Devil May Cry 5 - Denuvo Performance Impact
Published: 11th March 2019 | Source: OC3D Internal Testing | Price: |
Performance Impact - Three Scenes Tested
To test the performance impact of Denuvo, we decided to team up Nvidia RTX 2080 Ti graphics card with our 4GHz Intel i7-6800K processor, a CPU that offer performance levels that are way below today's high-end Intel offerings. By today's standards, this isn't a monster processor, as strong 6-core CPUs can be had for well under £200 these days, many of which offer out of the box boost clocks of over 4GHz.
In this set of tests, we will avoid Devil May Cry 5's cutscenes. Here we will use real gameplay, as this will offer the best possible representation of in-game performance. Devil May Cry 5's cutscenes are much more demanding that the game itself, but their pre-defined nature is not great for CPU testing. Denuvo is said to impact CPU loads the most.
When playing Devil May Cry 5, we will use the game's maximum quality settings, which pose no challenge to our RTX 2080 Ti. Yes, we can lower the game's resolution to below 1080p and lower settings to get the highest framerates possible, but Devil May Cry 5 almost completely CPU bound at 1080p with our RTX 2080 Ti Founders Edition, so it makes no sense to lower settings or the game's resolution further.
Test Scene One
In our first test scene, which is part of the game's 10th part/stage, we play as Dante during a short combat sequence. When replaying this sequence we were able to achieve a high level of repeatability, with relatively little run/run variation. We ran this test scene both with and without Denuvo three times. These tests were conducted under Devil May Cry 5's DirectX 12 mode.
As you can see below, our testing provides a very clear result in this scene, with all results showcasing differences that are well within the margin or error. We decided to test this scene at multiple resolutions, hoping to showcase how the performance differences disappeared ad the title became more GPU bound, but as you can see Denuvo makes no little difference to this section of the game.
If you look at our 1080p and 1440p results, we can see that both resolutions appear to be CPU bound, with both resolutions offering similar performance levels.
Test Scene 2 - A double check
In our game analysis pieces, we never test a single area of a game. Simply put, that leaves too much room for error. It is not uncommon for our CPU and GPU benchmarks to use completely different scenes, as a single area is often not enough to define maximum system load in all areas.
In our second test area, which is based in Devil May Cry 5's 4th stage/mission, we move through another small combat encounter and do a little bit more level traversal. In this test area, we found that Denuvo had a performance impact at 1080p, with our 1% minimum framerates increasing from 144FPS to 149 FPS and our average framerates rising from around 170 FPS to 175FPS.
When looking at these minimum framerates, we can see a performance increase of 3.47% when Denuvo is removed. This increase is lowered to 2.66% when we look at average framerates. These increases are notable, but not what we would call transformative. When the game is already running at over 144FPS, you will never notice the difference.
Testing Scene 3 - This is where we start to see big changes
When it comes to Denuvo, the two biggest complaints that the tech receives from gamers is how it allegedly increases load times, and how it squanders CPU cycles that could be better used on gaming applications. For our third test, we decided to push asset streaming as hard as we could.
To do this we started level 4 again, ensured our character's maximum travel speed by purchasing V's sprinting upgrade and cleared several areas of enemies to give us a clean run through a large section of a level. To reduce the impact of run/run variance, we tested this scene five times both with and without Denuvo to minimise performance variance.
On average, the Denuvo-less version of Devil May Cry 5 offered the highest performance levels, but again the performance difference was minimal. These changes amounted to a 3.4% and a 4.2% performance difference in average and 1% minimum framerates respectively. In this scene, performance variance could be larger than this, hence the need for averaged results over five test runs.