Graphics cards Buying Guide

Buyers' Guide to Graphics cards Contents

• 2D or 3D: that is the question
• PCIe or AGP
• Graphics card models
• ATI cards
• ATI variations
• Nvidia cards
• Nvidia variations
• What the specs mean
• Other features
• Other components
• Power usage
• Integrated graphics
• Getting the most from your card
• References

-----------------------------------------------------------------------------------

A vital component of any PC, the graphics card is a key performance component, especially in respect to 3D games. Put simply, it is an expansion card in your PC that processes image data and outputs it to your monitor.

graphics cards act as processing powerhouses, offloading from your CPU much of the hard work of calculating how scenes look, particularly in 3D graphics. Taking a 3D scene and rendering it to the screen takes an incredible amount of processing power. So much processing power, in fact, that the fastest graphics processors often have more transistors than mainstream CPUs, require fans to cool them and need direct links to power supplies.

A graphics card has multiple components: a graphics processor, or GPU; memory for graphics operations; a RAMDAC (more on RAMDAC later) for display output, and potentially other ancillary components to TV output and capture, SLI and the like.

2D or 3D: that is the question

A graphics card should be purchased in accordance with your needs. The simple rule is that all currently available graphics cards are up to snuff for 2D operations. If 2D is as far as you want to go, then you should look for a low-cost solution, perhaps even go for integrated graphics.

It's 3D graphics performance that really separates the wheat from the chaff. The performance of the graphics card will directly influence both the frame rate and image quality of 3D programs and games. There are huge differences between the low and high-end cards in this respect.

Detailing how 3D graphics works is an entire guide in itself, but the digest version is that there are two main tasks to be completed. The first is lighting and geometry, the second is rendering.

For the first, think of the 'wireframe' 3D images you've seen in documentaries and the like. These wireframes define the shape of the objects in the scene. This is the geometry of an image, and has to be calculated based on what the viewer can and can't see of the objects, the positioning, camera angles and the like. Lighting - figuring out where the light sources are and what effect they have on the objects also happens in this phase.

The second phase of drawing a 3D scene is the rendering -- that is, the painting of the wireframe. Textures are applied to surfaces, and modified according to light and other factors.

At one time, 3D graphics cards did not do any geometry processing, leaving that entirely to the computer's main CPU. Since the introduction of the Nvidia GeForce however, consumer graphics cards have possessed considerable geometry processing power - it was with the introduction of this chip that we first saw the term "graphics processing unit" (GPU) appear.

The GPU is a small microchip present on most modern graphics cards. They are specially geared towards processing complex, graphical algorithms that previously fell to the CPU. Thus, they both take the strain off the main CPU, and process information faster thanks to specialised design.

In addition to the raw ability to draw the wireframes, animate and paint them 60 or more times per second, graphics cards also have a large set of other 3D features designed to make scenes look good. Graphics cards can be differentiated by their ability to calculate shadows quickly, eliminate pixilation, render distant objects at low resolution, modify surface textures on the fly and a host of other capabilities.

PCIe or AGP?

Graphics cards will come in one of two varieties: AGP or PCI Express (PCIe). AGP and PCIe are types of expansion slots on your PC's motherboard where you plug in the graphics cards. Although there a few exceptions, your motherboard will only support one of the two types, so you need to be sure which it is before you go and buy a graphics card.

AGP is the older standard, and it comes in various flavours up to 8x (this is the most common one in motherboards less than two years old). In an AGP motherboard, there will be only one AGP slot. An AGP motherboard will also have PCI slots, but don't confuse those with PCIe slots. You can't put a PCIe graphics card in a PCI slot.

PCIe has recently supplanted AGP, and most new motherboards and systems use it. PCIe slots also come in different flavours: 1x, 4x, 8x, 16x and 32x. A motherboard with PCIe will have a mix of the types. For instance, a motherboard might have one 16x PCIe slot and four 1x PCIe slots. Most current PCIe graphics cards require 16x slots.

