As 3D games enter their second decade, their march toward photorealistic graphics moves forward as well. But this realism doesn't come for free, as evidenced by the software's ever-increasing processing demands on the graphics processing unit (GPU). To meet this burden, graphics cards must grow more powerful with each successive generation. CNET Labs' benchmarks are designed specifically to evaluate just how powerful these new cards are and how they stack up against one another. By using a handful of popular game titles, such as Half-Life 2, Doom 3, and Far Cry, and by testing at high resolutions with advanced graphics features enabled, our tests help determine whether a particular card has what it takes for your gaming needs.

Test environment

The graphics card market changes so rapidly that it is very difficult to keep our test methodology current. As such, we are constantly tweaking our processes to keep them up-to-date. One set of tests that we use to compare one group of graphics cards frequently differs from the suite we use to compare a different group released only a short while earlier. Additionally, new graphics hardware tends to push the performance envelope, so we change our graphics card test systems frequently to ensure that the surrounding hardware keeps up with the latest cards. Additionally, AGP- and PCI Express-based graphics cards need to be tested in different systems. Refer to the individual graphics card reviews for the specific test bed configuration. It's also important to remember that we compare only cards tested with the same test bed and with the identical methodology, which means that when you read a graphics card review, you should compare the testing results only from cards within that review.

We endeavor to keep the document you are now reading current, but there might be times when the published test results for a particular graphics card review are posted using a modified methodology before we have the opportunity to update this document.

We run all tests with the most recent versions of Microsoft's DirectX and the graphics card's display driver. We disable Windows Update, System Restore, and Remote Assistance. We ensure that Windows XP's Idle Task Scheduling is complete before running the tests. All tests are run with vsync and audio turned off. For all resolutions tested, we set Display Properties to a 75Hz refresh rate.

It is difficult to do a truly equal performance comparison between graphics cards based on GPUs from different manufacturers. For instance, the quality settings in the display drivers from different manufacturers do not necessarily perform similar functions using the same techniques. In order to come as close as possible to an equal comparison, we set the display drivers' quality settings to the highest-available common settings. While the techniques might differ among manufacturers, the expectations of the end user should be the same regardless of the card being used.

While we use real games in our testing, it is important to keep in mind that the results generated during benchmarking are usually somewhat better than you can expect to see when actually playing the games. For example, in order to isolate the graphics subsystem as much as possible, we disable all game sounds during testing; removing the sound subsystem from the equation puts less stress on the overall system and subsequently boosts overall performance.

Half-Life 2

Half-Life 2

Valve's Half-Life 2 is the latest addition to our gaming benchmark arsenal, and it is generally considered one of the most advanced 3D games available. In addition to advanced physics and AI engines, Half-Life 2 utilizes Valve's own realistic, shader-based render engine, the Source Engine. This DX9-based game's popularity with the gaming set is unparalleled, and the Source Engine has even been licensed to other game developers--so you can expect to see more games using this same render engine in the not-too-distant future. With proper selection of game levels and image-quality settings, Half-Life 2 can be fairly demanding on a graphics subsystem.

After installing the game, we download and install the latest updates from Valve. We then make the following settings within the game:

• Combat difficulty: Easy
• Enabled developer console (~): Enabled
• Aspect ratio: Normal (4:3)
• Display mode: Full screen
• Model detail: High
• Texture detail: High
• Water detail: Reflect all
• Shadow detail: High
• Shader detail: High
• Wait for vertical sync: Disabled

We created our own custom demo of actual gameplay using Half-Life 2's d1_canals_07 map. The demo is comprised of more than 20,000 frames and includes gameplay that takes place in both inside and outside environments. The demo incorporates both water and flashlight textures. The demo is played back in offline mode with DX9 mode forced on. The demo playback generates an average frame rate score reported in fps; a higher frame rate is better.

Depending on the type of card we're testing and the performance comparisons we want to make, we run these tests at different resolutions and different antialiasing and anisotropic-filtering settings. Each graphics card review indicates the resolutions and the settings at which we ran the tests. We run the tests with antialiasing and anisotropic-filtering settings selected via Half-Life 2's internal settings; therefore, the graphics card's driver interface settings for antialiasing and anisotropic-filtering are set to Application Controlled.

