Quick test: DLSS with dynamic resolution in Avengers and Doom Eternal

Doom Eternal was recently supported ray tracing and in connection with that, DLSS with support for dynamic resolutions was also added to the game. In the past, this combination has been unique to Marvel Avengers, a game in which “no one” seems to have tested this combination.

We therefore thought to throw ourselves over the task of figuring out what the interaction between dynamic resolution and DLSS looks like. What is really dynamic is not entirely obvious, and what about performance and image quality?

A little look back at which resolutions the fixed DLSS settings correspond to may be in place before we get started:

Experiences from Doom Eternal

The maximum resolution of the dynamic resolution in Doom Eternal is set by adjusting the quality setting for DLSS – with the exception that Performance completely turns off dynamic resolution. This is then combined with activating dynamic resolution and setting a frame rate target. framerate target) for the game.

It will be easier to illustrate with an example. If the target resolution is 2,160 vertical pixels (as in 4K UHD), DLSS Quality corresponds to an internal resolution of 1,440 pixels. If we add dynamic resolution, the game engine can render the game in a lower resolution than 1,440 pixels when required to achieve the desired performance goal.

The interval that the game has to work with is the resolutions between DLSS Performance and DLSS Quality. That is, between 67 and 50 percent of the target resolution. This also means that DLSS Performance cannot be combined with dynamic resolution because DLSS Performance is already as low as the setting can go. However, it does not prevent the game from letting one activate both at the same time without it having any effect.

It is painfully obvious that this is not simply easy to convey in a sensible way to all users. Just a look at which settings are related to resolution and how they are scattered across Doom Eternal’s menu shows that there are things to work on here.

DLSSQ_76.png

DLSS Quality with 76 percent resolution scaling

DLSSB_88.png

DLSS Balanced with 88 percent resolution scaling

DLSSP_100.png

DLSS Performance with 100 percent resolution scaling

The images above illustrate where each DLSS mode “bottoms out”, that is, the lowest upscaling factor that modes can use when combined with dynamic resolution. Doom Eternal is a game with a lot of effects, which makes an outdoor image like this difficult to get completely identical to each screen. But if you think the images are very similar, it’s no big mystery, as all three in practice correspond to DLSS Performance without dynamic resolution.

Hence the comment that it is ineffective to be able to combine DLSS Performance with dynamic resolution as it is not dynamic for five öre, the interval is literally between 50 and 50 percent. Balanced gets a margin of 50 and 58 percent and Quality’s range is between 50 and 67 percent.

To construct a test scenario where the solution actually has to stretch the legs and show what it works for, we chose a Geforce RTX 3060 Ti. All settings are at the highest possible except Texture Size Pool which is reduced to Ultra.

Decisive will of course be the existing performance and what performance goal you set in the game. If we set the goal too high, everything is rendered in the lowest quality setting, which corresponds to DLSS Performance. If we set it too low, everything is rendered in the corresponding DLSS Quality.

We therefore ran the test with static DLSS set to Quality and measured average FPS where we got a measured value of about 96 FPS. Then we switched to dynamic resolution with a target of just 96 FPS.

You Might Want To Read This:   AMD Radeon RX Vega in official pictures can be found on the web

Attitude

Average (FPS)

99:e percentilen (FPS)

99,9:e percentilen (FPS)

DLSS Quality (Static)

96,02

75,19

69,14

DLSS Quality Dynamic

99,20

82,45

76,66

Difference

3,3%

9,7%

10,9%

The performance gains are relatively small, which is not surprising precisely because the range of resolutions the game scales between is small. However, it seems that it helps especially during the heavier sequences of the test, just as expected.


In the video you can see a short sequence where we play with the configuration mentioned above with dynamic resolution. To the right in the picture you can see how the resolution changes dynamically. Partly in the form of a percentage next to the resolution and also in the form of the graph next to “RS:”. Do not pull on too large gears when inspecting image quality, Youtube slaughters our 7.2 GB large source file to only 844 MB.

DLSSDRS.png

Doom Eternal’s presentation of “Performance Metrics” is exemplary in many ways. However, Just Resolution Scaling (RS) is somewhat confusing. In combination with DLSS, it does not show what percentage of the target resolution is rendered, but what percentage of the DLSS setting’s internal resolution is rendered.

