Deep Learning Super Sampling (DLSS)

Kirk Ouimet
3 min readSep 28, 2020
An example of how a post-processing effect (anti-aliasing) can make an image more visual appealing

Background

I started playing computer games when my Dad brought home our first computer, an NEC with a 133 MHz processor. The first game I played was Wolfenstein 3D, created by John Carmack and the team at id Software.

As a kid I didn’t think too much about how how fast a computer could be until I started playing Quake 2. Our computer didn’t have a good video card, and I remember playing Quake 2 using software rendering at a resolution of 320x240. We had 56K dialup which would give me pings of around 180 to 280 milliseconds. It was in this hellscape of performance that my love for first-person shooters was born.

Later on I realized that I *needed* a better video card. I finally ended up getting two Voodoo2’s by 3dfx. I can still remember how good it felt to turn up the resolution to 800x600 and move my mouse around, feeling the smoothness.

Dual Voodoo2’s in SLI

Times have changed quite a bit and the graphics settings menu in video games have grown from a few options to 10+. Every time I launch a new game I do the same routine of going through each graphics setting, applying it, walking around a bit, changing it, and rinse and repeat until I get what I feel is the best look with maintaining 60 frames per second.

DLSS

I recently loaded up Fortnite and found a new graphics setting to play with, DLSS!

Deep learning super sampling (DLSS) is by far the coolest graphics setting I have seen in my 20+ years of gaming. It is such a clever and huge win from Bryan Catanzaro and the deep learning team at NVIDIA.

The premise of DLSS is this — can we take a low resolution image and feed it into an artificial…

--

--