I took the time to add BF4 into the mix, now that it's been released and we have drivers from both AMD and Nvidia optimizing BF4 performance. These results come from the single-player campaign. I'd like to test multiplayer eventually, but doing so and getting good, consistent results is not easy.
Since this game uses a 64-bit executable, it's not compatible with the FCAT overlay, so I used Fraps to record frame times.
Also, here's something different. As you may have noticed, Fraps frame time plots tend to involve more variance than FCAT plots. You'll sometimes see a "heartbeat" pattern in Fraps results—with one long frame time followed by one or two relatively short frame times—that doesn't show up in FCAT. That's because Fraps records frame submission times early in the pipeline, while FCAT measures frame delivery at the very end of the process. In between the two, triple buffering tends to smooth out small hiccups before the frames are delivered.
Depending on how the game's internal timers work, either Fraps or FCAT results may be more "correct" in describing the smoothness of the final animation. Each game seems to work a little differently, but our current understanding is that most game engines use some sort of moving average to determine how to advance their animation timing from one frame to the next. When that's the case, then FCAT results are the better indicator.
At the suggestion of AMD's Raja Koduri, we've attempted to simulate the effect of triple-buffering on our Fraps data by implementing a simple, three-frame low-pass filter (just a moving average, in this case.) As you'll see below, the filtered Fraps data is free from those quick "heartbeat" artifacts, and we believe it's a more faithful representation of BF4 animation smoothness. You'll still see frame time spikes in the filtered data, and those spikes are much more likely to have an impact on animation smoothness.
On another front, we've chosen a longer 90-second window for our BF4 test scenario and added a bit of an in-game warm-up period for each card prior to testing. That should help make sure our results reflect the true performance of Hawaii-based graphics cards with aggressive PowerTune settings.
Chalk up a clean, if narrow, win for the GTX 780 Ti across all metrics here.
|Nvidia admits, explains GeForce GTX 970 memory allocation issue||205|
|Here's my guest appearance on tonight's Alt+Tab Show||11|
|Watch John Romero talk about Doom level design||42|
|I'll be on Newegg TV's Alt+Tab show live at 3PM PT||15|
|Windows 10 build 9926 adds Cortana, Continuum, and more||40|
|Apacer exec sees 256GB SSDs falling below $70 this year||46|
|Deal of the week: IPS monitors with 4K and 1080p resolutions||11|
|Haswell CPU and Radeon graphics team up in ASRock's VisionX mini PC||14|
|HA. AMD in the red and nVidia in the green. Thats funny cause you know... *cough* oh forget it.||+81|