Forcing Alternate Frame Rendering
Out of the games we tested, at least three games (Homeworld 2, Richard Burns Rally and Star Wars: Knights of the Old Republic) gained little benefit with Crossfire. This may be caused by the absence of game profiles embedded in the drivers or to avoid compatibility issues with AFR rendering. Crossfire don't have any control panel options like NVIDIA's SLI which has a game profile manager and the options to force SLI multirendering mode. This is a good thing, because general users and gamers can just let the driver decide for them the profile that best suits the game or fall back to Crossfire default rendering mode to avoid any compatibility issues. However, this can be annoying at times - if a new game comes out, you'd have to wait at least a month for the Catalyst team to update the drivers with the profile for that game.Thankfully, all is not lost. There is a way to get around this situation. It may not be pretty and sometimes it won't work. What you need to do is rename the game's executable file into a game that has a game profile embedded in the drivers. Out of the various games we tested today, the most gains are with Quake 4 and Call of Duty for OpenGL based games and Splinter Cell: Chaos Theory, Serious Sam II and F.E.A.R for Direct3D games. So, we've decided to give this method a try. We rename the executable files of Homeworld 2, Richard Burns Rally and Star Wars: Knights of the Old Republic to force AFR mode in Crossfire.
We limit testing to 1024 x 768, since this is the sweet spot for two Radeon X1600XT in Crossfire mode. These are taken in Crossfire and Catalyst AI enabled and ran in both default settings and with AA and AF
| Homeworld2.exe | Renamed to Quake4.exe | Renamed to CODMP.exe | Single, Catalyst AI disabled | |
| Default | ||||
| Min | 37 | 43 | 43 | 44 |
| Avg | 105.64 | 111.44 | 111.38 | 105.64 |
| Max | 360 | 359 | 357 | 358 |
| AA AF | ||||
| Min | 34 | 38 | 40 | 34 |
| Avg | 94.1 | 104.54 | 103.39 | 92.91 |
| Max | 362 | 358 | 367 | 367 |
| richardburnsrally.exe | Renamed to SplinterCell3.exe | Renamed to Sam2.exe | Renamed to FEAR.exe | Single, Catalyst AI disabled | |
| Default | |||||
| Min | 90 | 98 | 96 | 99 | 90 |
| Avg | 147.19 | 146.85 | 147.38 | 147.18 | 145.17 |
| Max | 208 | 211 | 207 | 208 | 201 |
| AA AF | |||||
| Min | 52 | 97 | 97 | 97 | 47 |
| Avg | 78.67 | 136.25 | 136.33 | 136.35 | 68.945 |
| Max | 142 | 204 | 195 | 200 | 123 |
| swkotor.exe | Renamed to SplinterCell3.exe | Renamed to Sam2.exe | Renamed to FEAR.exe | Single, Catalyst AI disabled | |
| Default | |||||
| Min | 22 | 25 | 25 | 24 | 24 |
| Avg | 50.05 | 59.29 | 59.96 | 58.09 | 58.55 |
| Max | 63 | 71 | 71 | 72 | 71 |
| AA AF | |||||
| Min | 15 | 11 | 18 | 19 | 18 |
| Avg | 40.74 | 31.9 | 43.13 | 42.83 | 46.27 |
| Max | 51 | 41 | 54 | 54 | 58 |
all the numbers are frame rate per seconds
A note on Homeworld 2 - forcing AFR through renaming the executable file resulted in display corruption when running at default settings. However, these display corruption disappears once AA and AF is enabled. So this 'hack' doesn't work smoothly all the time, but when it works, it works beautifully. We gain several frame rates with AFR, though it's not that significant. It's important to note the increase is present on both default and AA / AF settings.
The same also applies to the Richard Burns Rally and SW: KOTOR. For Richard Burns Rally, it seems that ATI's driver developers simply haven't had the chance to include a profile yet. We saw no display corruption or rendering issues (although we didn't test the entire game for any problems) with renaming the executable file and you can see AFR works wonders with Richard Burns Rally. Instead of dropping half the frame rate, we're now seeing a mere 10 fps drop between default and AA / AF results! You'll also notice there's no slowdown when running at default settings, we even gain around 10 fps for the minimum fps. This is far better than what we achieve with SLI.
With SW:KOTOR, instead of a slowdown now we're getting a slightly higher frame rate. It's not much, but can still be quite significant - a 10 fps average in default settings and 2 - 5 fps with AA and AF enabled. Renaming it to Sam2.exe or FEAR.exe seems to work better than SplinterCell3.exe. Unlike the other two games, we can see we can see nothing is gained with Crossfire - minimum frame rates are similar to what we're getting in a single card setup with Catalyst AI disabled.
These results cast some new light at Crossfire. Foremost, Crossfire AFR mode works beautifully, in most cases you'll see a definite increase in performance, both at default and with AA / AF enabled. The increase does vary, mostly depending on your processor and the chosen graphics cards vertex shader performance. In pixel shader limited situations, the gains can be very impressive - 70 to 100 percent. At it's worst, you'll get a minimal impact in performance, mostly on the upper limit (SW: KOTOR) so it's virtually penalty free. This is very different from SLI, where driver overhead can incur quite a significant performance penalty.
Temporary Conclusion:
If we have to make a conclusion right now based on the information we got during testing for this article, ATI really have a solid offering in Crossfire. NVIDIA should be worried. Even with quite a slow card and a limited system, a pair of Radeon X1600XT in Crossfire scale beautifully in performance, although the increase is mostly limited to 1024 x 768. If we have a gripe with Crossfire, it's ATI's insistence of not allowing users to control Crossfire rendering modes directly through the drivers (via Catalyst Control Center). There's is a way around that, but it would've been nice to have since not all games react 'kindly' to renaming their executable files (Hitman: Blood Money).However, let's postpone making a final conclusion for now. We've only looked at the low end of the spectrum. We're currently testing a pair of Radeon X1900 in a Crossfire setup to see what a high end Crossfire setup has to offer. Since even a single X1900 is system limited with our Athlon 3500+, we'll be focusing more on quality rather than performance in part 2 of this article.
[Previous Page]
[Go to top]
[Home]