PCIe is considerably better than AGP, although that fact may not be realised in the current generation of graphics cards. A card in a 16x PCIe slot can communicate with the PC's CPU and memory at twice the speed of one in an 8x AGP slot.

In addition to speed, PCIe has another advantage - the ability to put more than one graphics card in the PC. A motherboard can only have one AGP slot, but more than one 16x PCIe slot is possible (though not common). Some Nvidia graphics cards are designed to work together to render a scene, effectively splitting the load if you have more than one card in the PC. This configuration is called the scalable link interface (SLI).

Motherboards with two PCIe slots (such as those based on the Nvidia nForce4 SLI chipset) will also allow you to install two different graphics cards in a non-SLI configuration (that is, the cards are not connected to each other to boost rendering performance) so that you can use up to four monitors on one PC!

This type of functionality is suitable for those of you who work with large image or audio editing applications and require as much screen real estate space as possible. Conveniently, to implement this type of multi-monitor support, you don not have to use identical cards or even cards of the same make. A card based on an Nvidia chipset will work in conjunction with an ATI chipset-based card. You will need to use an operating system that supports multiple monitors in order to be able to enable this feature though. Windows XP Pro or Home does the job very well.

Graphics card models

Although there are a large number of graphics card vendors, there are actually only two major manufacturers of chips for graphics cards. ATI and Nvidia dominate the business, and nearly every graphics card you buy will use chips from one of these companies.

Usually, the chip in use is detailed in the product name or specifications. Cards using the same graphics chip are fairly homogenous: one graphics card using the Nvidia GeForce 6600 chip will perform roughly the same as another card using the same chip; that is not to say however, that you should not shop around. Some brands are known for offering more reliable cards that often come with slightly higher clock speeds out of the box, as well as increased, manufacturer guaranteed overclocking (see our overclocking section later) for a slightly higher price. Other factors that should come into your decision include the amount of memory a card has (128MB, 256MB etc) and its features, such as TV-Out and dual-screen support.

Unfortunately, figuring out which is the best chip for you can be challenging. The graphics chip business has become a nightmarish monstrosity involving a seeming infinite number of chip models and variations. For instance, in the Nvidia "5000 series" alone there are more than 14 discreet chips, each with different clock speeds, memory support and multitudinous other specifications. Factor in variations in memory size, interface, output support and other features and the task of choosing a graphics card becomes almost overwhelming for a casual buyer.

Nvidia are rectifying this with their next series of cards. The 7800 GTX comes in a single model. No complex variations in numbers or builds. There will be differences based on a number of the factors mentioned above, but the process of graphics card selection will grow much simpler over the next six months.

If you're not much interested in the Herculean task of unravelling the performance specifications of the various chips, we suggest the best strategy is to choose your graphics cards based on price.

The market price of graphics cards (especially those using chips from Nvidia and ATI) is very much dependant on the performance they offer. As a general rule, one $300 graphics card is going to perform roughly the same as another $300 graphics card. This is generally true across model types and across brands. The ATI and Nvidia lines mirror each other quite closely on a price/performance scale - one of the consequences of their duopoly on graphics chips.

An exception to this rule are the now rare Matrox graphics cards, which tend to provide an host of extra features at a slightly higher cost (the Matrox Parhelia can drive up to 3 monitors on its own.)

At the present time, there are broadly three price categories of graphics cards:

• The lowest-cost cards (sub $150): Some can even be purchased for less than $70. These cards are not going to perform at all well on recent 3D games such as Doom3 and Half Life 2. However, they will be perfectly fine for most non-gaming applications, such as Web surfing and office applications.
• Mid-range cards ($200-$400): For around $300 you can get a card that performs pretty well in most games, allowing you to switch on most, if not all, of the graphics features (shadows, anti-aliasing and the like) in most 3D games and still get a playable frame rate.
• The elite cards (over $600): These are premium cards, for those who simply must have the best of everything, no matter what the cost. Much like CPUs, the performance of cards does not scale linearly with price (a $600 card will not be twice as fast as a $300 one, for instance). These cards exist for people who take their 3D games very seriously.

