I think there's some misunderstanding of what the new latency benchmarks are actually showing. I'm going to just discuss a few TechReport charts for illustration, as they've been doing it the longest and have the best handle on why this testing makes sense:
Average FPS (the standard benchmarking method):
99th Percentile Frametime (a close approximation of frames per second):
Now, what I believe everyone is focusing on is how the dual-card setups perform better in the 99th percentile frametime. That should not be surprising - they are
faster. Therefore, on average, their frametimes
should be shorter. Now, there are some anomalies, such as the 7970CF's 99th percentile frametimes being worse than the Titan despite being faster, and that's a hint of what's to come, but it's quite close.
Time spent beyond 50ms (the
latency benchmark):
Source:
http://techreport.com/review/24381/nvidia-geforce-gtx-titan-reviewed/12
It's when you get to the frametimes over 50ms, which is a rough approximation of a frame that the user could experience as a stutter, where this new technique really provides us more information than FPS. Even though the 690 and 7970CF are
faster than the single card equivalents, they are
not smoother. Yes, the 690 appears to be smoother than the 680 in some of the other games, but I'm not sure why this should be such a surprise
if Nvidia has a method to smooth perceived latency. This does not mean SLI is more consistent, however, and I believe this is where there is confusion. All the frametime graphs show more variability in the frametimes on SLI setups, but "smoothness" is impacted only when frametimes are long enough to actually be perceived as out of sync with other frames.
There are enough reviewers out there who've provided descriptions of actual gameplay to know that in some cases, 7970CF does not work as well as the FPS would suggest. That is not to say that it will not be improved by the major driver re-work, but it is short-sighted to just say these new benchmarks don't make sense and therefore cannot be used to support what reviewers have long been reporting subjectively.
Now, here's a graph for the same game from PCPerspective:
Source:
http://www.pcper.com/reviews/Graphi...ance-Review-and-Frame-Rating-Update/Far-Cry-0
What you'll see is that the SLI setups are very fast, have average frametimes that are much lower (necessarily), but are also less consistent, and more importantly for the user, have more latency spikes. Where you report the cutoff in a chart will very much color how readers interpret the findings, but these raw data plots tell the whole story. In fact, I could make the argument that setting aside the initial spike for the GTX680, the 680 is
smoother than Titan SLI.