AMD Crossfire issues?
In one area Nvidia's FCAT method has provided some very interesting insights, and that's with multiple GPU configurations. The frametime story is a bit more complicated for SLI and Crossfire, because when you combine two cards a technique called Alternate Frame Rendering (AFR) is employed. It basically means that the two cards alternate in processing frames: one takes the even ones, the other the uneven ones.
To keep things smooth the two cards have to be in completely sync, but as you can imagine this isn't always a given considering the variances in frame times. When the two cards aren't in sync, micro-stuttering can occur.
For smooth graphics GPUs in SLI/Crossfire have to synchronize, with constant and even timings between frames.
If they're not in sync, as can happen with AMD Crossfire, this can result in micro-stutters.
Nvidia has incorporated a technique in its GPUs that tries to maintain synchronization as much as possible, which it calls frame metering. AMD does not yet a similar technique operational. On various forums you can read that people feel that Nvidia SLI is "better" than AMD Crossfire, but it's been difficult to objectively establish why that is. It wasn't really reflected in frames per second performance, or even in the frametime tests.
However, the FCAT tests did suddenly show a major issue with Crossfire. Many frames that had been processed were only displayed very briefly, or not at all. When such 'runt' frames, as Nvidia calles them, are filtered out in the analysis, Crossfire appears to perform significantly worse than SLI.
AMD is promising a new driver with frame synchronization that's supposed to be available in June or July. So if the new dual-GPU Radeon HD 7990 is released before the new driver is available, that could be an issue.
This chart from Tech Report says it all. When you filter out the runt frames, Crossfire performs worse than SLI.