Due to their ability to get playable frame rates out of even relatively weak and old graphics cards, upscaling techniques such as AMD FidelityFX Super Resolution ("FSR") and Nvidia Deep Learning Super Sampling ("DLSS") are currently the talk of the town among gamers and developers alike. But where exactly are the similarities and differences? We have taken a look at the topic of upscaling using the latest AMD FSR 3.x and Nvidia DLSS 3.x offshoots as examples and explain them in detail below. AMD FSR 3.x goes back to AMD FSR 2.x, while Nvidia DLSS 3.x goes back to the first generation at its core. The first generation of AMD's FSR, on the other hand, was still a so-called "spatial upscaling," which had to manage entirely without a temporal component.

Nvidia created its Deep Learning Super Sampling (DLSS) technology to help improve performance in games when you turn on ultra-strenuous ray traced visuals, all the way back when both ray tracing and DLSS were introduced alongside the GeForce RTX 20-series. DLSS 2 greatly improved the visual quality of upscaled images, while DLSS 3 added AI-generated frames to boost performance even more. Now, Nvidia returns to DLSS's ray tracing roots with DLSS 3.5, introduced today at Gamescom in Germany. While DLSS 3 boosted performance, DLSS 3.5's "Ray Reconstruction" aims to improve the visual quality of upscaled, ray traced games, specifically by turning Nvidia's AI models on a critical process called "denoising." Ray tracing is limited by the number of rays a GPU can "cast" into a given scene, to create the data needed for the realistic lighting effects.