Of course, that's just one game and surely one game can not represent all or even some of the games out there. So, we ran the test again, but this time with a different game - Richard Burns Rally. Unlike Splinter Cell: Chaos Theory, this is a much older, pre DirectX 9 game. There's a lot less complicated shaders and graphical features in this game, so it's a very straight forward 3D graphics engine. Let's look at the results below.
Richard Burns Rally - renamed to rollover.exe
| Forceware 91.31 AA AF | |||
| Single | |||
| 1024 x 768 | 1280 x 1024 | 1600 x 1200 | |
| Min | 85 | 61 | 42 |
| Avg | 129.459 | 89.329 | 59.245 |
| Max | 231 | 151 | 90 |
| Auto Select | |||
| 1024 x 768 | 1280 x 1024 | 1600 x 1200 | |
| Min | 96 | 74 | 55 |
| Avg | 153.265 | 119.868 | 82.743 |
| Max | 227 | 220 | 125 |
| Alternate Frame Rendering | |||
| 1024 x 768 | 1280 x 1024 | 1600 x 1200 | |
| Min | 106 | 105 | 71 |
| Avg | 154.914 | 144.921 | 98.238 |
| Max | 225 | 218 | 144 |
| Split Frame Rendering | |||
| 1024 x 768 | 1280 x 1024 | 1600 x 1200 | |
| Min | 97 | 74 | 54 |
| Avg | 152.828 | 119.851 | 82.728 |
| Max | 229 | 222 | 125 |
| Alternate Frame Rendering 2 | |||
| 1024 x 768 | 1280 x 1024 | 1600 x 1200 | |
| Min | 107 | 105 | 70 |
| Avg | 155.307 | 144.864 | 98.258 |
| Max | 224 | 222 | 144 |
all numbers are in frame rates per second except for 1024 x 768, 1280 x 1024, and 1600 x 1200
Ah, much better. All three modes of SLI multi rendering works they way they're supposed to and all offer a significant increase in performance. Looking closely at the numbers, it looks like the default SLI rendering mode is Split Frame Rendering or Scissor mode. This confirms what NVIDIA told reviewers when SLI came out. However, we can also see this may not be the most optimal mode. Alternate Frame Rendering and Alternate Frame Rendering 2 (obviously the same mode but with a workaround) can actually offer higher frame rates than Split Frame Rendering. During testing, these three modes are pretty much identical in image quality, right down to the same problem. At 1600 x 1200, there's a green tint on the gravel road throughout the whole replay, obviously something that's not supposed to be there. This is probably why NVIDIA felt it was necessary to include an embedded profile for this game in the new Forceware drivers.
OK. We have two games with two very different outcomes when we tried forcing SLI multi rendering mode. What to make of this? Actually, we think the answer is pretty simple. Splinter Cell: Chaos Theory has a much more advanced graphics engine that no doubt needed special 'handling' for it to scale with SLI. A much more simpler, straight forward graphics engine like the one with Richard Burns Rally will probably not need any such special handling or workaround. Looking ahead, this will probably mean games with a very advanced graphics engine may take a while to be fully supported by SLI. Quake 4 seems to be a perfect example of this. Even with the new Forceware 91.31, this game still won't scale with SLI (no such problem with Crossfire though ). To avoid this kind of scenario, NVIDIA has to be very aggressive in encouraging developers to code their games' graphics engine so it will be SLI friendly. While this means good news for SLI users, that's not exactly a strong point for SLI, since in doing so, that means the game will be Crossfire friendly as well.
Conclusion:
Has SLI improved with the new GeForce 7600 series? The short answer is a disappointing no. Even with the addition of vertex shader units and a much higher clocked core, the GeForce 7600GT SLI behave in a similar way to the GeForce 6600 SLI and 7800GTX SLI. We are not that much faster in Call of Duty and Quake 4 from a single card, we still saw a general decrease in performance with default settings in Call of Duty, Homeworld 2, SW: KOTOR, Quake 4, Serious Sam II and there is no change in SLI's anti aliasing that's remotely close to Super AA 10x and 14x from Crossfire. The games that respond well to SLI remains a limited few - F.E.A.R, Richard Burns Rally and Splinter Cell: Chaos Theory. In contrast, Crossfire only 'failed' in delivering a performance increase in Homeworld 2 and SW: KOTOR. In every other game, Crossfire pretty much stomped SLI, at least in the various games in our benchmark suite.One aspect that we didn't explore in this article is what would happen if we were to use a much faster processor. A faster processor will be able to handle the driver overhead much better, possibly even enough for SLI to break free and offer a performance increase. However, we don't think that's actually a strong point or even a strong argument for SLI. Both Crossfire setup we've tested scale very well with our Athlon 3500+. So, that actually means Crossfire needs less processor power than SLI! Even if we force ourselves to use a faster processor such as the Athlon FX 57 or FX 60, it would be very odd indeed to those processors just to have an increase with a pair or GeForce 7600GT.
With a hit-or-miss proposition on performance and a surpassed quality offering, the only advantage SLI has at the moment for users is ease of installation. Even the promise of ease of use (forcing multi rendering) is apparently not that valuable after all. Newer games will likely use more shader focused graphics engine, with new graphics technology such as normal / parallax mapping, multiple render targets, high dynamic range rendering that needs special tweaks for them to be scalable with multi rendering solutions. So, it's pretty unlikely forcing multi rendering from the control panel will offer you an increase in graphical performance with newer games that have not been profiled yet by NVIDIA.
The next big question will be what SLI can offer for physics. Of course, until Havok FX enabled games arrive, we won't have the answer to that question.
[Previous Page]
[Go to top]
[Home]