Testbed and Methods

We used the following hardware for this test session:

• AMD Athlon 64 FX-60 CPU (2x2.60GHz, 2x1MB L2)
• ABIT AN8 32X mainboard (nForce4 SLI X16) for Nvidia GeForce 7 cards
• ASUS A8R32-MVP Deluxe mainboard (ATI CrossFire Xpress 3200) for ATI Radeon and Nvidia GeForce 8800 GTX cards
• Corsair TwinX 2048-3500LLPRO memory (2x1GB, CL2-3-2-6)
• Maxtor MaXLine III 7B250S0 (Serial ATA-150, 16MB buffer)
• Creative SoundBlaster Audigy 2 sound card
• Enermax Liberty 620W power supply (ELT620AWT)
• Samsung SyncMaster 244T monitor (24? 1920x1200@75Hz max display mode)

• Microsoft Windows XP Pro SP2, DirectX 9.0c
• ATI Catalyst 6.11
• Nvidia ForceWare 93.71 for the GeForce 7 series
• Nvidia ForceWare 97.02 for the GeForce 8800 GTX

The graphics card drivers were set up in such a way as to provide the highest possible quality of texture filtering.

ATI Catalyst:

• Catalyst A.I.: Standard
• Mipmap Detail Level: High Quality
• Wait for vertical refresh: Always off
• Adaptive antialiasing: Off
• Temporal antialiasing: Off
• High Quality AF: On
• Other settings: by default

Nvidia ForceWare:

• Texture Filtering: High quality
• Anisotropic sample optimization: Off
• Trilinear optimization: Off
• Force mipmaps: None
• Threaded optimization: Auto
• Gamma correct antialiasing: On
• Transparency antialiasing: Off
• Vertical sync: Force off
• Other settings: by default

We selected the highest possible graphics quality level in each game using only those settings that were accessible to the player. We didn’t modify the games?configuration files that might contain additional settings. Performance was measured with the games?own tools or, if not available, manually with Fraps 2.8.1. When measuring the speed of a card with Fraps, we chose simple routes in the game to achieve repeatability of results and reduce the measurement error. For the same purpose we would take a long enough fragment when using a cut-scene on the game engine. We also measured the minimum speed of the cards where possible because this is often the criterion that determines how comfortable it is to play a game.

We tested the cards in three standard resolutions: 1280x1024, 1600x1200 and 1920x1200. We didn’t use the “pure speed?mode (without full-screen antialiasing) because it is not relevant for top-end graphics cards. Instead, we used two additional high-quality antialiasing modes. The test session including both single cards and multi-GPU configurations, different antialiasing methods were used in the FSAA 6x/8x + Aniso 16x and FSAA 14x/16x + Aniso 16x test modes. They are listed in the table:

As you see, we focus on quality and choose the best-quality, although resource-consuming, 8x MSAA (called 8xQ in the Control Panel) and 16xQ CSAA modes on the GeForce 8800 GTX. The latter mode features a double number of color/depth samples in comparison with 16x CSAA

We enabled anisotropic filtering and standard FSAA modes from the games?own menus. If not possible, we forced the necessary mode from the driver. The high-quality antialiasing modes were forced from the driver. On the GeForce 8800 GTX, if the game supported standard FSAA modes, we used the Enhance the application setting option described in our theoretical review. We benchmarked the cards without FSAA in those games that do not support it due to the specifics of their engines.

Besides the GeForce 8800 GTX, we included the following graphics cards into this review:

• GeForce 7950 GX2 (2xG71, 500/1200MHz, 48pp, 16vp, 48tmu, 32rop, 256-bit, 512MB)
  http://www.xbitlabs.com/articles/video/display/nvidia-gf7950gx2.html
• GeForce 7900 GTX (G71, 650/1600MHz, 24pp, 8vp, 24tmu, 16rop, 256-bit, 512MB)
  http://www.xbitlabs.com/articles/video/display/geforce7900gtx.html
• Radeon X1950 XTX (R580+, 650/2000MHz, 48pp, 8vp, 16tmu, 16rop, 256-bit, 512MB)
  http://www.xbitlabs.com/articles/video/display/ati-x1950xtx.html

Multi-GPU systems were represented by the following configurations:

• GeForce 7950 Quad SLI
  http://www.xbitlabs.com/articles/video/display/quad-sli-two.html
• GeForce 7900 GTX SLI
  http://www.xbitlabs.com/articles/video/display/nvidia-gf7950gx2.html
• Radeon X1950 XTX CrossFire
  http://www.xbitlabs.com/articles/video/display/quad-sli-two.html

So, we’ve got the fastest single graphics cards as well as all modern dual- and even four-processor configurations for a more accurate comparison. Unfortunately, we couldn’t benchmark a SLI tandem made out of two GeForce 8800 GTX due to problems with our nForce4 SLI X16-based mainboard, but we are going to do that in an upcoming review.

We used the following games and benchmarks:

First-Person 3D Shooters

• Battlefield 2
• Battlefield 2142
• Call of Duty 2
• Far Cry
• F.E.A.R.
• F.E.A.R. Extraction Point
• Tom Clancy’s Ghost Recon Advanced Warfighter
• Half-Life 2: Episode One
• Prey
• Quake 4
• Serious Sam 2

Third-Person 3D Shooters:

• Hitman: Blood Money
• Splinter Cell: Chaos Theory
• Tomb Raider: Legend

RPG

• Gothic 3
• Neverwinter Nights 2
• The Elder Scrolls IV: Oblivion
• Titan Quest

Simulators:

• Pacific Fighters
• X3: Reunion

Strategies:

• Age of Empires 3: The War Chiefs
• Company of Heroes
• Rise of Nations: Rise of Legends

Synthetic Benchmarks:

• Futuremark 3DMark05 build 1.2.0
• Futuremark 3DMark06 build 1.0.2