NVIDIA Teases the Future of DLSS with Improved Image Processing in Control

Tsing

The FPS Review
Staff member
Joined
May 6, 2019
Messages
12,595
Points
113
Initial reactions to NVIDIA's Deep Learning Super Sampling (DLSS) were lukewarm, but the technique's potential is beginning to unravel as company engineers refine their AI research models and develop better algorithms for greater image fidelity and performance. Their latest advancements can be seen in Remedy's newly released third-person shooter, Control, which improves upon the blurriness of prior implementations.

While there are still plenty of improvements to be made (check out the fire comparison below), NVIDIA stresses the importance of DLSS and its ability to create "super resolution" for ensuring that ray-traced titles can be played at acceptable frame rates in 4K. Turing's RT Cores are just one part of the recipe for maximum visuals.

The increasing computing demands of next generation, ray-traced content requires clever approaches, such as super resolution, to deliver great frame rates. The new DLSS techniques available in Control are our best yet. We’re also continuing to invest heavily in AI super resolution to deliver the next level of image quality.
 
Does "super resolution" = "native resolution" yet?
Unless it can beat the quality of native I dont see the "super" having any worth.
NVidia please quit exaggerating, this is in effect a lie because instead of being better that what we expect, it is worse.
Exaggeration like this is bringing standards of expression down, and with it, trust.
 
I agree that there have been some questionable claims and screenshots may fail to accurately capture all relevant information in comparisons. I have, however, experienced some impressive results with SOTTR and Metro Exodus.

Metro was a train wreck at launch but improved by leaps and bounds. Play it on a 4k panel that is at least 65" with RT Ultra and DLSS on and it looks great. I did read an article somewhere that said playing it 1440p with DLSS looked horrible but I'm not sure when they did their testing. Only downside is that HDR is still flubbed and doesn't seem likely to be fixed. I tested it in 27" 1440p/G-Sync just so I could 60+ fps but I did see some degradation of IQ with DLSS.

SOTTR had the usual weekly, monthly, driver updates, patches that eventually got it to a 'finished' status. Played this one exclusively in 4k/HDR. This game was interesting in that switching between DLSS and AA(smaax4) to me only seemed to shift dots or patterns. I wouldn't say either was better than the other but DLSS delivered higher FPS by using the tensor cores and relieving the main cores for other duties. Wasn't always able to hold 60fps with RT Ultra but able to when underground and many places above ground. I use this game as a reference when friends come over and I want to show off the technology and all are in amazement at the realism of the lighting, shadows, and environments.

Haven't got Control. Too many mixed reviews about game play and bugs. I also read about this at DSOG and they too mentioned how fire lost quality but also pointed out that numerous things actually improved with DLSS. I think "Super Resolution" is nothing more than an attempt at some kind of branding. Wonder if NV has used Super in anything else recently? I think they've been watching too many super hero movies.
 
I think this is nice. I remember when RT cards came out that the very idea of playing a game at 4k with Ray traycing was laughed at. Now with the same hardware and updated logic/drivers we are able to achieve playable experiences at 4k.
 
For anyone interested in some in depth RTX testing of this game: https://www.techpowerup.com/review/control-benchmark-test-performance-nvidia-rtx/

They got some nice slider image comparisons but do report that the game takes about 40-50% fps performance hit with max settings. I also did a little research and found out the game does not support HDR. Kind of a shame. In my experience RT and HDR really enhance each other. Regardless of the performance loss they do strongly recommend playing the game with max settings.
 
lostin3d said:
For anyone interested in some in depth RTX testing of this game: https://www.techpowerup.com/review/control-benchmark-test-performance-nvidia-rtx/

They got some nice slider image comparisons but do report that the game takes about 40-50% fps performance hit with max settings. I also did a little research and found out the game does not support HDR. Kind of a shame. In my experience RT and HDR really enhance each other. Regardless of the performance loss they do strongly recommend playing the game with max settings.
Click to expand...
Great review for the most part but its not at all clear if RTX was used for their benchmarks on capable cards, there is no mention.
For the results to have any meaning, RTX must be off, otherwise the comparisons with other cards as shown are invalid.
That means they dont have any fps results for RTX despite the article title saying "RTX Performance Analysis".
Most odd.

Also no DX11 results (or did I miss a mention why they didnt show them?).
 
Last edited:
Go back through the article. There are results showing RTX further in the article.
 
Grimlakin said:
Go back through the article. There are results showing RTX further in the article.
Click to expand...
I can only see RTX screenshots, no performance chart.
The Performance & VRam usage page only shows RTX off.
Not that it matters, there are links to other reviews in the user comments.
 
