The Core Chip

The core chip is surely the most important element present in any video card. Without it, the card wouldn't be able to do anything! Until a few years ago it was a very simple element, as it only had to manage textual information and - after the launch of the first GUIs, like Microsoft Windows and the X Server - to display graphical outputs, without performing any operation. But nowadays, as 3D video games invaded the market and always more complex GUIs were launched, at the high resolutions allowed by 17'' monitors, it has become one of the most complex circuits in the PC - sometimes even more complex than the system CPU. This because today 3D acceleration is often supported via hardware, since this approach allows great performance boosts, and there's only one way to do this: adding new circuits, able to perform more and more complex operations on 3D scenes and video images in general. Moreover, also 2D acceleration has improved greatly in recent years, and this adds complexity to complexity. The result is that, while a modern x86 CPU can have from 30,000,000 to 50,000,000 transistors, very recent high-end video cards can have more than 100,000,000 transistor - more than the double! Since they operate at high clock frequencies (up to 300-400 MHz), there is a big power consumption: in fact, in the last 5 years, passive heatsinks began to appear on video cards, and after a while even fans, necessary to dissipate the heat produced by the device. In some cases the power requirements are so high that the AGP slot is not able to feed the board which needs an external connection, direcly tied to the main power supply.

Why are video cars so complex? Which are the most transistor-hungry elements? Here is a list of the most important components of the video core, which can give you an idea of how it's made:


The rendering pipeline


It is the main elaboration unit in a graphical chip; there can be more than one pipelines operating in parallel and, depending on the clock frequency, they can generate a given number of pixel per second. It is possible to obtain the theoretical fill-rate of a graphical chip multiplying the clock frequency by the number of pipelines present.
top

The RAMDAC

The screen image information stored in the video memory (RAM) is of course digital, because computers operate on digital numbers. Every value is stored as sets of ones and zeros; in the case of video data, the patterns of ones and zeros control the color and intensity of every pixel (dot) on the screen.
CRT monitors, however, doesn't use digital information - they are analog devices. In order to display the image on the screen, the information in video memory must be converted to analog signals and sent to the monitor. The device that does this is called the RAMDAC, which stands for Random Access Memory Digital-Analog Converter. "RAM" of course is a well-known acronym that refers to memory.
Many times per second, the RAMDAC reads the contents of video memory, converts the information and sends it over the video cable to the monitor. The type and speed of the RAMDAC has a direct impact on the quality of the screen image, how often the screen can be refreshed per second, and the maximum resolution and number of colors that you can display.
Refresh rate is measured in Hertz (Hz), a unit of frequency. Support for a given refresh rate requires two things: a video card capable of producing the video images that many times per second, and a monitor capable of handling and displaying that many signals per second. The refresh rates are somewhat standardized; common values are 56, 60, 65, 70, 72, 75, 80, 85, 90, 95, 100, 110 and 120 Hz. This is done to increase the chance of compatibility between video cards and monitors.

Note: Do not confuse the refresh rate with the term "frame rate", often used for games. The frame rate of a program refers to how many times per second the graphics engine can calculate a new image and put it into the video memory. The refresh rate is how often the contents of video memory are sent to the monitor. Frame rate is much more a function of the type of software being used and how well it works with the acceleration capabilities of the video card. It has nothing at all to do with the monitor.

The refresh rate is important because it directly impacts the viewability of the screen image. Refresh rates that are too low cause annoying flicker that can be distracting to the viewer and can cause fatigue and eye strain. The refresh rate necessary to avoid this varies with the individual, because it is based on the eye's ability to notice the repainting of the image many times per second. However we can say that:

