Image Quality

Anti-aliasing

For quite some time ATI had a sizeable lead on NVIDIA in AA image quality, RADEON 9700 PRO (and ATI’s follow-up 9800 products) provided sharper visuals than GeForce FX, but starting with the GeForce 6 series, NVIDIA has narrowed the gap substantially, thanks to a new rotated-grid sampling pattern, this is the same method ATI uses for X800. Today, the AA quality of cards from both manufacturers is quite close.

You can see this pretty clearly in our Battlefield 2 AA screenshots, both the X850 XT PE and GeForce 6800 Ultra do a very good job of removing the jaggies in our reference image:

X850 XT PE 0xAA

GeForce 6800 Ultra 0xAA

X850 XT PE 4xAA

GeForce 6800 Ultra 4xAA

Zooming in on the antenna however reveals a slight edge for ATI’s X850 RADEON XT PE:



RADEON X850 XT PE 4xAA




GeForce 6800 Ultra 4xAA


You can also see edges on ATI’s card are slightly sharper on the top edge of the bunker in the screenshot above:



RADEON X850 XT PE 4xAA




GeForce 6800 Ultra 4xAA

Anisotropic filtering

While you can clearly see ATI’s edge in AA with Battlefield 2, anisotropic filtering is a much more subjective topic, what looks “superior” to one person’s eyes may look inferior to another. In addition, the environments in BF2 are so large that turning on this feature does come with a significantly greater performance hit than we’ve seen in previous titles. You’ll see this in our performance benchmarks in the following pages.

Therefore, rather than providing a winner and loser, we’ll just provide a sample image for representing each card’s AF quality:

X850 XT PE AF disabled

GeForce 6800 Ultra AF disabled

X850 XT PE 8xAF

GeForce 6800 Ultra 8xAF

How we tested

Normally whenever we benchmark with a new title, we like to use custom demos we’ve recorded of actual gameplay. Once we’ve recorded our demo, we then use FRAPS or the game’s built-in timedemo feature to record an average frame rate, and in some cases the minimum and maximum frame rates as well.

By using demos to measure performance, we’re assured that each card tested is handling the exact same identical workload. This ensures repeatable results regardless of the hardware used, or graphics driver used. The other method we will sometimes use is running through the game and recording the frame rate with FRAPS.

This method is a little more tricky, as variables within the game can prevent you from getting an identical workload from graphics card to graphics card. For instance, during periods of inactivity in games like BF2, your character may automatically perform involuntary actions such as flexing his hands or toying around with the weapon you’re currently holding. In Far Cry, birds would randomly fly across the screen, while in BF2 the grass in front of you will flow with the wind.

Basing performance on a combat sequence with this testing method is a definite no-no, as your enemy may use his rifle to shoot at you with one card, and toss a grenade at you while you’re testing performance with a second card. As you can certainly imagine, this would have a profound difference in performance. This isn’t a problem when using recorded demos to test cards, as your enemy does the same thing every time.

In this case, since BF2 doesn’t provided demo recording functionality, we had to use the second method. To ensure the results were as repeatable as possible we created an empty server, hopped on it, and ran down the same path each time, with FRAPS running in the background recording our frame rate. We repeated this procedure three times per card for each resolution/setting, so at 1600x1200 we ran the GeForce 6800 Ultra three times at 1600x1200 before proceeding on to 2048x1536. We took the average of the three runs.