Related Articles

Hardware

With the performance of this last generation of video cards reaching new heights, we felt it was time to upgrade our benchmarking rig. Compared to our last setup, this new testing bench offers a significant increase in processing power, allowing us to better stay clear of any bottlenecks:

• CPU: Intel QX9650 at 3.6 GHZ (400x9)
• Motherboard:  Asus P5Q Deluxe
• Memory: 4 gigabytes of Mushkin XP2-9200 DDR2 RAM (2 * 2048MB @ 1203 MHz[5-5-4-12]) 
• Power Supply: Enermax Galaxy DXX 1000W
• Hard Drive: 250 GB Seagate Barracuda 7200.10 SATA, w/16 MB cache
• CPU Cooler: Noctua NH-U 12
• OS: Vista-32,SP1

To compare against the Diamond HD4870, we selected the following cards: NVIDIA's GTX 280, an Asus EN8800 GTX, a VisionTek HD3870, a XFX 9800GX2, a Palit 9600GT Sonic, an Asus EN3850X2, and a VisionTek HD4850.

Software

Driver-wise, I used NVIDIA's 175.19 Forceware drivers with everything but the GTX 280 which used Forceware 177.35's, while for the ATI cards, we used Catalyst 8.6 drivers, except for the HD4870, which used an unreleased driver supplied by ATI ('sample 5').

On the software side of things, here is the list of programs used:

Bioshock: For this benchmark, all of the Detail settings were set to 'High'. All of the graphic option switches were set to 'On', with the exception of the following three settings: Vsync, Windowed mode, and Force Global Lighting. We used FRAPS to measure frame rate performance. The FRAPS run was 138 seconds, triggered from pulling the switch in the sub at game's beginning. The sub's dive involves many big models moving around, which should strain the GPU's and be a good measure of the game's engine.

Call of Juarez: We used the stand-alone Call of Juarez DX10 benchmarking program for these results. For our AA testing, we used a setting of 2x.

Crysis: These benchmarks were performed using 'fly-by' GPU test found within the single-player pre-release demo version of the game. All graphic settings were on High.' For AA, we used a setting of 4x. DX10 mode was used. The game has also been fully patched.

Enemy Territory: Quake Wars: We use this id FPS benchmark to test out higher resolutions. We used the highest possible detail settings. We tested the resolutions at 4x AA as well as at 8x AA. 16x AF was also used.  

3DMark06 and 3DMark Vantage: These popular synthetic benchmarking programs were used at a resolution of 1280x1024. Vantage was run in 'Performance' mode, and only the two GPU tests were used.

World In Conflict: We used the built-in benchmark of the demo version of this game. We ran the benchmark in DX9 rendering mode, with a 'High' level of quality. For the AA testing, we used a setting of 4x, and a setting of 16x for AF.

Media Error demo: A runner-up winner in the scene.org best demo category for 2007, this demo made by Fairlight , CNCD & Orange. If you have no idea what a demoscene demo is, you might want to go here for an explication, and it is recommended you follow the link above to see for yourself. Media Error in particular presents a serious work out for a video card. Although to my knowledge, no one in the history of hardware reviewing has used a demoscene demo for benchmarking purposes before now, demos such as these are naturally well-suited to the task. This demo was run at a 1600x1200 resolution, with no AA and with 4xAA,  and FRAPS was used to measure framerate averages.   

If you would like any further information about our benchmark settings, feel free to ask us in the forums.