The Elder Scrolls V: Skyrim
Our test run for Skyrim was a lap around the town of Whiterun, starting up high at the castle entrance, descending down the stairs into the main part of town, and then doing a figure-eight around the main drag.
Since these are pretty capable graphics cards, we set the game to its "Ultra" presets, which turns on 4X multisampled antialiasing. We then layered on FXAA post-process anti-aliasing, as well, for the best possible image quality without editing an .ini file.
The plots above show the time required to render the individual frames of animation produced by the game during our 90-second test run. The uninitiated will probably want to read my article explaining our new testing methods in order to get a sense of what we're doing. Just looking at these raw plots, though, one can see that the vast majority of the frames are rendered in under 30 milliseconds, which is quite good. A steady stream of frames at 30 ms would translate into an average frame rate of over 30 FPS. More importantly, the relatively small number of high-latency frames from these video cards means all of them should deliver a relatively smooth sense of motion in the game.
However, I saved one plot for last. Notice that the vertical axis stretches to higher values here, and there are lots of really rather long-latency frames. The GeForce GTX 560 Ti is the only card of the bunch with just 1GB of video RAM onboard, and its memory capacity is overwhelmed at this four-megapixel resolution and image quality level. Several of the other cards have 3GB of memory, which is probably overkill for most single-display configs these days, but having more than 1GB on tap can sometimes be very helpful.
The 7970 and 7950 take the top two spots in the traditional average-FPS sweeps. Notice how the GTX 560 Ti manages to pull off an average of 36 FPS. Just looking at that outcome, one might be tempted to think it performed reasonably well here, which is not the case. Truth is, averaging out frame rates over a full second doesn't offer enough resolution to capture those frame time spikes that can make games feel laggy and sluggish.
Another way to quantify gaming performance that sidesteps some of the problems with FPS averages is to consider the graphics subsystem's transaction latency, much like one might do in benchmarking a database server. Seems odd, perhaps, but I think consistent delivery of low frame times is what a graphics card should do. The chart above shows the 99th percentile frame time—that is, 99% of all the frames were returned within x milliseconds—during our five 90-second test runs.
As you can see, the differences between the cards are really very minor here. Everything but the GeForce GTX 560 Ti delivers the vast majority of frames in well below 40 milliseconds. This measure tends to focus on the pain points, the most difficult spots during the test run. In this case, that would be the first few hundred frames of the test session, where you can see in the plot above that the 7950's frame times are just slightly higher than those produced by the GTX 580 and the GTX 560 Ti 448.
This last graph looks at how much time each video card spent spinning its wheels on frames that simply took too long to render. The idea is to quantify the depth of the problem when a card's performance begins to dip into unacceptable territory. Our threshold of 50 milliseconds would average out to a rate of 20 FPS. We figure anything slower than that probably isn't getting the job done in terms of maintaining a fluid sense of motion. For more on our rationale for this one, please see here.
Only the GTX 560 Ti, with its memory size issues, spends any time at all beyond 50 ms, it turns out. The 560 Ti wastes nearly two seconds during a 90-second run working on frames above our threshold, which is pretty awful. This is not smooth animation by any stretch, and our seat-of-the-pants impressions back up that assessment. I don't know if I'd say Skyrim is entirely unplayable at these settings, but it's definitely not fun or even acceptable.
On another note, I think we can say with confidence that there's almost no perceptible difference between the 7950 and the 7970 here.