ATI Cards

Radeon 9000 series - a largely outdated series of cards, superseded by the X series. The 9000-series cards are still widely available as low-cost graphics solutions, however. There are a lot of different card models in this series, but the most commonly encountered are the Radeon 9250, 9550, 9600, 9700 and 9800.

Although the cards have largely been supplanted by the X-series of cards, they can still offer decent performance at the top-end of the range. A Radeon 9600 has a comparable rendering speed to a X300 card, and a 9800 Pro can rival the X600 line. Neither can match up in terms of geometry performance, however, and they lack some of the image quality features of the X series. They also lack the X series' support for video compression acceleration.

Radeon X series - currently the mainstream and top-end processors of the ATI line, these PCIe and AGP solutions go head to head with the Nvidia 6 series. At the bottom end of the range you find the sub-$150 X300 cards, while at the top end, the X850 cards can set you back more than $750.

Pound for pound, they're a little better than the 9000 range of ATI cards, although the top-end of the 9000 series will outperform the low end of the X series. The X series does not see a dramatic improvement in the fill rate over the 9000 cards, but the geometry and image quality capabilities are considerably better. For instance, the Radeon X700 Pro has roughly the same fill rate as the Radeon 9800 XT (3.4 gigapixels) but 55 per cent better geometry performance (637.5 million triangles per second, as opposed to 412 million for the 9800 XT).

The X range includes the following base chips:

• X300
• X600
• X700
• X800
• X850

ATI variations

As with Nvidia, each ATI chip comes in multiple flavours. You have, for example, the Radeon 9800, Radeon 9800 Pro, Radeon 9800 SE and Radeon 9800 XT. For the most part, these model numbers indicate variations in the clock speed of the chip and of the memory on the card.

Here's a quick rundown of the variations ATI uses:

SE - a cheaper card that has half the memory bandwidth of the base card. A X300 SE, for instance, has a 64-bit wide connection to the graphics card memory. A normal X300 has 128-bits.
 

Pro - indicates that the card is faster in some respect than the base card. It can mean the card has a faster processor, faster memory, or more pixel pipelines to increase the rendering rate.
 

VIVO - indicates that the card has support for composite or S-Video inputs and outputs. A VIVO card will have TV-out and TV capture, so you can plug a video recorder into it and digitise the video. It does not necessarily mean it has a TV tuner, however.
 

XT - the next step up from the "Pro" cards. It indicates that the card runs at a higher clock speed again than the Pro and Basic cards.
 

XT PE - a step up in performance yet again from the XT cards.
 

So, the order of performance within a given model, from slowest to fastest, is the SE, the basic card, the Pro, the XT and the XT PE.

Nvidia cards

Nvidia GeForce MX series - The oldest series of cards still available for sale, finding them now is rare, as they are surpassed by both the 5000 and 6000 series. They can be found for as little as $30-40 now, and are only suitable for basic computing operations such as word processing and internet browsing.

Nvidia GeForce FX 5000 series - An older but still widely available range of cards, the 5000 series of cards are equivalent to ATI's 9000 series, ranging from very low cost to mid-range cards. They do not have nearly as much geometry processing power as the 6 series, but can hold their own in terms of fill rate.

There are a very large number of variations, ranging from the ultra-cheap GeForce FX 5200 to the very powerful GeForce FX 5950. There are too many to go through in detail here, and most are very difficult to find, but here is the full list of graphics chips in this range available from Nvidia:

• GeForce FX 5200
• GeForce FX 5200 Ultra
• GeForce PCX 5300
• GeForce FX 5500
• GeForce FX 5200
• GeForce PCX 5750
• GeForce FX 5700LE
• GeForce FX 5700
• GeForce FX 5700 Ultra
• GeForce FX 5900
• GeForce PCX 5900
• GeForce FX 5900 XT
• GeForce FX 5900 Ultra
• GeForce FX 5950 Ultra

The performance across the range varies considerably. For example, the still-available FX 5500 makes a rather poor video card by today's standards - but it can be bought very cheaply. Its fill rate power is about 80 per cent that of the GeForce 6200, and its geometry processing power is roughly quarter of the newer 6200 cards.