Note that this also depends on the size of the monitor. Flicker is easier to see on a larger monitor than on a small one for two reasons: first, you're just looking at that much more screen; second, you are seeing much more screen using your peripheral vision, which is much more sensitive to noticing flicker.
Also note that going to very high refresh rates generally has no positive impact; most people cannot even tell the difference between a video system running at 120 Hz and one running at 100 Hz. In fact, going to too-high a resolution can be counter-productive. The reason is that the higher the refresh rate, the faster the electron guns have to switch between colors for adjacent pixels. At very high refresh rates there can theoretically be a reduction in the contrast in the displayed image. So, the best solution often is running only at as high a refresh rate as you need to eliminate flicker. 85 Hz is generally more than enough for virtually everyone.
The refresh rate is related directly to the resolution of the image - higher resolution images generally have lower refresh rates. Higher refresh rates require the RAMDAC to generate the video images more times per second. The ability of the RAMDAC to do this depends on several variables: Refresh rates are normally specified for non-interlaced operation, since that is what modern video systems typically use. Some very old monitors can only display some of the higher resolutions when using interlacing. Interlacing allows the refresh rate to be double what it normally would be, by displaying alternating lines on each refresh. In essence, half the screen is redrawn at a time. Interlaced operation is normally done at 87 Hz (really 43.5 Hz because of the interlacing) and hence produces flicker that is noticeable by most people.
The table below shows the relationship between screen resolution, refresh rate and the amount of data the RAMDAC must process. The numbers in the table are in MHz, representing how many millions of pixels per second the RAMDAC must output to support a given resolution at a given refresh rate. Many video cards rate their RAMDAC in MHz and you can use this table to see if the card is likely to support the resolution and refresh rate you need. This table includes a 1.32 conversion factor to take into account retrace times (the time that the electron guns are in non-visible areas of the monitor):

Resolution

43.5 Hz (87 Interlaced)

60 Hz

72 Hz

80 Hz

85 Hz

90 Hz

100 Hz

320x200

3.7

5.1

6.1

6.8

7.2

7.6

8.4

640x480

17.6

24.3

29.2

32.4

34.5

36.5

40.6

800x600

27.6

38.0

45.6

50.7

53.9

57.0

63.4

1024x768

45.2

62.3

74.7

83.0

88.2

93.4

103.8

1280x1024

75.3

103.8

124.6

138.4

147.1

155.7

173.0

1600x1200

110.2

152.1

182.5

202.8

215.4

228.1

253.4

Note: Don't forget that at higher resolutions and color depths, video memory bandwidth becomes a limiting factor. The speed of the RAMDAC doesn't matter if the video memory can't provide the necessary data fast enough.
top

The Memory Controller

This part of the video chip has the important function of controlling the video RAM and for this reason it is used in all I/O memory related operations. It is a critical component in modern video cards and one of the most difficult parts to project: since modern boards operate at incredibly high clock frequencies, they need very fast memories to talk with in order to maintain fully operating they data-hungry pipelines. If memories are too slow, the GPU will waste a lot of time waiting for new data to process, so its potential will not be used. In fact, nowadays the performance of a video card are often "bandwidth limited", meaning that rising the memory clock will improve dramatically the board's performance, while rising the core clock will leave them unaltered. For this reason, video RAM is always very fast - and often the most expensive part of the board - and the video memory bus is as wider as the system's one, if not wider: in cards produced about 5 years ago it often was 64 bits-wide, while in today's cards it is 128 or even 256 bits-wide, in high end devices.
top

The Transform & Lighting Unit

This unit is present in all recent 3D accelerated video cards. Its purpose is to help the system CPU in performing the 3D -> 2D conversion (see Transform & Lighting for more information): in fact this process needs a huge quantity of floating point operations and, although they often are very simple, they are so many that if made in software (i.e. by the CPU) they slow down very much the program. To solve this problem, some years ago video card began to implement this unit, which is specifically designed to perform these operations instead of the main processor. This allow a great increase in performance, so almost all recent video cards have this unit, although not all programs can take advantage of it: in fact the T&L unit can work only if supported by the program itself and for this reason it needs specifically written programs; otherwise, it is totally useless. However, it is a so important features that video chips implementing it are no more called simply "video chips" but GPU, Graphic Processing Unit.
top