You Might Want To Read This:   AMD hypers the Radeon 300 series in video clips

For example, “RS: 100%” when driving with DLSS Performance and dynamic resolution are the same as “RS: 76%” when driving with DLSS Quality. It does not feel completely intuitive or obvious.

Experiences from Marvel Avengers

As soon as we tackled Marvel Avengers, we began to understand why, in principle, no one bothered to explore these settings. Dynamic resolution in this game works slightly differently in Doom Eternal. With dynamic resolution enabled, you must set one framerate target, in this case limited to the same as the maximum refresh rate of your monitor. What is not as clear is that framerate target also means framerate cap. Of course, this adds to the total in terms of the possibilities of performance testing the situation.

Marvel Avengers offers very few settings but is still a more logical way to expose the combination of DLSS and dynamic resolution

avengers.png

Show more

While Doom Eternal sets a maximum internal resolution using a quality mode for DLSS, Avengers simply offers “DLSS Dynamic”, which in practice gives us the entire range from 50 to 67 percent of the target resolution. Although the game offers the Ultra Performance setting, it is not used in combination with dynamic resolution. Other than that, the game does not offer any kind of overlay so we can not see what internal resolution is currently being used.

You Might Want To Read This:   New test results with Gigabyte R9 290X Windforce 3X

Difficulties in exposing dynamic resolution and DLSS

Exposing dynamic resolution in combination with DLSS is not easy. On the one hand, the solution in Avengers is at first glance more logical and simple than that in Doom Eternal. On the other hand, it’s like a black box where you do not really know what’s really going on.

And perhaps it would be easiest if that was how it was described in the game’s menus, rather than having it as two separate settings. That is, the four existing quality settings were simply accompanied by a fifth “Dynamic” setting, without mixing in a separate setting.

Marvel Avengers turned out to be a mess in terms of “testability” both in terms of performance and image quality, instead we have only focused on showing how the settings are presented in the game.

The bottom line is that neither of the two implementations feels completely clean in terms of settings. But Doom Eternal’s version is clearly preferable. The best thing in this context would probably be if Nvidia developed guidelines for how these settings should be exposed to the player.

You Might Want To Read This:   The Asus ROG Strix Radeon RX 6800 XT houses "Rage" mode in the BIOS

Something to put on the wish list for the future is that the solution is not locked to Nvidia’s preset quality modes for DLSS. That is, a solution that could have scaled all the way to native resolution but also lower than DLSS Performance would be desirable.

The performance tests then?

Testing performance at dynamic resolution is largely harmless because you set the frame rate target yourself. That is, you as a user decide what performance you want to achieve. In Doom Eternal we managed to find a somewhat realistic scenario as you could see above, while in Avengers it does not work, because the frame rate target also acts as a frame rate limitation.

Another test scenario that actually worked is to play Doom Eternal with everything on Ultra Nightmare and ray tracing activated in 8K on a Geforce RTX 3090. The card can just at the limit reach stable 60 FPS with DLSS Quality without dynamic resolution, with both activated we reach almost in Port. In fact, it would be one of the few examples where we would have liked the opportunity to activate the Ultra Performance mode as it would in all probability have locked us to 60 FPS even in the worst scenes.

You Might Want To Read This:   AMD's upcoming graphics card "Big Navi" is rumored to be launched in November

Thoughts and recommendations

If you have a card with support for DLSS and have a frame rate that is exactly on the border of what you want to achieve when you run Doom Eternal with DLSS Quality, of course dynamic resolution is a nice addition. It does no miracles but it can even out the worst scenes somewhat.

We can not really say the same about Avengers, because dynamic resolution is exclusively combined with one framerate cap in the game. We also have a hard time really seeing what is happening in the game as the image quality differences are small and the performance can not be tested.

For the future, we would have liked free blow for dynamic DLSS as it is currently limited to scale from 50 to 67 percent of the target resolution. It is a game of 17 percentage points in other words and it is with the hand on the heart a little too little for it to be something to go and crave.

You Might Want To Read This:   AMD Radeon RX Vega in official pictures can be found on the web

In the best of worlds, we wanted to see a solution that had a much larger range in the style of 40 to 100 percent. It combined with a little more sensibly designed menu choices so that users, who do not want to spend a whole day analyzing the settings, can pretty easily get an insight into what they are actually doing.