The presentation of AMD’s FSR 2.0 technology generated a lot of expectation, and the truth is that there were plenty of reasons for it. The Sunnyvale giant confirmed that this new version of its well-known rescaling algorithm overcame a major limitation. that was present in FSR 1.0, its inability to use temporary elements.
With FSR 1.0, the rescaling was generated taking into account only spatial elementssomething that simplified the workload that this technology entailed and that allowed for a very marked performance improvement even in less powerful graphics solutions, but that in the end had a very large impact on image quality.
I have had the opportunity to test this rescaling technique in several games and with different graphics cards, and I can attest that When you go down from the ultra quality mode, you begin to notice such a great loss of quality which is just not worth it.
The performance mode presents a loss of sharpness and image quality so enormous that it does not directly make sense. I delved into this topic at the time in the technical analysis of Dying Light 2, where I attached videos showing the impact of FSR and DLSS in image quality.
AMD FSR 2.0: Higher image quality at the cost of a minor performance improvement
The FSR 2.0 technology works, as we anticipated, with temporal and spatial elements, which means that it takes into account information from previous frames to offer a more precise and higher quality rescaling. This translates into a much higher result than what we could obtain with FSR 1.0 technology, this does not support discussion, but it is clear that It has not been a “free” improvement, and it is that in the end it represents an additional workload which makes FSR 2.0 offer less performance improvement than what we get with FSR 1.0, and regardless of the graphics card we use.
That’s why there’s that performance loss vs. FSR 1.0, and this occurs even on new generation graphics cards, such as the powerful Radeon RX 6800 XT and GeForce RTX 3080. I don’t want to go into technical complexities, as I want all our readers to understand it without problems. Consider that when using a rescaling algorithm like FSR 2.0, all of this happens:
- Reduced workload on the GPU due to rendering resolution clipping. Depending on the mode used, the total number of pixels can range from 67% of the original resolution to only 33%.
- Although the workload is reduced from that perspective, a new task derived from the application of the algorithm is introduced, and this consumes rendering time on each frame. If you only have to work with spatial elements, the process will be completed sooner than if you also have to add temporal elements.
- Given the does not use specialized hardware to speed up the workload represented by the algorithm, in the end the impact on performance ends up being greater than what we have, for example, with the second generation DLSS, although it is true that the difference is small, at least in Deathloop.
Realistically, we only have one game to compare to right now, Deathloop, so I’d rather wait until I can do a larger comparison to draw my own conclusions, and with my own tests. However, it is clear that by adding this layer of work with temporary elements AMD has managed to clearly improve the quality of the FSit is also evident that it has done so at the sacrifice of performance.R2.0, but
AMD has done an excellent job and has cut back, but second generation NVIDIA DLSS technology remains the great rival to beat, both for image quality and performance, at least until we see what Intel XeSS is capable of. In the attached video you can find the complete comparison of HardwareUnboxed.