At the top end of this range, however, the GeForce FX 5950 Ultra is an excellent card, in spite of its age. It has fill rates and vertex processing power roughly equivalent to the GeForce 6600.

GeForce 6 series -- The GeForce 6 series is Nvidia's current mainstream and high-end line, currently comprising of the GeForce 6200, GeForce 6600 and GeForce 6800. Cards using the 6200 chips are difficult to find, although if you can find them they should be in the sub-$200 category of products. The 6600s cost around the $300 mark, while the powerful 6800 cards can cost $600 or more.

To give you an indication of the scale of difference across the range, the GeForce 6600 GT has more than three times the fill rate of the 6200 cards, and its vertex processing power is about 40 per cent greater than that of the 6200. A GeForce 6800 Ultra has more than twice the memory bandwidth of a 6600 GT. It has a fill rate of 6.4 gigatexels (compared to 4 for the 6600 GT) and a vertex processing capacity of 600 million per second (the 6600 GT can process 375 million. We'll talk more about texels later.)

Being the most recent series of graphics cards, the chips used in the GeForce 6 series support many of the latest features required by developers to make scenes look good, including support for the most recent DirectX Shader Model (3.0).

The 6 series is available in both AGP and PCIe cards.

Nvidia variations

Much like the ATI cards, Nvidia has special versions of different chips, clocked at higher speeds or with faster memory. Letters next to the name (such as GeForce 6600 GT) indicate a variation from the standard specifications.

Here's a quick rundown of the variations:

Turbocache - a variation of the GeForce 6200, the Turbocache is available in only the PCIe version of the card, and is designed as a low cost solution that partially uses main memory for graphics operations.
 

TD and TDH - The T indicates the card has TV-out, the D indicates it has DVI out and the H indicates it has Hardware Monitor Function.
 

LE - a trimmed down version of the basic card, using a chip running at a slower clock speed with possibly fewer pixel pipelines.
 

XT - the chip and memory run at slightly slower clock speeds than the "basic" card. XT cards are thus a little slower and a little cheaper than the basic card . This can be confusing, since ATI uses "XT" to indicate a high-end card.
 

GT - in the 6600 and 6800 cards, a "GT" designation means that the card uses GDDR3 memory instead of DDR memory. GDDR3 memory is a much faster type of memory than DDR, running at up to 1050MHz. A GT card will be somewhat faster than a "vanilla" version of a card.
 

Ultra - a considerably faster version of the chip than the vanilla version. The chip and memory run at higher clock speeds than the normal version.
 

Ultra Extreme - faster again than the Ultra.
 

SLI - the card is capable of working with an identical card in an SLI configuration (see below). SLI cards are all PCIe.

What the specs mean

Processor clock speed: Much as with PCs, a higher clock speed for the graphics processor does not automatically mean it runs faster. A 400MHz processor that can do twice as much per clock cycle as a 600MHz processor will outperform it. In terms of rendering speed, the per-clock performance of the graphics processor is indicated by the number of pixels it can process per cycle.

Memory size: This is the amount of memory on the graphics card, to be used exclusively for graphics operations. Nearly all chips support multiple memory configurations, some up to 512MB. The effect of memory size on performance will vary depending on the application. The graphics processor can talk to the graphics card's memory much quicker than it can talk to the PC's memory, so the more memory on board, the less it has to go to main memory for the information it needs to render a scene.

Generally speaking though, if you currently have a 128MB GeForce 6600-based card, for example, and want to upgrade to a 256MB GeForce 6600-based card, you may not notice a performance difference at all during gameplay unless you are playing games with big enough texture sizes that can take advantage of the extra memory, such as Half-Life 2, Doom 3 and Far Cry.

Memory bandwidth: This is the speed at which the graphics processor can talk to the memory on the card. One of the big bottlenecks in 3D performance is the speed at which the PC can deliver information to the graphics processor. Faster memory means that this bottleneck is eliminated, resulting in faster rendering speeds. There are two types of memory used on current graphics cards - DDR and GDDR3. GDDR3 is faster, and better able to keep the graphics processor occupied.

