CONTROL Gets DLSS 2.0, How Does It Look And Perform?
CONTROL not only get a new DLC, The Foundation, this week but also got a fresh coat of paint in the form of NVIDIA's DLSS 2.0. When CONTROL first launched it was one of the only games to go all-in on RTX features with ray traced shadows and reflections as well as DLSS support, making it quite ambitious and a blast to test for myself. But, the initial implementation of DLSS was version 1.X and not the same as the original that shipped out with games like Battlefield V and Metro Exodus. Rather than using the Tensor Cores, it used a form of temporal reconstruction to improve the image, and while it was a decent implementation we now see the game move on to a true DLSS 2.0 version.
With the DLSS 2.0 injection, we're back to utilizing the Tensor Cores to reconstruct the image in a way that much better represents the native image you would get at full resolution. Added into the mix is an extra option for the base resolution as the DLSS 1.X only gave you two variables to work with, now with DLSS 2.0 you get three. The results should be, according to NVIDIA, much better than what we saw in the past and that's what we're here to take a look at.
I do want to point out that we're having to work with a mix of old and new so we had to run the game in one version and capture the information and images before moving to the updated version with the latest drivers that implemented the new DLSS 2.0. I know that MechWarrior 5 also got the DLSS 2.0 update and Mark at OC3D did a great job covering it, but I wanted to focus on CONTROL because it can show us the changes from one version to the next rather than no DLSS to the new DLSS 2.0.
The series of images here are captured using GeForce Share with an RTX 2080Ti at 4K in an effort to get the highest resolution image for inspection that we could get. But remember there are some slight differences between the two versions as I had to update the game and drivers to get DLSS 2.0 functional. As it replaces the options for the old DLSS there's not a way to flip back and forth, so we had one shot to get this in. Unfortunately, I had already updated the game without realizing it since we were on a 'Review' version of the game, so we had to pull older images that I already had archived and use those as a point of comparison. Because of this some assets are moved/missing/added, do not mistake these changes as DLSS 2.0 'moved objects' because it didn't.
Now with those images, you'll be able open them up in their full resolution to inspect them more closely, but I would like to take a couple of side by sides that I took the liberty of zooming in greatly on to take a look at some of the more noticeable issues of the past DLSS implementations and show the impact of moving to this new solution.
The foliage of the tree in the middle of the cafeteria was a point of contention in our original look at the game in article and video form. It seemed to be the area of greatest detail loss in that case, but not so much anymore. When zoomed in we can see the leaves and the small branches become more defined and sharp, looking past that we can see the tiles on the wall have a much less soft look to them, even over the native image.
Another place I would like to bring some focus to is distant text. You can see a very clear impact in the original DLSS 1.X implementation where the text on the sign is much muddier and even the small square sign over in the upper left-hand portion of the image gets impacted. But, with DLSS 2.0 the image is sharp and clear all around.
Performance And Conclusion
Testing a game like Control presents its own set of issues since it's a non-linear game, but primarily indoors and does not have an in-game benchmarking utility we had to set up one for ourselves. Just because the game takes place indoors doesn't mean there aren't large scale rooms, one being an insanely large furnace room for example, along with a plethora of geometry in each part of the Oldest House. We decided to go with a run from the Central Executive room through the Cafeteria and just into the Dead Letters room. Along this was we have plenty of varied light sources and surfaces to give an indicator of game performance, although we lack in combat sequences in this run the framerate doesn't appear to be too heavily impacted during fight scenes, that is something to take into consideration.
For our testing here we had the game set to the default HIGH settings and for the Ray Traced features we also selected HIGH (All RT functions enabled) in order to see just how the game handled the new improved visuals from the DLSS 2.0 implementation.
Once we had the results from 3 runs, after discarding an initial burner run for loading purposes, we took the average of average frame rates as well as the 99th percentile results from the run. We report our performance metrics as average frames per second and have moved away from the 1% and .1% reporting and are now using the 99th percentile. For those uncertain of what the 99th percentile is, representing is easily explained as showing only 1 frame out of 100 is slower than this frame rate. Put another way, 99% of the frames will achieve at least this frame rate.
|Graphics Card||GeForce RTX 2060 SUPER FE|
|CPU||Intel Core i9-9900k @ 5GHz|
|Memory||32GB Mushkin Redline DDR4 3600|
|Motherboard||EVGA Z370 Classified K|
|Storage||Kingston KC2000 1TB NVMe SSD|
|PSU||Cooler Master V1200 Platinum|
|Windows Version||1909 with latest security patches|
|Driver Version||GeForce Driver 445.75|
The RTX 2060 SUPER was used for 1080p because this is the class of card typically under the most scrutiny for its DXR performance and visuals as well as the 1080p resolution being one of the more difficult resolutions for the older Tensor Core based algorithm to work with. But we see a solid visual improvement over the older method and the performance is fine. We were told to expect the performance of the DLSS 2.0 to be a bit down from the original 1.X implementation in CONTROL and that was what we saw in the higher quality options. However, the visual improvement and consistency, especially in motion and around a metal chain link fencing the ever so slight performance penalty is worth it.
We checked in with the updates for Wolfenstein Youngblood in the early part of February and found the use of DLSS there to be much better than what we saw early on. We only later found out that that game was using the DLSS 2.0 revision and it only makes sense that it did the same increased fidelity while using it here with CONTROL. The new ability to rapidly move on DLSS 2.0 with no longer needing to be designed per game nor relying on DirectML might give developers the confidence they need to move forward with supporting DLSS 2.0 in their upcoming games.