AntiQuark said:
I can only see RTX screenshots, no performance chart.
The Performance & VRam usage page only shows RTX off.
Not that it matters, there are links to other reviews in the user comments.
Click to expand...

Ok I'm totally scratching my head here. There was this article I read with Directx 12 no RT for AMD and Nviida cards... a bunch of them at high detail at various resolutions. Then later a bunch of screenshots showing the RT/non RT difference, and then performance with all RT options turned on across various cards at various resolutions.

I feel like a jerk for calling you out on that. I'll see if I can find the link and come edit this response with it.

It was the techpowerup link that @lostin3d posted. Sorry this post should have included a link to that one as it was a worthwhile look.
 
Grimlakin said:
Ok I'm totally scratching my head here. There was this article I read with Directx 12 no RT for AMD and Nviida cards... a bunch of them at high detail at various resolutions. Then later a bunch of screenshots showing the RT/non RT difference, and then performance with all RT options turned on across various cards at various resolutions.

I feel like a jerk for calling you out on that. I'll see if I can find the link and come edit this response with it.

It was the techpowerup link that @lostin3d posted. Sorry this post should have included a link to that one as it was a worthwhile look.
Click to expand...
lol, that is the one I am talking about, it has no RTX benchmarks :)
 
AntiQuark said:
Great review for the most part but its not at all clear if RTX was used for their benchmarks on capable cards, there is no mention.
For the results to have any meaning, RTX must be off, otherwise the comparisons with other cards as shown are invalid.
That means they dont have any fps results for RTX despite the article title saying "RTX Performance Analysis".
Most odd.

Also no DX11 results (or did I miss a mention why they didnt show them?).
Click to expand...

My mistake. A little too tired when I was reading that one. The only significant mentions of RTX testing was on page 4 and conclusion which was screenshots with slider comparions including FPS but that only included the 2080. On the conclusion page they mention it in the closing paragraph but no real charts. I took it for granted because their charts are usually more thorough and in dept but TBH RT/DLSS has really put reviewers thru the wringer with all the extra info to track.

The charts they have on page 5 are all RTX off. They do have fps on the screenshot page but it's not in charts but instead just in the top corner of each screenshot and only for a 2080.
 
Grimlakin said:
I think someone must have slipped me something because I totally remember seeing charts that are not present now... I... I have no excuse.
Click to expand...
I think we both had some of that because I thought so too but just went and checked and nothing.
 
I'll do some looking around and see if I can find a more detailed review. I know I saw one at OC3D but haven't checked it yet.
 
O.k. so just finished reading the "Control RTX/Tay Tracing PC Analysis" over at OC3D. I recommend checking it out. They actually have 2 different articles. One recently released for the game as a whole along with some optimizing tips for users and the RTX analysis. The RTX analysis goes into depth with comparisons of DLSS on/off and RT on/off along with settings and resolutions of each. Only one chart that I linked above. They compare a FE edition 2060 and 2080TI. Below I only reference the 2080TI numbers and 4k resolutions.

This game does have something I think is unique as I haven't heard or seen another do yet. You can set the resolution that DLSS is upscaling from. If you want your best 4k performance for using RT high then DLSS 1080p is the best bet but there's significant quality loss. That will get you closer to 60 fps but still shy of it. Using DLSS to upscale 1440p to 4k will give the least visual IQ loss but brings performance to 30-40 fps. This IQ comparison is similar to what I've witnessed with Metro Exodus. In 4k w/ DLSS the difference was more like shifted dots than clarity(mostly) but at 1440p it became much more apparent.

Highest frames per second at 4k occurred with RT off, of course, and DLSS at 1080P. This got you to around 100fps but it seems pretty pointless to me with all the compromises. DLSS 1440p brought it down to 58-70 fps which is a pretty solid 60 FPS experience but again, RT off. I could see myself bouncing between this and max OC settings for the card to use RT High/DLSS 1440p. A properly OC'd 2080TI will produce a bit of heat and noise while potentially gaining 5-10 fps. Coming from 30-40 that would be a helpful improvement.

I have to give props to OC3D for all the effort they've put into testing this game. They also mention how it'll likely be 1-3, or more, card gens before anyone can really play this game at max settings but still praise it for the IQ RT implementation. That's the 2nd site I've seen do that now. It also the 2nd time I've seen a site do 2 sets of reviews for game. One just for the games as a whole and another just for RT. [H]ard did one for something before shutting down but I can't remember which game it was.
 
You must log in or register to reply here.
Become a Patron!
Back
Top