Nvidia Introduce DLSS 2.0
At NVIDIA, we’ve been working on graphics and artificial intelligence for a long time. And now we have to opportunity to redefine real-time graphics by applying artificial intelligence to make games more beautiful and more fun. DLSS stands for Deep Learning Super Sampling, and it’s an artificial intelligence rendering technology that makes your games render almost twice as fast while preserving the detail you need to play at high resolutions. We’re able to do this because RTX GPUs have Tensor Cores, which are specialized AI processors that allow us to execute AI models really fast. In the past, it would have been crazy to think about running an artificial intelligence model for every frame of a high-resolution video, but we can do that now.
The whole DLSS journey started in 2016, when an NVIDIA researcher showed us some incredible results, obtained by using a convolutional network to fix damaged photographs. We immediately saw an opportunity to make games run faster by first reducing the rendering resolution and then you use AI to fill the missing information and generate high-quality images at much higher performance. What makes DLSS truly special is being the first AI algorithm that can do that while also generating temporally stable images, meaning without popping artifacts. For each game, we would capture very high-quality image sequences, that would be otherwise too costly to render in real time.
Then, the DLSS network would be trained to reproduce these super high-quality examples from the kind of lower quality frames a game can render very rapidly. But we wanted to do more. We wanted to make it faster, we wanted to generate sharper images, and we wanted to make it easy to train and deploy.
Introducing DLSS 2.0:
DLSS research has definitely continued nonstop. And in the last year, we basically improved every aspect of DLSS. Ultimately, reinventing a new architecture that is simply much better at doing super resolution for game rendering. With this new architecture, we have dramatically improved image quality. The images are now looking more detailed, more temporally stable, and also have less artifact, or aliasing. And there are places where it looks even better than native resolution rendering.
We’ve also drastically increased the maximum upscaling ratio. You can now render an image at 1080p, and DLSS would generate a beautiful looking 4K image for you. In addition, we’ve made the neural network run twice as fast by utilizing the Tensor Cores more efficiently. And as a result, you can now use DLSS across all the RTX GPUs and all the resolutions. Since the new DLSS support up to 4x pixel scaling, we now provide three different quality modes. Performance, Balance, and Quality. Finally, the new improved DLSS is a fully general solution, so we don’t have to train DLSS on a per game basis anymore. And this makes integrating DLSS into games easier than before.
Let's now look at some examples of the new DLSS
2X Zooming Sample..
Game Name : Deliver Us The Moon
Let’s now look at some examples of the new DLSS in action. In Delivers Us The Moon, if you zoom in on the gloves here, you’ll notice that when DLSS is enabled, we’re getting more detailed images and we’re getting higher framerate. If we look at Wolfenstein: Youngblood, you’ll notice again that we’re getting a really comparable image quality to native rendering and we’re providing a tremendous boost in framerate. If you zoom in on this particular spot, you’ll notice that with DLSS enabled, we’re actually getting better definition on the barrel than native rendering, even though we’re rendering fewer pixels. In MechWarrior 5, you’ll see the same positive results, really similar image quality to native rendering with huge boost in performance. We’re also adding the latest version of DLSS in Control, which will further improve the image quality in this already great-looking game.
Games just get more beautiful and more detailed every year. DLSS gives us new tools to try out new effects like real-time ray tracing in ways that you could never have dreamed of in the past. This is really opening the door to more beautiful, more exciting experiences for all of us gamers.