A quick explanation of AMD FidelityFX Super Resolution

With FidelityFX Super Resolution (FSR), AMD has introduced a simple and effective open-source solution for improving image definition and performance in video games.

History of the main FSR versions

FSR 1 is a spatial upscaler based on the Lanczos algorithm, which requires an anti-aliased lower resolution image.

FSR 2 is a temporal upscaler based on a modified Lanczos algorithm requiring a low-resolution image and using temporal data (such as motion vectors and image history), then applying its own antialiasing pass that replaces the game's temporal antialiasing solution.

FSR 3, launched in September 2023, adds image generation and "native antialiasing". Frame generation increases the perceived frame rate of a game. "Native antialiasing", like Nvidia's DLAA, can be used without upscaling to improve antialiasing; it can also be combined with image generation and Anti-Lag (RX6000 Series) / Anti-Lag+ (RX7000 Series).

How it works: (source :AMD GPUopen)

FSR is initially a two-pass solution, but the introduction of FSR3 adds a third pass for image generation.

The first pass consists of increasing image definition via the EASU algorithm: Edge Adaptative Spatial Upsampling. This consists of enlarging the image in its entirety (spatial), in a variable way within it (adaptive), with analysis at the level of the salient points (edge).
This allows an increase in resolution while maintaining a relatively low impact on performance, thanks to the use of an improved Lanczos filter, which is quite common in image reconstruction but requires considerable computing power. To reduce the need for computing power, EASU introduces a rational approximation which enables the computing tools to produce a result as close as possible to the theoretical one.

The second pass is made via RCAS: Robust Contrast Adaptative Sharpening, which consists of applying a filter to improve image sharpness.

The algorithm adjusts the amount of sharpening per pixel to obtain a uniform level of sharpness throughout the image. Areas of the input image that are already sharp are given less emphasis, while areas that lack detail are given more. This results in greater natural visual sharpness with fewer artifacts.

The third pass generates additional images after the user interface has been added. This allows it to be considered during image generation and limits the visual impact it may have on it.

The generated image is presented directly, which then allows the real image to be presented in the middle of the next calculation, resulting in a doubling of the number of frames per second displayed.

To combat the additional latency inherent to the image generation process, AMD has a driver-level function called Anti-Lag, but this is limited to AMD GPUs, unlike FSR.