Deep Learning Super Sampling is NVIDIA’s homebrew method to make games look better with a smaller performance cost than Anti-Aliasing. And supersampling. And increasing (or decreasing, depending on what you want to do) your game resolution. NVIDIA hardware owners benefit today from DLSS in two major ways:
- Games look better at a low performance cost
- Gamers with lower-resolution displays can upscale to get the most out of their games
DLSS specifically has done a great job of improving over time, thanks to the “Deep Learning” aspect. According to a report by Moore’s Law is Dead, the upcoming DLSS 3.0 coming to next-gen Ampere GPUs, will “work in any game with TAA”. This means as soon as we get our hands on an RTX 3000-series cards, a huge number of games will suddenly have supersampling. And that NVIDIA’s deep-learning system will have the time and data to crank out both better visuals and better performance.
That’s…well, that’s the future I signed up for.
The coolest aspect is games don’t have to, in theory, be made with NVIDIA’s spec for DLSS in mind. Developers can just include “standard” TAA and focus on making their games great. NVIDIA’s deep learning process will improve both the performance and display of any game, by default, with no extra work. And it’ll get better over time too.
Originally reported by TweakTown