AMD FidelityFX Super Resolution will soon receive an update to version 2.0

According to current rumors, AMD could present version 2.0 of its upscaling technology FidelityFX Super Resolution with an important change and optimization in the course of GDC 2022.

Enlarge
AMD released FidelityFX Super Resolution image upscaling technology on June 22nd.
© AMD
FidelityFX Super Resolution (read the article) is the name of AMD’s upscaling technology, which makes it possible to get higher frame numbers in modern computer games at the expense of image quality. According to their own statements, the developer of CapFrameX is said to have received access to version 2.0 of FSR demo material. The update should bring a fundamental change compared to its predecessor: the feature should no longer rely on spatial but on temporal upscaling, just like Deep Learning Super Sampling (DLSS) from Nvidia. However, the hardware requirements should not change, so FSR 2.0 should be compatible with the GPUs from all providers and do not require artificial intelligence.
AMD FidelityFX Super Resolution 2.0
According to a tweet from
CapFrameX
FSR 2.0 should bring a fundamental change. The feature should no longer rely on spatial, but on temporal upscaling, which could drastically improve the image quality together with optimized anti-aliasing. In contrast to DLSS or Intel XeSS, however, no artificial intelligence should be necessary. As a result, FSR 2.0 requires no special hardware and is compatible with graphics cards from all vendors. However, the tweet does not reveal whether there will be restrictions regarding the age of GPUs.
FSR 2.0 to come/announced soon. Saw some footage today. Very impressive. 😮
▶️ Temporal upscaling + optimized AA
▶️ Doesn’t need AI
▶️ Runs on GPUs from all vendors
▶️ Impressive performance and image qualityAMD even claims that it can be better than native. ^^#FSRGen2
— CapFrameX (@CapFrameX) March 11, 2022
On March 14th, CapFrameX published another tweet regarding the performance of FSR version 2.0. The performance gain using the example of Deathloop should decrease somewhat compared to the first version. Specifically, there is talk of the fact that the performance mode of FSR 2.0 can almost double the frame rate, with FSR 1.0 the FPS could be increased by a factor of 1.3. In return, however, the image quality with FSR 2.0 should be significantly better than the first version.
New FSR 2.0 comes at a cost. According to AMD’s slides the Performance mode can almost double (<100%) the frame rates at 4K whereas FSR 1.0 can offer ~130% more fps using the Performance mode in Deathloop.
Of course, FSR 2.0 Performance mode provides much better IQ. 1/2
— CapFrameX (@CapFrameX) March 14, 2022
The Game Developers Conference 2022 (GDC) will take place next week from March 21st to 25th. As part of this event, AMD will hold a session called “Next-Generation Image Upscaling for Gamers”. Accordingly, it is not unlikely that we will see the first demo material from AMD’s FSR 2.0 as early as next week. The GDC itself requires you to register online, but AMD’s video is scheduled to be publicly available from March 24th at 3pm German time.
AMD FSR vs Nvidia DLSS in comparison
While gamers can get higher frame rates when using FSR than when using DLSS, the image quality with AMD’s solution is usually significantly worse. The reason for this is that AMD relies on spatial upscaling for FSR. In short, this means that an image is upscaled from a lower resolution to the monitor’s native resolution and then post-sharpened. However, it is perfectly understandable that an image in Full HD resolution cannot provide the same amount of image information as an image in UHD resolution and, accordingly, the content cannot be reconstructed.
This is different when using DLSS, since temporal upscaling based on motion vectors is used here. As a result, past and future image information can be taken into account in order to reconstruct an image with a higher resolution. Ideally, an image scaled up by DLSS contains even more image information than in the native monitor resolution. However, Nvidia requires an RTX graphics card with so-called tensor cores and the implementation is quite complex for the game developer. AMD’s FidelityFX Super Resolution, on the other hand, is open source, easy to implement, and works on any modern graphics card from any manufacturer.