AMD today announced that a new version of the resolution upscaling technology will be available in the second quarter of this year.
FSR 2.0 will be a temporal resolution boost, moving away from the spatial methods used in the first iteration of FSR. The technology is said to provide better image quality compared to native resolution. It will also include anti-aliasing, which will greatly improve quality over FSR 1.0.
Unlike NVIDIA DLSS and Intel XeSS, AMD FSR 2.0 will not rely on machine learning algorithms, meaning no need for Tensor Core or XMX acceleration. The technology should work on all GPUs. FSR 2.0 will also be supported by NVIDIA GeForce and Intel ARC GPUs, according to ComputerBase, but AMD has yet to provide details.
Here is the list of changes in FSR 2.0:
- Provides image quality that is equal to or better than the original image using temporal data.
- Enables high quality anti-aliasing.
- Better image quality than FSR 1.0 at all quality presets/resolutions.
- Requires no special machine learning (ML) hardware.
- Improves frame rates in supported games on a wide range of products and platforms, both AMD and some competitors.
AMD provided comparison screenshots from a game called Deathloop. They show performance and quality modes compared to Native 4K and FSR 1.0.
Scaling technologies often look better on static images. Therefore, AMD also provided a demo video from the Deathloop game, but be aware that this is a rather slow video:
AMD will share more details about FSR 2.0 during its GDC 2022 keynote on March 23rd.