Performance will also be affected by the width of the memory bus. A graphics card with a 128-bit memory bus can haul double the amount of data between the memory chips and the graphics chip than a card with a 64-bit bus. Generally, only mainstream cards have a 64-bit memory bus.

Shader model: DirectX Shader Models give developers a great deal of control over the appearance of the scene on screen, and can be used for various neat effects like complex shadows, reflection, fog and the like. Microsoft has improved the Shader Models over time, giving more power to the developers (and thus greater potential to make scenes look realistic). Shader Model 3.0 is the latest version, found in DirectX 9.0c. At this stage, only Nvidia supports Shader Model 3.0.

Fill rate: This indicates the speed at which the graphics card can "paint" a scene. Each surface in a 3D scene has to have a texture applied. The pixels (or texels) per second rating indicates how many pixels of textures can be painted per second. What is a texel? It is kind of like a 3D pixel. For instance, a flat surface may be painted (textured) with a 640 by 480 image of bricks, in order to make it look like a brick wall. Each pixel in that texture is known as a texel, and each has to be processed to take into account of the distance from and angle to the viewer of the wall. Other textures, such as light maps, can then overlay the original texture.

The fill rate is a product of the clock speed of the processor and the number of pixels it can process per clock cycle. The number of pixels that can be processed per clock cycle is gleaned by the number of pipelines the graphics chipset supports.

Vertices/triangles: Where the fill rate tells you the rendering performance of the graphics chip, it does not cover the performance of the card in geometry calculations. The graphics chips cannot work on curves - they can only process flat surfaces; but if you have enough flat surfaces you can make them look like curves. In a process called tessellation, all objects in a 3D scene are broken down into a set of triangular surfaces - the more triangles, the better the curves and angles of the real world can be represented. A 3D object can be made up of hundreds or even thousands of triangles. This is what game developers are talking about when they talk about the number of polygons in a character - the number of discreet (flat) surfaces in the character model.

Unfortunately, to further confuse matters, ATI and Nvidia talk about geometry processing power using different standards. Nvidia chip specifications refer to vertices per second (that is, the points at the corners of the triangles), while ATI specifications list triangles per second. (You cannot, incidentally, just divide the Nvidia numbers by three to get triangles - triangles next to each other share vertices).

Anti-aliasing: A technique used to make scenes look better by eliminating visual artefacts such as the jagged stepping effect caused by diagonal lines and square pixels. (Open up a paint program and draw a diagonal line, and you'll see what we mean). Anti-aliasing does improve image quality, but usually causes a performance hit. There are multiple levels of anti-aliasing, and some reduce performance more than others.

RAMDAC: This stands for Random Access Memory Digital to Analogue Converter. This is the chip that takes the scene and converts it to the format that your screen uses. The graphics processor creates a final image of a scene - the RAMDAC takes the image and puts it on the VGA or DVI cable for your monitor to display it. RAMDACs have different speeds - usually 350MHz or 400MHz, and a graphics card may have more than one. A faster RAMDAC indicates that the graphics card can support higher output resolutions, and multiple RAMDACs indicates that the card can support multiple displays, possibly in a dual-head configuration.

Other features

Much like washing powder companies, graphics chip vendors are rather inclined to throw technobabble at users about new and unique technologies in their products. CineFX, SmoothVision, HyperZ and a host of other graphics technologies are detailed by Nvidia and ATI in their marketing materials. Some of these technologies do have a significant impact on the performance or visual quality of the display, or on the ability of the card to compress video streams, but others may be just marketing fluff.

Other components

TV-out - you may want the ability, via either a composite (RCA) or S-Video connector to display the output of the graphics cards on a TV screen. TV-output can be a bit iffy, especially on older cards, because of its low resolution, interlaced scan and poor pixel precision. Therefore, you should only use the TV-Out port if you want to watch video from your PC on a TV screen. For more on TV-out, we recommend taking a look at our Media Centre PC Buyer's Guide. ^link^

DVI - a graphics card with DVI has a connector for DVI monitors. This is handy as there is a large range of LCD screens that support DVI. DVI is a type of monitor connector that offers better image quality than the standard VGA connector. graphics cards that have DVI connectors usually ship with a DVI to VGA adapter that allows you to plug in an analogue monitor cable.

