AMD's FidelityFX Super Resolution leverages standard upscaling techniques and appears to offer worse image quality than DLSS 2.0
AMD is set to launch FidelityFX Super Resolution later this month as a competitor to NVIDIA's AI-based DLSS upscaling solution. During a recent announcement, AMD stated that FSR would not be using machine learning like DLSS. This raised questions about image quality, since conventional upscaling methods are limited in terms of the information they can reconstruct from a lower-resolution frame.
A recent comparison by WCCFTech of AMD's own side-by-side FSR and native footage for Godfall indicate that image quality might actually be an issue for the new upscaling tech. The comparison highlights how high-frequency detail on textures and transparencies - like the leaves on trees - is lost in the FSR footage, compared to native resolution rendering. The difference is stark enough that it's apparent, despite the fact that the footage was from YouTube and heavily compressed.
If the Godfall comparison reflects FSR's general image quality across games, the feature might not be as useful as AMD touts it. DLSS 2.0, in contrast, leverages deep learning to infer details in lower-res frames that simply wasn't there. This is what allows current iterations of DLSS to deliver image quality that's as good as or better than native, without the performance hit. If FSR turns out to be little more than a souped up version of FidelityFX upsampling, it might not be worth the performance hit compared to conventional resolution scaling.