SLI Performance Test Scores
In the past, I'd noticed a big difference between running with two video cards in SLI and running with just one of the two, with the single card scoring a good deal higher in Performance Test. However, I tested my current system with all the other settings identical and the difference was only eight points in scores over four thousand.
I don't know if the difference is in the latest nVidia cards and drivers or in the way the x58 chipset and/or Core i7 CPU interacts with them, but there doesn't seem to be a significant difference between SLI and single card processing.
We did do some testing with PT7 and multiple card configurations. In the end the results were fairly underwhelming and pretty much as you describe. In the complex test we were able to get a moderate frame rate increase by manually selecting AFR2 in the Nvidia control panel. Other than this we weren't able to get any speed-up to occur.
It's interesting that you used to get an increase in PT6. Up until PT7 we haven't done much multi-GPU testing. It may be the drivers as you say. There hasn't really been any changes to the existing 3D tests, other than updating the DirectX SDK from October 2006 to August 2008.
We also tested a Crossfire setup, and found that performance degraded compared to a single card unless we turned off Catalyst A.I. Turning it off produced results comparable to a single card.
I used to get a decrease in scores running SLI, so I had to remove one video card in order to get the best score. With my current system, SLI doesn't improve the score but it doesn't hurt it, either, so the second video card can be left in place. In 3DMark Vantage, adding the second card caused the score to jump from around 15 thousand to around 25 thousand, so it does make a difference in some situations.
Originally Posted by Michael (Passmark)
SLI (and Crossfire) generally only helps for a small number of applications for which the video card manufacturers have optimised the drivers.