessarily require hardware 3D functions of the graphics card, but use them if they are available. Older 3D applications can, in principle, run on computers with integrated graphics or a simple 3D card , but relatively slowly or less visually appealing.
The universal panorama driver was developed for the operating system EComstation , which operates all common graphics cards.
Software problems with graphics cards
Since many graphics cards nowadays allow the smooth viewing of videos using the computer through hardware acceleration and many graphics cards also have a TV-out connection, it is obvious to connect the computer to a television or video recorder . However, some manufacturers prevent the graphics card driver or the graphics card itself from combining both capabilities. So it happens that when playing videos, the entire user interface is visible, but the video itself is not. Under Linux then for example, the works XVideo - Implementation only when the primary display (ie the computer monitor), but not at the TV-out port. This problem can usually be avoided by switching off the hardware acceleration for decoding videos, but the video is then often no longer smooth.
It is believed that such restrictions are built in to prevent the user from recording the video by a VCR . In any case, you can read in some of the manuals supplied that products from Macrovision (a company known for a copy protection process ) have been integrated into the graphics card .
A specific case is the fglrx driver from AMD, which (currently) does not support the hardware-supported playback of videos at the TV output.
Another problem was and is the use of multiple VGA-compatible graphics cards, as can be the case in PCI systems. The operating system does not support every free combination, not even graphics cards from the same manufacturer. However, this can sometimes be remedied by updating the ROM on the card.
Graphics chip manufacturers : 3dfx , 3DLabs , AMD , Alliance Semiconductor , ARK Logic , ArtX , ATI Technologies , Avance Logic , Bitboys Oy , Chips & Technologies , Cirrus Logic , Intel , Matrox , NeoMagic , Number Nine , Nvidia , Oak Technology , Rendition , S3 Graphics , S3 Inc. , SiS , Trident , Tseng Labs , Western Digital , XGI .
Graphics card manufacturers : Abit , Albatron , AOpen , Asus , ATI Technologies , AXLE3D, Club 3D , Connect3D, Creative Labs / 3DLabs , DFI , Diamond Multimedia , ELSA Technology , EVGA , Elitegroup , Gainward , Galaxy Microsystems Ltd., KFA2, GeCube, Genoa, Gigabyte , Hercules Graphics, HIS, Inno3D, Leadtek , Matrox , MSI , miro, Number Nine , Orchid Technologies, Palit Microsystems Ltd., Paradise Systems, PixelView, PNY , PowerColor, Quantum3D, Sapphire , Sigma, Sparkle, SPEA, STB Systems, TerraTec , VideoLogic , Video Seven, XFX, XpertVision, Zotac .
Today only AMD / ATI Technologies , Nvidia , Matrox and S3 Graphics are still on the market as graphics chip manufacturers, as well as AMD, Intel , Nvidia, SiS and VIA Technologies as manufacturers of integrated graphics solutions.
- Jens Dünow: Pixel art. Heise Verlag, January 15, 1999, accessed on January 22, 2017 (explanation of how a 3D graphics card works).
- Archived copy ( memento of the original from July 1, 2011 in the Internet Archive ) Info: The archive link was inserted automatically and has not yet been checked. Please check the original and archive link according to the instructions and then remove this notice.
- c't 2009/2, p. 126, construction workers - graphics cards for professional CAD and 3D applications
- https://www.heise.de/newsticker/meldung/Monster-Grafikkarte-Nvidia-Titan-V-3100-Euro-fuer-5120-Kerne-und-12-GByte-Speicher-3914138.html. Accessed March 31, 2018 .
- http://grafikkartenvergleiche.de/nvidia-geforce-gtx-1070/. Retrieved June 20, 2016 .
- Kevin Lee: AMD Radeon VII release date, news and features. techradar, January 30, 2019, accessed February 5, 2019 .
- Sapphire Radeon RX 480. In: GrafikkartenVergleiche.de. Retrieved July 1, 2016 .