AMD planning a potential Next-Gen upgrade for its FSR technology

At the GDC 2022 conference later this month, AMD will discuss its “next-generation image upscaling technology” for games. Because the technology is based on AMD’s ongoing research, it could be a new implementation of AMD’s existing FidelityFX Super Resolution (FSR), however, it’s unclear exactly what is being updated.

FSR is a set of open-source upscaling algorithms developed by AMD to increase performance at higher resolutions by upscaling a lower resolution source (e.g., upscaling 25601440 to 38402160, etc.). Unlike Nvidia’s deep learning super sampling (DLSS), AMD’s current implementation of FSR does not use AI/DL-based image analysis to increase image quality, which is why it works on nearly all GPUs available today, including Intel and Nvidia’s.

However, AMD’s FSR is not flawless in its current implementation, and while it delivers on its promise of enhanced performance, the rendering quality of AMD’s algorithm falls short in many circumstances. This is especially true at higher upscaling factors, and there’s no doubt that Nvidia’s DLSS outperforms the competition. That’s why AMD’s next-generation image upscaling technology is in the works.

We can only speculate because the corporation hasn’t revealed any details regarding its next-generation image quality-improving procedures. There are several ways that AMD could improve FSR. FSR currently employs spatial upscaling, which was an attempt by Nvidia with DLSS 1.0. Nvidia eventually modified the algorithm to use a temporal upscaler in conjunction with machine learning, and DLSS 2.0 has shown to be significantly superior to the previous version.

Nvidia’s RTX series GPUs are equipped with Tensor cores, which helps it get away with the difficult DLSS calculations. These can do a lot of less precise math rapidly since they use FP16 rather than FP32 calculations. Because AMD’s GPUs lack Nvidia’s Tensor cores, it’s unclear whether this next-generation upscaling technique will operate with current and previous-generation GPUs, or if it’ll be reserved for AMD’s future RDNA 3 design.

We don’t know if AMD’s next-generation upscaling technology employs AI/DL analysis to improve performance and quality at this time. Perhaps AMD’s next-generation GPUs have hardware support for high-quality upscale methods used by Adobe Premiere Pro and Blackmagic Design’s DaVinci Resolve, but given GPU development tendencies, we’d bet on something more intelligent and less special-purpose.

Also Read:

NVIDIA GeForce MX550 discrete-GPU has a hard time performing against AMD RDNA 2 ‘Radeon 680M’ Integrated GPU

Source