Here's a commentary on Techreport's excellent video card evaluation article - pertinent to microstuttering on GTX 690 in 4 popular games compared against GTX 590 and HD 6990. It deserves a thread of its own, but I wanted to add it here to show at least some supporting evidence as to the reduced 7.5% penalty for GTX 690 compared against other multi-GPU cards that receive a 15% penalty against the overall benchmark ratings.
First, here's a link to the wonderful article, starting here: http://techreport.com/articles.x/22922/3
In this chart, we see that for GTX 690, the line across the chart is practically as flat as for GTX 680 and GTX 670 (while being a little lower with less frame times):
Now, let's compare this relatively flat line of the GTX 690 to the super-erratic of HD 6990 (while GTX 590's line is still pretty flat also but not quite as flat as GTX 670 below, or even the multi-GPU GTX 690 above):
HD 6990 performs considerably worse here:
Both GTX 590 and HD 6990 are at the bottom of the chart, with increased time spent below 50ms (likely for the badly "microstuttered/stuttered" frames during the demanding parts of the game):
However, for Batman: AC, we see that GTX 690 does exhibit a line that's never as flat as the single-GPU cards:
Above, the sharp spikes for the single-GPU cards are simply "stuttering" (or "hitching") rather than microstuttering. Microstuttering is constant, in that the line never gets to be flat (or really thin instead of "fat").
GTX 590's line below is only a bit worse than GTX 690, with more regular spikes:
And we can see that HD 6990 once again has frame times all over the place, consistently.
For Battlefield 3, GTX 690 once again expresses a relatively "plump" line that tells us that microstuttering is not completely fixed, but not too bad either:
GTX 590 shows spikes all over the place, while HD 6990 paints nearly half of the entire chart with its own color:
Above, HD 6990 is basically equivalent to a GTX 670, since the bottom spikes are so low that they're almost irrelevant to the top alternating spikes (which is what we're basically seeing with their longer frame times- being on the screen for perhaps 3-4 times as long as the "quick" alternating frames that are so instant). One thing to note, though is that GTX 590 does stutter (, not constant microstutter) badly with super-high spikes quite frequently.
By the way, Techreport made a new kind of chart which is rather interesting (at around 50%, HD 6990 shows a sharp rise in the average frame time, which directly translates to microstuttering - 50% being quick, and 50% being slow):
For the last game - Crysis 2, GTX 690's line is pretty flat, but not perfectly flat..
Just draw a flat line across the middle of this lime-green line (not exactly middle of it, but maybe 10% higher than the middle of it) and this is pretty much where GTX 690 is at with the actual "feel" of the frame times, which is still only a bit more than half of the single-GPU cards:
Below, we see that microstuttering is quite profound for BOTH GTX 590 and HD 6990 cards:
HD 6990's microstuttering makes it feel just like GTX 670 - with each alternating frame taking up most of the display time on the screen, the "median" line for HD 6990 is right about where the line for GTX 670 rests at. HD 6990 might have more than 25% higher fps than GTX 670, but the frame count is skewed in that HD 6990 does not really look any more fluid than GTX 670 across any TWO frames, one displaying after the other.
Again, we see a problem here:
In every one of those games, HD 6990's line looked like as if it were the Richter Scale, we'd be experiencing a constant earthquake!
The GTX 690 review article explains it more, covering it in some detail with a chart zooming in to the first few frames for closer analysis:http://techreport.com/articles.x/22890/8
And here we can see that microstuttering is present for GTX 680 SLI, much worse than GTX 690:
For more, on Crysis Warhead, see: http://www.pcgameshardware.de/aid,88116 ... st/?page=3
And Serious Sam 3: http://www.computerbase.de/artikel/graf ... kroruckler
(where microstuttering still exists for GTX 680 SLI rather than GTX 690)
Nvidia's attempt at curing the microstuttering "defect" is a very welcome one, albeit a long-overdue one after nearly 8 years of supporting SLI technology. Hopefully AMD could also attend to this problem for once and for all.
At least GTX 690 did not show any serious microstuttering or stuttering problems above....