NVIDIA DLSS – With great graphics comes great performance

If you’re gonna be dubbing yourself a bonafide Gamer with a capital G, naturally you’d be wanting that shiny maximum settings, ray tracing, high frame rate, the whole works. Here’s where NVIDIA comes in, with their Deep Learning Super Sampling (DLSS) technology. With over 55 games and counting, including game engines, supporting DLSS, it begs the question: how does it work?

HOW IT WORKS

The simple way to describe NVIDIA DLSS would be to call it a deep learning AI algorithm, where it takes a lower resolution image, then generates an upscaled image based on what it thinks it should be. For your pictures, think of it as being amazingly detailed, with a really tiny file size in comparison, as a very rough idea. DLSS applies that across the board, where the game you’re running doesn’t have to start at its best quality, but technology helps it reach a higher level. You probably won’t expect any random blade of grass to have the same loving detailing as Norman Reedus’ face in Death Stranding, but DLSS means that both things could be equal in quality.

In processing the inputs, DLSS also takes multiple samples of the same pixel to deliver better detail and edge quality. The output comes from four different inputs:

• The base lower resolution image from the game engine
• Motion vectors of the image from the game engine, which tell the algorithm the direction objects are moving in from frame to frame
• High resolution output of the previous DLSS-enhanced frame
• A large data set of 16K resolution ground truth images of different game content used in training the AI network

The technicalities sees the algorithm predicting pixels: 4K DLSS output is 8.2 million pixels, while 1080p is 2 million pixels, requiring an additional 6.2 million pixels. Drawing from a database of high resolution images, it learns how to best turn that lower quality frame to a much better looking one. All this is done without taxing your PC, for the best of both worlds: no performance loss while enjoying crisp graphics.

For comparison’s sake, take NVIDIA’s own Image Sharpening solution (NIS). NIS is a more traditional upscaling method that does not utilize AI, therefore not having that database that it may learn from. As such, NIS can only average an existing frame’s pixels and throw on a shader: while the result can certainly look good, it cannot reach the heights that DLSS can offer. As the AI continues to learn and improve, it’ll be able to better process more complicated pixels, like things in motion or fog effects, at a speed where hand-coded AI just can’t match.

DLSS AND YOUR GAMES

Fortuitously enough, since its initial introduction in 2019, NVIDIA DLSS is currently in its 2.0 iteration, doubling its efficiency on the Tensor cores it runs on. It’s also customizable, for more user choice and correspondingly bigger performance boosts, just to pick at a couple of its improvements.

In DOOM Eternal‘s case, your machine won’t be having a meltdown when the gore and sparks quite literally begin to fly when you begin revving up your guns and the demons start swarming to meet their doom. Maybe you just want to preen in the ray-traced surfaces, admiring the Doom Slayer’s physique in Photo Mode. Those are options open to you.

Join Us!