Graphics cards

At the heart of every graphics card is the graphics processing unit (GPU), and the power and complexity of this chip defines how fast the card will be at playing games, and the effects it can display. Not surprisingly, the more powerful the GPU is, the more expensive the card will be.

This is a very simple concept, but graphics card manufacturers manage to make it appear far more complicated, thanks to their use of pre-overclocked GPUs, third-party coolers, different memory configurations and the addition of seemingly random letters to product names. The situation isn't helped by the rapid rate at which ATi and Nvidia release new GPUs, and the price slashing of older cards that follows. Whether to buy a mid-range card now, or wait until the high-end cards drop into your price range, is perhaps the hardest decision of all.

The Custom PC graphics card buyer's guide is a comprehensive review of 35 PCI-E graphics cards. In fact, we've evaluated the performance of every GPU available on the market today, and even some that are no longer available. These obsolete models, which were highly popular in their time, have been included so that you can determine for sure whether there's still life left in your current card, or if it's time to upgrade.

As there are so many different variants of each card, we've tested only models that run at the official reference speeds, and have the standard amount of memory intended for that GPU.

From this Labs test, you'll discover whether ATi's new flagship, the Radeon HD 2900XT, has the guns to defeat Nvidia's champion, the GeForce 8800 Ultra, which mid-range and budget cards are the best buys and if the time is right to switch to DirectX 10, plus much more. In short, the following pages contain the complete guide to buying a graphics card.

The most meaningful way to test graphics cards is to play modern games and measure the frame rates that the card is able to produce. By testing in this way, the frame rates that we record are representative of actual gameplay.

We test graphics cards by playing three very different games at four different quality settings. Obviously, we don't expect a low-end card to be able to handle high resolutions with plenty of AA and AF, so we judge each card based on what it's designed to do. Therefore, high-end cards are judged mainly on their ability to handle high resolutions, while mid-range cards are judged on how well they can run games at medium resolutions.

However, we expect all modern cards to be able to handle the latest shader-heavy games at maximum detail settings. After all, what's the point in buying a new graphics card if it doesn't allow you to turn up the settings and enjoy the game in the way that the developer intended?

The three games we use are F.E.A.R., Need for Speed: Carbon and S.T.A.L.K.E.R. - Shadow of Chernobyl. In F.E.A.R., we use the benchmark built into the game to record the average and minimum frames rates, although we also play through several different levels at each of the test settings, using FRAPS (www.fraps.com) to measure the frame rate and determine the best playable setting. In Need for Speed: Carbon, we play the third Challenge race with the quality settings set to maximum, while in S.T.A.L.K.E.R., we play through a section of the game involving a frenetic battle with a squad of Russian soldiers.

By looking at these figures, which are listed in the graphs starting on p87, you can clearly see the resolutions and image quality settings at which each graphics card can smoothly run the games. The definition of 'smoothly' in this context is also important, since we consider both the minimum and the average frame rate. The reason for this is that, since the eye can perceive the lowest frame rate, a low minimum will mean that you see the game stutter and jerk, even if the average frame is high. For this reason, we define a game as being playable only if the minimum frame rate is 25fps or higher.

We start by testing each card in each game at 1,024 x 768 with no AA or AF; in our opinion, every modern GPU that claims to be able to play modern games should be capable of delivering a smooth frame rate at this setting. The other test settings vary slightly depending on the game being tested. For example, we test F.E.A.R. at 1,280 x 960 with 2x AA and 2x AF, then at 1,680 x 1,050 with 2x AA and 8x AF, before finally testing at 1,920 x 1,200 with 4x AA and 16x AF.

Additional testing is carried out in Need for Speed: Carbon at 1,280 x 1,024 with 2x AA and AF enabled, before testing at 1,680 x 1,050 with 2x AA and AF enabled, and then finally at 1,920 x 1,200 with 4x AA and AF enabled. S.T.A.L.K.E.R. is tested at 1,280 x 1,024, 1,680 x 1,050 and 1,920 x 1,200, with AA and AF enabled in each instance.

We then try to overclock the GPU and memory of each card to their maximum stable frequencies before we encounter image corruption or stability problems. The overclocked performance results are shown on the F.E.A.R. graphs at 1,280 x 960 with 2x AA and 2x AF, and at 1,920 x 1,200 with 4x AA and 16x AF.

Each graphics card was tested in the same test rig, which comprises an Intel Core 2 Extreme X6800 overclocked from 2.93GHz to 3.19GHz, 2GB of OCZ PC2-8000 RAM and an Intel D975XBX2 motherboard. The test system runs Windows XP Professional SP2. To ensure that even the most power-hungry cards received a stable supply of voltage and current, the test rig is fitted with the DXX version of the awesome 1kW Enermax Galaxy PSU. Where available, each card was tested with the latest WHQL-approved drivers. For the Nvidia GeForce 6- and 7-series cards, this meant using ForceWare 93.71, and we used ForceWare 158.22 for the Nvidia GeForce 8-series cards. All of the ATi Radeon cards were tested using Catalyst 7.4, apart from the Radeon HD 2900XT, which was tested using Catalyst 8-37-4. Custom PC would like to thank Enermax for supplying the PSUs used in this Labs test.