NVIDIA’S DLSS Computational Imaging GPU For Graphics

NVIDIA DLSS (Deep Learning Super Scaling) is a graphics rendering technique used in Tensor Core AI processors in their GeForce RTX GPU line. It introduces that increase frame rates and renders sharper images for games and graphics applications. It’s main feature is the use of ray tracing techniques for increased output resolution. For games this means boosted frame rates and sharper beautiful images.

Many good things have been said about DLSS, but what exactly is it for? DLSS comes from the computer vision field of AI, which uses deep learning as a way to enhance and upscale images. This has plenty of significance with video games because of animated renderings. DLSS helps push pixels from digital frames to appear more like real still images with less latency and flicker, which appears smoother on the eyes.

Sharper and more realistic graphics renderings using DLSS (Source Nvidia)

DLSS Overview

Let’s first discuss the Ampere architecture that will be a big part of DLSS implementation on the RTX series. Ampere is an improvement on Turing, but does not necessarily make it faster. The main improvement is in the techniques with accelerated concurrent processing. From Ampere we get new designs in the RT Cores and Tensor Cores as well.

The architecture uses a third generation Tensor Core design. Each Tensor Core is faster than previous versions, but with half the SM (Streaming Multiprocessors). It delivers the same performance at minimum, or 2x the performance in some workloads.

What does DLSS really do?

It renders images at a lower resolution to provide a performance boost, then applies various effects to raise the resolution. This is what leads to higher frame rates without a big loss in visual fidelity (i.e. less artifacts and aliasing). All this was done with the help of Nvidia’s supercomputers with trained data sets, to optimize the best performance that the system can provide.

RTX Series GPUs

Nvidia implements DLSS in their RTX line of GPUs which makes use of the Ampere 2nd Generation RTX architecture. This utilizes the new RT Cores, Tensor Cores, and streaming multiprocessors for realistic ray-traced graphics. This provides AI accelerated performance for realistic and immersive gameplay.

RTX hardware specs (Source Nvidia)

These cards also claim to use the fastest GPU processors. In performance testing provided by Nvidia, the frame rate can be increased from 39 fps to 107 fps with DLSS enabled (demo with Metro Exodus game). When you boost frame rates, the transitions are smoother and the graphics appear more sharp and crisp. This is ideal for 4K resolution displays and higher.

With DLSS enabled, higher frame rates are possible (Source Nvidia)

Frame stability is one thing that these cards can provide. This is in part a result of the performance from AI rendering cores. This helps to maximize graphics settings and resolutions that are able to support up to 8K. Tests from Nvidia reveal that playing various titles showed performance increases on an RTX 3080 GPU at 3840 x 2160 resolution (4K). On Cyberpunk 2077, the performance increase was up to 333% (in terms of frame rate).

Data from an RTX 3080 at 3840 x 2160 resolution with Max Graphics settings, DLSS Performance Mode, and Ray Tracing enabled. (Source Nvidia)

A benefit of having an RTX GPU is how it brings out the tiniest details in a visually stunning form. You get realistic shadows and reflections, along with detailed texture and color gradients. The lines are blurred between vector and raster graphics. It is clearly not a photograph, but it looks realistic enough to fool your senses.

DLSS with Ray Tracing enabled (Source Cyberpunk 2077)

Overall the RTX series GPUs have more CUDA cores, more memory bandwidth and more RT cores. This is what makes it one of the fastest cards for games with ray tracing enabled.

The DLSS SDK

A software developer’s kit or SDK (release 2.4 as of posting) is provided for DLSS. It requires Windows 10 64-bit with DirectX End-User Runtimes (June 2010) or a Linux kernel 2.6.32 and higher. The development environment recommended is Visual Studio v15.6 or later versions.

For the engine requirements:

  • DirectX11, DirectX 12, or Vulkan based
  • The raw color buffer for the frame (in HDR or LDR/SDR space).
  • Screen space motion vectors that are: accurate and calculated at 16 or 32 bits per-pixel; and updated each frame.
  • The depth buffer for the frame.
  • The exposure value (if processing in HDR space).

Once you have the SDK, you will also need the UE and Unity plugin from Nvidia’s website.

DLSS Support Required

Not all games will play with DLSS support, even if you have an RTX GPU. It depends on the title and the game developers who support DLSS features. This is available in games from Cyberpunk 2077 and Fortnite. Geekflare provides a list of DLSS supported games, and there are new ones being added.

Synopsis

The GPU alone does not boost overall performance. Another way is to complement the GPU with a monitor or display that can support higher frame rates and faster refresh rates. You wouldn’t want to have your GPU overwhelm your monitor, if it cannot handle the stream of packets. That is why G-Sync gaming monitors would complete the solution to any lags that might occur.

Another way to get the most of DLSS is to use GeForce Game Ready Drivers. These are quality tested components from game developers, that Nvidia has validated. They optimize game settings so gamers get the performance that is suitable for their configuration.

There are other GPU makers who will implement a form of their deep learning architecture (e.g. AMD FSR). This makes graphics much better from a visual perspective, and for gamers this creates a more immersive and realistic user experience that makes playing all the more interesting.

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Vincent Tabora

Editor HD-PRO, DevOps Trusterras (Cybersecurity, Blockchain, Software Development, Engineering, Photography, Technology)