AMD announces a second version of its FidelityFX Super Resolution. The updated upscaling technique will be available in the second quarter of this year and uses temporal upscaling, just like Nvidia’s DLSS.
Although AMD has yet to reveal few details and refers to a presentation it will give next Wednesday on the GDC, it becomes clear from the advantages of the technique mentioned that temporal upscaling is involved. In doing so, information from previous frames is used to sharpen the current frame. The basis of the technique is therefore more comparable to Nvidia’s DLSS than to the first generation FSR, which uses a spatial upscaling algorithm.
According to the chip designer, FidelityFX Super Resolution 2.0 offers image quality comparable to, or even better, native rendering. In addition, the quality of the anti-aliasing would have been improved compared to the existing FSR. However, unlike modern Nvidia GPUs, the Radeons do not have separate cores for machine learning calculations, so the upscaling has to be done on the normal GPU cores. In theory, this makes it slightly less efficient.
The announcement of FSR 2.0 comes along with a driver update that adds Radeon Super Resolution. Like the first generation of FSR, FSR 2.0 will require game integration and thus be compatible with far fewer games. RSR works at driver level and therefore for practically every game, but cannot render the UI elements separately from the in-game images, so that the end result is less beautiful. In addition, RSR required a relatively modern AMD video card, while FSR 1.0 was compatible with many more cards. The exact system requirements for FSR 2.0 have not yet been disclosed.