Frame Rates & Times
First, let's take a look at average and minimum frames per second, both of which are certainly interesting. We're using three different resolutions with the High preset: Full HD (1920x1080), WQHD (2560x1440) and Ultra HD (3840x2160). The results are presented in galleries you can scroll through.
The minimum frames per second reflect the subjective experience best, especially when it comes to stuttering in a few spots. With only a few exceptions, Nvidia edges out AMD. Then again, results like this don't necessarily tell the whole story, as we'll see below.
Frame Rates Over Time
Let's take a look at the frame per second curve for the benchmark's entire 180-second run next. Our interpreter summarizes the log file's data so that we can put it in a graph that shows how frame rate changes across the benchmark's different scenes with their different loads.
Frame Time And Smoothness
The graph above is still just not detailed enough to provide us with a good representation of the subjective experience. The only way we can achieve this is by looking at the actual render times. Since different graphics cards render different numbers of frames during the benchmark run depending on their speed, we're using a complex mathematical procedure to have our interpreter analyze the log file. Larger deviations are weighed differently compared to consistent sequences. This way, the subjective stuttering is retained.
The 2560x1440 resolution with the High preset is great for us because the faster Nvidia graphics cards hit their limit earlier due to the increased CPU load. Even though the frame rates and frame rates over time don't show this at all, this graph indicates that the CPU sets a clear cap on performance due to the draw calls.
Finally, let's take a look at the smoothness, which is to say the relative time differences between frames. Looking at the benchmark results this way shows any stuttering that you might subjectively experience better without being influenced by the actual rendering time.
Using the same settings to examine the subjective user experience, AMD's Radeon R9 Fury X offers a calmer and smoother picture across the entire benchmark run. Because we also statistically analyzed the CPU and driver data, we'll see why this is the case shortly.
Frame Rates And Times Bottom Line
AMD's graphics cards do particularly well in scenes that have a lot of AI activity (and its accompanying compute load).
The older Radeon R9 390X is especially noteworthy in this context, since it's actually playing in Nvidia’s GeForce GTX 980 Ti's league, and not just in the markedly slower non-Ti version's. Similarly, an overclocked AMD Radeon R9 390 handily beats both Nvidia's overclocked GeForce GTX 970 and 980.
At Full HD, AMD's Radeon R9 380X does pretty well, whereas Nvidia's GeForce GTX 960 can't compete.
All of these results are based on just one benchmark, of course, which isn't representative of most games. Still, it gives us a bit of an idea where optimizations for asynchronous shading/compute might lead, though.