Testing Setup and Performance
For this test we used the highest possible settings at 1080P and 1440P to see how it performs at those resolutions. SMAA T2X was shown to have no performance affect when compared to FXAA, and it looks better, so that was chosen as the base AA mode in these tests. We'll explore how it performs with maximum settings, then show the difference with Shadow Map size and also show the difference in memory usage with different Shadow Map sizes as well.
Test System
| Component | Selection |
|---|---|
| CPU | Intel Core i5-6600K |
| Motherboard | ASRock Z170 Extreme 4 |
| Power Supply | EVGA SuperNOVA 1300 G2 |
| SDD | SanDisk Extreme II 120GB |
| Storage Disk | Seagate 2TB |
| Memory | 16GB Crucial Ballistix DDR4 2400 |
| Monitor | Dell P2715Q |
| Video Cards | AMD R9 Fury, AMD R9 Nano, AMD R7 370, GeForce GTX Titan X, GeForce GTX 980, GeForce GTX 960 |
| Operating System | Window 10 64-Bit |
| Drivers | NVIDIA - 358.91 AMD - 15.11 Beta |
Thank you ASRock for providing the Z170 Extreme 4 motherboard for our test rig.
First lets take a look at 1440P, then 1080P. Because there's no internal benchmark, this is only really comparable to those that benchmark in much the same way. So take these results with skepticism, as it should improve as more patches are released as well.
Not exactly the most encouraging of results here. And the 290X being able to produce higher performance is just odd. But it was able to do so consistently on the newest beta drivers. I'd imagine this is a driver issue, however, and that future updates (Crimson and the official 15.11) should help to improve performance appreciably for Fiji based cards.
Even though the Fury is theoretically faster than the GTX 980, and it is in most situations, here it was not. Again, this is likely a not so greatly optimized driver. The Fiji cards had some very interesting slow-downs in the beginning mission, stuttering quite badly in some spots. This wasn't there on the 290X or the 370, nor was it there on any of the NVIDIA cards. Further testing concluded that the shadow map could be to blame, as it might render, and put into memory rather inefficiently, an extra-large shadow-map for perhaps the entire map. We'll look at a high-end and a low-end GPU from each manufacturer to see how performance is affected by this setting.
Those results are much more akin to what one would expect, and want, from higher end hardware. Simply reducing the size of the shadow map has an impressive impact on performance. The Fiji-based Fury is able to keep adequate numbers throughout, and the stuttering and frame-dips are gone. But the hypotheses is that high memory usage is causing swapping to system RAM, which negatively effects performance with the transfer of assets causing stuttering and even far lower framerates. But let's look at GPU memory usage for the above examples. The Titan X replaces the 980 to see how it uses larger amounts of VRAM.
Memory usage is certainly higher with the shadow map setting at Extra, but it never reaches the heights that it supposedly has with other tests. It never exceeds the maximum amount available, though that doesn't mean it isn't swapping into system memory.
To check out memory usage I recorded one of the tests using HWiNFO64 over 25 seconds.
Memory usage is high, though it isn't unexpectedly high considering. The question is what that memory is being used for. Is there swapping being done in the background, or is it something else completely unrelated? Unfortunately I was unable to test whether it scales with less RAM installed. If the VRAM swaps to the system RAM, and that swaps to the HDD, then it could result in some very poor performance. Thankfully there is a sharp drop during the playthrough, meaning that at least it doesn't continually increase and isn't a terrible memory use bug.
Conclusions
Right not it seems that Black Ops III isn't the best performer per-se. There are some obvious performance issues that can have a dramatic impact, though those associated settings can be adjusted so as to not really be of consequence. The range of adjustments is actually quite good, though still slightly limited compared to other AAA titles that are cross-platform.
The graphics are actually quite good, though. Texture resolution is good, the implementation of SMAA is fantastic, and the shadows, despite being memory intensive and performance killing, look and work well. Though there is little in the way of physics for clothing, the entire game looks great even if it doesn't perform like we'd like it to. So is this caused by the continued re-working of older code? Perhaps. The engine is a bit old, but it still is very capable despite the performance issues that are present.
Hopefully more patches are released so we can further test the improvements that are bound to happen. It shouldn't play like this on release anyway, and we certainly didn't expect this to be so poor, but it's definitely improving since it was first released.
Contents
Follow Wccftech on Google to get more of our news coverage in your feeds.