Dual-head- now sported by many graphics cards, dual-head is the ability to use two monitors, usually side by side. Your desktop stretches across both monitors. Cards with more than one output often support dual-head, although the heads may not be symmetrical. For example, an ATI card might have both a VGA connector and a DVI connector. You can use them in a dual head configuration if you want to run two monitors side by side.

SLI - this stands for Scalable Link Interface. Currently only supported by Nvidia in some of its GeForce 6600 and 6800 chips, SLI allows you to run two graphics cards side by side in your PC. When processing geometry and rendering a scene, each card takes half the scene, theoretically doubling the performance of the graphics subsystem.

SLI only works on PCIe motherboards, and the motherboard must have at least two 16x PCIe slots. You also obviously need to buy two SLI-supporting graphics cards, and both cards must be identical.

Motherboards that support SLI are also a great investment if you want to use more than two monitors on the same computer. Because you have two graphics card slots, you are free to install any two PCIe graphics cards in these slots (they don't have to be identical or even have the same chipset) to take advantage of up to four monitors, assuming the cards you install have both VGA and DVI connectors (dual-head).

Power Usage

Whilst older graphics cards are capable of drawing all necessary power through the AGP or PCIe slot they sit in, many modern cards have powerful fans and GPUs that actually require a separate power connection. If you are purchasing a high powered video card such as Nvidia's 6800 range, be sure to check the power requirements beforehand. You will need a spare power connector to plug into the card, and a power supply strong enough to cope with the extra wattage.

Integrated graphics

We have not touched on the integrated graphics solutions. Some motherboards using chipsets from Intel, Nvidia and SiS have graphics processing capabilities integrated into the motherboard itself, obviating the need for a graphics card.

Integrated graphics are perfectly fine if all you ever want from your PC is to use productivity applications and surf the Web. They can save you a good deal of money.

They tend to fall down a bit when it comes to 3D graphics, however. They use system memory instead of dedicated graphics memory, and they usually do not perform nearly as well as discreet graphics cards.

Often, a motherboard with integrated graphics will still have an AGP or 16x PCIe slot into which you can insert a graphics card. In this case, inserting a graphics card will often override the integrated graphics capabilities of the motherboard (sometimes on older models you may have to disable it manually through the BIOS. See your motherboard manual for details). These motherboards are good buys - you can run them without a discreet graphics card for a while, but if you find you need to upgrade, you can purchase a graphics card at any time.

Getting the most from your card

Another element that may be important to some buyers is the overclockability of the graphics card. Overclocking refers to pushing a particular piece of hardware, be it RAM, CPU or graphics card, past the clock speeds available out of the box. Most hardware can be overclocked to some degree, with the limiting factor being the temperature of the item.

Depending on the brand of card you buy, the cooling you have in place and the quality of the memory it uses, you may be able to push the card well beyond its standard specifications, with some cards overclocking to speeds past that of the next tier up. Several companies even guarantee their cards to overclock to a certain level, so if you are interested in pushing your PC that little bit further, and can afford to spend a small amount more, be sure to look out for those. You can also shell out a bit of extra money and buy some third party cooling devices for your card, including fans and heatsinks, which may increase the level to which you can overclock, but this will obviously cost more, and may also be louder than any out of the box cooling available.

It is important to be aware that overclocking can damage your hardware, and many manufacturers will not honour the warranty if it is damaged in this way, so overclock at your own risk.
 

References

If you're serious about getting the best graphics cards, benchmarks usually provide a better indication of performance than the specifications ever will. There are a number of sites that regularly run reviews of graphics cards, including comprehensive performance benchmarks. We recommend:

Anandtech

Ace's Hardware

The Tech Report

[H]ard|OCP

Your department store source to buy cheap used and new items online