Doom 3

Doom 3

Activision's Doom 3 is currently one of the most advanced 3D games on the market. The game uses OpenGL with its own lighting model where lighting effects and shadows are generated in real time--a fairly new rendering technique for games. The result is very impressive-looking graphics with realistic-looking lighting and shadows that respond accurately to moving objects. With the quality settings cranked high, the game can be very demanding on the graphics subsystem.

After installing the retail game, we patch it to version 1.1, which is available for download here.

Doom 3 settings:

• Video quality: High quality
• Full screen: Yes
• High-quality special effects: Yes
• Enable shadows: Yes
• Enable specular: Yes
• Enable bump maps: Yes
• Vertical sync: No

We've created our own custom demo of actual gameplay using Doom 3's Caverns-Area 2 map, and we play it back using the game's precaching option. The test generates an average frame rate score, reported in frames per second (fps); a higher frame rate is better.

Depending on the type of card we're testing and the performance comparisons we want to make, we run these tests at different resolutions and different antialiasing and anisotropic-filtering settings. Each graphics card review indicates the resolutions and the settings at which we ran the tests. The tests are run with antialiasing and anisotropic-filtering settings selected via Doom 3's command-line interface; therefore, the graphics card's driver interface settings for antialiasing and anisotropic-filtering are set to Application Controlled.

Far Cry

Far Cry

UbiSoft Entertainment's Far Cry is a Direct3D/DX9-based game that uses Vertex Shader 2.0 and Pixel Shader 2.0, as well as plethora of advanced rendering techniques, which all combine to produce some of the most realistic scenery and physics we've seen in a game title to date. As such, it is very demanding on a graphics subsystem and is therefore an excellent tool for evaluating high-end graphics cards.

After installing the retail game, we patch it to version 1.3, which is available for download here.

Far Cry settings:

• Antialiasing: None
• Full screen: Enabled
• Vertical sync: Disabled
• Play background video: Enabled
• Render mode: Default
• Advanced options: Very high
• Enable music: Disabled

We created our own custom demo of actual gameplay using Far Cry's Rebellion map. We play back the demo, which generates an average frame rate score reported in fps; a higher frame rate is better.

Depending on the type of card being tested and the performance comparisons we want to make, we run these tests at different resolutions and different antialiasing and anisotropic-filtering settings. Each graphics card review indicates the resolutions and the settings at which we ran the tests. We run these tests with antialiasing and anisotropic-filtering settings selected via the graphics card's driver interface.

Unreal Tournament 2003

Unreal Tournament 2003

We no longer use Epic Games' Unreal Tournament 2003 as part of our graphics card testing suite; the information is included here for reference for our older graphics card reviews that were tested using this benchmark. We use the popular first-person shooter, Unreal Tournament 2003, to measure a graphics card's 3D graphics gaming performance.

After installing the retail game, we patch it to version 2225, which is available for download here.

We use a subset of the tests from the retail game's built-in benchmark: the Flyby-Antalus test. By default, Unreal Tournament 2003 automatically adjusts settings based on a graphics card's capabilities. To prevent this from happening, we make a number of changes to some of the game's files. By doing this, we ensure that the same game settings are used for each graphics card tested so that we can compare the performance between graphics cards.

We edit the MaxDetail.ini file, located in the \UT2003\Benchmark\Stuff directory, to change the values of the Fullscreenviewport's X and Y axes to correspond to the resolution we are testing; we set the minimum desired frame rate to 0; and we delete all of the text in the [ALAUDIO.ALAudioSubsystem] section in order to disable the game's audio. We then set the MaxDetail.ini and the MaxDetailUser.ini files as read only. We edit the Flyby-antalus.bat file, located in UT2003\Benchmark\ directory, and add the following text:

-ini=..\\Benchmark\\Stuff\\MaxDetail.ini-userini=..\\Benchmark\\Stuff\\MaxDetailUser.ini

Depending on the type of card we're testing and the performance comparisons we want to make, we run these tests at different resolutions and with different antialiasing and anisotropic-filtering settings. Each graphics card review indicates the resolutions and settings at which we ran the tests. These tests are run with antialiasing and anisotropic-filtering settings selected via the graphics card's driver interface. Unreal Tournament 2003 performance is reported in fps; a higher frame rate is better.