NVIDIA DLSS (Deep Learning Super Sampling) is groundbreaking AI rendering technology that increases graphics performance using dedicated Tensor Core AI processors on GeForce RTX™ GPUs. DLSS taps into the power of a deep learning neural network to boost frame rates and generate beautiful, sharp images for your games. This week on Inside Unreal we will walk you through the DLSS plugin for UE4. We’ll look at first steps for setup, implementation, example projects and much more.
In further discussion regarding depth of field + DLSS. Correct me if I’m wrong, but for subtle depth of field, it seems that the image is being treated as a lower resolution image that needs reconstruction via DLSS rather than what’s intentionally trying to be blurred by depth of field. I know many don’t like depth of field in their games, but I do, particularly used with subtlety. However I’ve yet to get DLSS to play well with subtle amounts of DOF, or even falloffs distances that are smoother in nature. It tends to either remove the DOF effect by reconstructing it, or there are harsh contrasty divides between what it considers intentionally blurry versus reconstructing a detailed sharp image.
Sorry for the long winded post, I suppose what is like to know is, are there any tips or tricks to get desirable DOF with DLSS. I know you stated there are differences, that makes sense, but what have you found in ways to combat those differences to achieve the desired results?
I do have a few things to add but it may not be the answer you’re looking for. As with all things in games or realtime rendering, context usage of your content does matter. If say you were doing some very subtle DOF in your realtime cinematics maybe force it to switch to DLSS Quality mode during the cinematic. Or perhaps don’t use DLSS in certain cinematics (since you can turn it on and off when you want) and use it for everything else.
Here’s a question to ask yourself: How many scenes need DOF to look a very particular way, vs how many scenes would benefit from the framerate boost? It’s totally going to depend on your project and what’s important to you. Some games would say this is an easy choice, they barely make use of DOF and the fps boost is an easy win. Some projects won’t mind the changes in DOF behavior, DLSS works fine for them. But yeah some projects might say DLSS doesn’t work for this reason. DLSS has to fit into the current rendering pipeline for Unreal Engine 4 so this is a limitation of where technology is at right now.
Now it is true also that if you are a coder with access to the source, you can change some fundamentals about how and where postprocess effects work. I think experts in this area will tell you that can cause new problems however, and will probably also harm the performance benefits of DLSS. So this is not necessarily an easy win, but it is worth thinking about if you have those resources.
So the good news is that since we’ve identified the issue and are starting to understand it, we can take steps to improving and fixing it. DLSS is not done yet, the current shipping plugin of v2.1.5 will not be the last version. And of course software is generally improving in this area, especially as we continue to understand the issues.
Thanks for tuning in! I really wish I had a better answer on this today.
I’m looking into it, I’ll probably replace this post when I have something. I can say one of the features of DLSS is to support multiple displays and viewports on the same RTX machine, which makes editor workflows and VR possible. Ndisplay is another matter of course.
Correct. I think its fair to say we would like to support it and figure out what we can do. This is a continuing investigation for us. Keep in mind that MRQ + DLSS is all brand new as well. MRQ was added pretty recently I think and Epic considers it a beta in 4.26 (according to the documentation if I’m not mistaken). So give us a chance to catch up to new features and I’m sure we’ll get there. =)
Can we dive into that question? In what sense do you mean? It absolutely can speed up any GPU bound editor workflow, basically rendering the viewport faster. So that’s a very broad yes, but maybe there’s something more specific you have in mind?
thanks again for the live demo the other day, DLSS is very impressive, however I came across an issue that few other users also encountered( from other post) that is the defocus seems to be disappeared as soon as DLSS switched on . I wonder if we are just missing some cVar to have defocus to work with DLSS? If you can shed some lights that would be great!
Question: It appears the Unreal Editor comes up with DLSS setting OFF. I understand that I can turn it on by the viewport dropdown menu by setting DLSS Setting to Quality and it appears the Sequencer rendering uses this setting. Is there a way to tell the editor to come up with DLSS Settings = Quality by default so that it is automatically set for Sequencer rendering whenever I start up the Editor? (It appears that r.NGX.DLSS.Quality = 1 consol command sets this only for Play mode, and not for the Editor or Sequencer mode.)
Thank you, this is a great answer, unfortunately for the project, it’s an arena combat fighter, and the DOF was used to take away focus from the background set dressing and let players focus mainly on the combat within the arena, our current solution is with subtle DOF, that being said, we really want to include DLSS, so we will get creative with other solutions that solve our underlying problem, so we can still use DLSS, but perhaps maybe not use DOF for our solution. Looking forward to the future updates and improvements on the tech and implementation in UE4, and hope to see you back on another stream in the near future. Thank you for your response.