Today we're testing the performance of the Final Fantasy XV CPU against the new standalone benchmark that was released before the PC launched next month. The reason we want to look at CPU performance first is that the game is extremely CPU intensive, far more than we expected. Developer Square Enix recommends players to pack at least one Core i7-3770 or FX-8350, which means significantly more firepower from Intel. We will take care of it shortly.
The point here is that the game requires an 8-thread CPU for the recommended specs, while on the GPU front they think you can get away with a GTX 760 or R9 280. Oddly enough, AMD offers far more firepower in this second comparison. On the other hand, this is a Nvidia-sponsored title that GameWorks is branded into.
According to Nvidia, the game offers the integration of Nvidia Flow, Nvidia HairWorks, Hybrid Frustum Traced Shadows, lawn effects and Voxel Ambient Occlusion. As far as I know, these GameWorks features are only enabled with the "High Quality" preset. Although the CPU test was done with a GeForce GTX 1080 Ti, I also tested the default settings for standard and lite quality.
The standalone benchmark lasts more than 5 minutes and includes various events, cards and characters that are used in the game. For our test, however, we give frame rates for a 90-second part of this benchmark. All results are based on an average of 3 runs. The test begins at the beginning of the car journey and ends shortly after everyone has got out of the car. Please note that I have noticed that the frame times between crossfading to and from transitions have often been negatively impacted, so I avoided testing them.
As usual, all automatic overclocking functions in the BIOS were deactivated and locked CPUs were coupled to the corresponding memory. For example, the Core i3-8100 was tested with DDR4-2400 memory. The unlocked processors used DDR4-3200 memory, and so did all AMD Ryzen CPUs.
At the beginning we have the high quality results at 1080p. This is the default for maximum quality, for which all GameWorks features are enabled. Here, the GTX 1080 Ti caused a system bottleneck with an average of 91 fps. It is therefore quite shocking to see that such a high-end GPU is limited to less than 100 fps at this relatively low resolution.
In this diagram, we focus on the average frame rates before sorting the data by the results of the frame time. Here we see that the Ryzen processors all perform very well compared to 8th generation Intel CPUs, although Intel is limited by the GTX 1080 Ti.
Even the much older Core i7-2600K and the FX-8350 look pretty good, although the A12-9800 falls on a bunch and can't deliver playable performance.
If we sort the data by the results with 1% lower frame time, we see that the 8700K is 23% faster than the Ryzen 7 1800X and the Ryzen 5 1600X. The 1800X is roughly on par with the old Sandy Bridge 2600K and quite shocking the FX-8350. With the exception of the A12-9800, however, we see reasonable performance across the board.
If we then focus on the 0.1% data, the older 2600K and FX-8350 processors are omitted, as is the case with the modern quad cores such as the Ryzen 3 1300X and the Core i3-8100, although the higher clocked 8350K still does quite well.
Since we were so severely restricted with the GTX 1080 Ti at 1080p, I decided at least for the high-end core processors of the 8th generation for a resolution of 720p. Here the 8700K is now 11% faster than the 1800X for the average frame rate, although the 1% low data was similar, although it was pulled away with an 18% margin for the 0.1% low result.
There are high quality results and for the most part things looked pretty good. However, because the results are based on an average of three runs, they somewhat smooth out some of the frame mistakes we've seen every now and then.
Even with the Core i7-8700K and GTX 1080 Ti, we occasionally saw peaks in frame rate lag, and this was really only a problem when using the high quality settings. When we turn off GameWorks features by smoothing out the default quality performance and seeing some interesting things, we look at that.
Here are the results of standard quality at 1080p and we immediately see more consistent performance, possibly with the exception of the older 2600K and FX-8350 processors.
However, when we look at the 8700K and 1800X, we now see a much smaller performance difference. The 8700K is now 12% faster when comparing the 0.1% results compared to almost 30% faster before. The variation in the average frame rate and the 1% low result is also far less significant.
In fact, Ryzen looks very impressive with the standard quality settings because the Ryzen 5 1400, for example, offers a much more even performance compared to the Core i3-8100.
If we sort the chart by 1% low results, we can see how close it is between the top 6 processors, which include the Ryzen 7 1800X and the Ryzen 5 1600X and 1500X. Meanwhile, the Ryzen 3 1200 is comparable to the Core i3-8350K and the Core i7-2600K, which is a solid result. The FX-8350 also does surprisingly well in this group.
Why not invest the time to check the lite quality preset and this led to some very interesting results …
Typically, Intel's higher clocked CPUs run low-quality, low-quality tests because they can deliver much higher maximum frame rates. We see this here when we look at the average frame rates.
The 8th generation Core i5 and Core i7 processors were able to maximize the GTX 1080 Ti at 171 fps and were at least 11% faster than the Ryzen 7 1800X. However, you may have noticed something strange when looking at the frame time data. So let's focus on that.
Here we find something unexpected. When sorting the data according to the 1% low result, the Ryzen CPUs have the edge, even the Ryzen 3 1300X can beat the Core i7-8700K. This is not the case with the 0.1% result, but here the Ryzen 5 and Ryzen 7 CPUs still beat the powerful Core i7-8700K. This is very unusual and we have to dig a little deeper to find out why we see it.
Waiting for the actual game
These were some interesting results and should give you a good idea of the processing power you need to play Final Fantasy XV on March 6th.
We usually assume that the benchmark tool is very accurate since the developer released it beforehand so that players can prepare their systems. However, evidence has emerged that the tool may not be suitable for measuring actual / expected GPU performance for the game. What is probably an unintentional mistake in the benchmark, it renders objects that it does not need (outside the field / view, using GameWorks functions) and therefore places an excessive load on the graphics card. Therefore, it may not be accurate for GPU testing, but we believe the CPU results should turn out to be accurate.
If you have a quad-core clocked below 4 GHz, our tests will likely make you struggle with more than the Lite quality setting. The game is very CPU intensive. For example, the 8700K with 12 threads rarely fell below 40% load and spent most of our time over 50%, sometimes up to 80%. The situation is similar with the Ryzen 7 1800X, although more time was spent on the 30-40% mark. Nevertheless, the Ryzen CPUs in the lower price range with 12 or fewer threads were used very well.
We should note that 1600X and 1800X were 5-10% faster depending on the test with SMT disabled, but I didn't go into that in detail here because we doubt that many of you will disable SMT just for a bit more performance in Final Fantasy XV, but note you could if you wanted to.
The high quality settings that appear to activate all GameWorks features were somewhat worrying. Stuttering was a problem here, even with the most extreme hardware configurations. However, this may be due to the error that causes objects to no longer appear. With that in mind, AMD has announced that they won't be releasing their optimized driver until next month when the game is released.
Of course, we're keen to test the game after it's released, and hope we have a lot of settings that we can tweak for better results. We'll do a proper GPU benchmark test for Final Fantasy XV as we get closer to launch, and can confirm that it reflects the actual gaming experience.