Updated Video Card Testing Methodology

Articles

Event Coverage | Editorials | Guides

Author:
Editor:
Sponsor:
Published:

Max Slowik
Kurtis
N/A
May. 23, 2008

Article

Specifications

Images

User Reviews (0)

Comments (0)

<< Previous

Next >>

DirectX 10 Titles

Finally, the results. The games we've picked cover each API, DX10, DX9, and OpenGL. I picked these titles because they are ridiculously popular (even the hot-but-dumb Crysis) and also span the spectrum on performance requirements.

We bench at two resolutions, 1440x900 and 1680x1050, the first, equivalent to 1280x1024 megapixel-wise, makes up about half the displays people use (according to the Valve survey statistics). The second is the de-facto "Hi-Def" resolution, and is nice viewing if you can get it. We dropped 1920x1200 because so few people use monitors of that resolution. We'll continue to use it for special image quality reviews of high-end video cards, but it's just too damn big for the rest of the hardware.

As far as driver-specific settings go, we avoid them. We stick with standard box-filter anti-aliasing and no transparency anti-aliasing. Those too, we'll include in image quality reviews, but they're decidedly not universal.

Company of Heroes

Company of Heroes has average system requirements for a DX10 game. We all want a machine that can play Crysis with a billion-x anti-aliasing across three displays, but yeah, those don't exist. CoH (as they call it in my country) represents your vanilla DX10 experience. The detail settings are all set to "high", with model detail at full, and we only test 2xAA. We use the included timedemo, and average the results of three runs. (Oh God does Hans' coffee suck.)

Crysis

The, if I may, ball-breaker game. It might be an uninspired title plot-wise, but even the timedemos look like movie footage. This is the game that makes even the best computers cry, and because reviewers are sadistic, we all like it. Truly, if a video card can rip up this title, it should be sought by all. Anti-aliasing isn't an option since most cards have a hard time playing the game on "high", which we benchmark after "medium". Also, we don't use the GPU timedemo, because it doesn't reflect the gameplay in the least. We instead use "CPU1" (three runs, results averaged).

Unreal Tournament 3

UT3 is has the leanest system requirements of the DX10 titles we test, and it uses a popular engine; Unreal Engine 3 is licensable and other games use it as a backbone, including Gears of War and BioShock. So it's pretty important and you can turn up the AA--we test with 4xAA. While it doesn't have a timedemo, it's very easy to set up a botmatch and benchmark actual gameplay; we run three three-minute botmatches to get our results.

Performance Notes

Appending the graphs are the notes. While I'll probably just translate the numbers to good! and crap!, with some low-end hardware I have to tweak the settings, or lower the anti-aliasing just to get results. Here's where the justifications live.

<< Previous

Page 3 of 6

Next >>

Page 1: Introduction, The Card & Bundle
Page 2: Specifications and Setup
Page 3: DirectX 10 Titles
Page 4: DX9, OpenGL, and Synthetics
Page 5: Video, Power, and Overclocking
Page 6: Conclusion

Subscribe to Editorials [more info]

Add Comment

0 User Comments

Add Comment

To add a comment without being a member, you may omit the password field, but you must enter your name (or nickname) along with your comment. * Denotes required fields.