Performance, Frametimes DX11 and DX12

Frametime and latency performance

The charts below will show you graphics anomalies like stutters and glitches in a plotted chart. Frame time and pacing measurements. 

• FPS mostly measures performance, the number of frames rendered per passing second.
• Frametime AKA Frame Experience recordings mostly measures and expose anomalies - here we look at how long it takes to render one frame. Measure that chronologically and you can see anomalies like peaks and dips in a plotted , indicating something could be off. 

We have a detailed article (read here) on the methodology used. Basically the time it takes to render one frame can be monitored and tagged with a number, this is latency. One frame can take say 17 ms. Higher latency can indicate a slow framerate, and weird latency spikes indicate a stutter, jitter, twitches; basically, anomalies that are visible on your monitor. What these measurements show are anomalies like small glitches and stutters that you can sometimes (and please do read that well, sometimes) see on screen. Below I'd like to run through a couple of titles with you. Bear in mind that Average FPS often matters more than frametime measurements. 

As you might have observed, we're experimenting a bit with our charts to give it a little more clarity. At the left side, you can see the frame time in milliseconds (ms). At the x-axis, 30 seconds of the game spread out over roughly 1400~1500 frames. Please understand that a lower frame time is a higher FPS (!), so for these charts, lower = better. Huge spikes would be stutters, thicks lines would be bad frame pacing, and the graduate streamlining is framerate variation.  

DirectX 12 - on or off?

With the frame times and pacing out of the way, we can do some other stuff. The technique used is really handy to measure other stuff to check out behaviouristics. We've talked about DX12 many times in the past. This game specifically offers an option to turn it off (DX11), or on(DX12. History has proven that poorly optimized games actually can suffer from DX12 ASYNC on. That's totally different from this one. We take two brands again and have a peek by running both modes.

For NVIDIA I took a GeForce GTX 1070. The dark line is the card in the 30 seconds test with DX11, light grey is DX12. As you can see, NVIDIA benefits from DX12. Also, frame times and pacing overall look better. Now, 1070 is GPU limited at 2560x1440 so the difference is smaller than I expected as I've also tested an RTX 2080 Ti, the difference between DX11 and DX12 was hugely in favor of DX12. However, I will show you some perf FPS results later on with that, for now, we're just talking frame times. 

AMD then, I think I can safely say out loud that AMD less is struggling a bit more on the lower resolution than NVIDIA. Here is where you can find API and driver overhead, so DX12 should help, and indeed it does. Look at that Vega 64 going at DX12. That is a substantial performance increase at DX12. you can also see a much smoother pacing at DX12.

Now I specifically point you here, these are the results from both cards at DX12. The pacing is great, we see a few petit spikes for AMD, but these are below ~22ms you would not notice it. 

Average FPS between DX11 and DX12 with flagship GPUs

I just told you at the 1070 frame time entry that a 1070 at 2560x1440 is a bit GPU limited, so no matter what processor you'd fire off at it, it cannot go faster. So in the upper chart, I did some runs from Full HD towards Ultra HD on the GeForce RTX 2080 Ti, the fastest card on the planet that no-one can afford. 

The result yes, is staggering. DX12 up-to 2560x1400 massively benefits from the DX12 API. Considering the frame times and pacing also look better, we cannot advise anything else other than running DX12 on NVIDIA GeForce graphics cards. Even when GPU limited, you'll benefit from smoother frame times compared to DX11. 

So what about AMD? Well, there is just no doubt about it, the VEGA64 also benefits greatly from DX12, here as well up-to 2560x1440. After that resolution, GPU limitation kicks in, and no CPU or renderer API kind shy away from that.