AMD FidelityFX Super Resolution, review: higher performance for your games

AMD FidelityFX Super Resolution (FSR) is the image scaling technology offered by AMD for all GPUs, an interesting option to squeeze the full potential of our hardware.

Thanks to AMD’s algorithm system with FSR we can obtain a good quality image from a lower resolution that requires less graphics power. This AMD system differs from NVidia’s DLSS in that it does not use AI to improve graphics quality, something that makes it less demanding and more compatible with all GPUs, but it also does not offer such good results.

How AMD FidelityFX Super Resolution works

The basic concept of AMD FidelityFX Super Resolution (FSR) is to offer a higher resolution image from a lower resolution, the game consumes less resources and we will have a better performance. Depending on the selected scaling level we will have a worse or higher quality.

The FSR algorithm has two phases, first it performs a scaling pass to a higher resolution, with an edge detection component that works to generate a good outline, the goal is to improve the classic bicubic filtering. As a second stage a sharpening filter is applied which appears to be basically AMD’s CAS (Contrast Aware Sharpening). CAS contrast filtering at native resolution, most of the time already improves the image quality by itself and is relatively lightweight.

Officially supported GPUs are the AMD RX 4xx series up to the present including the Ryzen 2xxx APUs with integrated GPU, also the NVidia GeFroce 1xxx up to the present, although it may work on lower models. It also works on integrated Intel GPUs where if the game has that setting it can be used without limitations by meeting the basic requirements.

To be able to apply these functions in the supported games we will need an API compatible with our GPU with support for DirectX 11, 12 and Vulkan as well as drivers higher than version 21.6.1 if we use an AMD GPU.

In addition, this optimization system can be used in other applications thanks to its open nature. We can see more details of this open source solution on the GPUOpen.com website, where this technique and others based on FidelityFX are explained.

Scale and performance factors

The FSR system offers us in the supported games different pre-configured performance modes. As we can see in the table below we can choose to achieve high resolutions from different scales from a lower resolution. The higher the scaling factor the more image defects it will generate, depending on our need for quality or performance improvement we will choose one scaling factor or another.

• Example: Output resolution 1080p in Quality mode the game is internally rendered at 720p.

AMD FidelityFX Super Resolution supported games

The list of games compatible with AMD FidelityFX Super Resolution (FSR) is growing every day thanks to the fact that it requires few requirements from developers and there is no need to train the AI as with DLSS. Moreover, we can use it with any GPU, including NVidia, which makes its adoption much more attractive. Apart from these games with official support there are other applications such as emulators or SteamVR where they have already begun to use this technique to get more performance.

• See complete list of games supported by AMD FidelityFX Super Resolution.

AMD FidelityFX Super Resolution Performance Tests

We test several computers with the AMD FidelityFX Super Resolution (FSR) system, as we can see the results are quite good especially if we scale to 4K resolution, if we have as output target 1080p the margin is lower as we start from less information for scaling. In any case the scaling figures shown in AMD graphics are quite correct and performance varies between games as is logical. We would like to have even more control over the Super Resolution process or to have an automatic mode, but for the first version of this system the result is positive.

• For our benchs we used the Riftbreaker demo that allows you to change and test FSR comfortably.
• You have to launch it from Steam, not from the desktop icon to access settings and benchmarks.

Benchs with discrete GPUs

The image quality in Quality and Ultra mode is very good, there are no errors in moving scenes and if we are missing a few frames to have a more fluid frame rate is certainly interesting to activate them. The Balance and Performance mode gives us a slightly blurrier image, but if we need performance it is something we will have to live with. In any case it seems a better solution than simply lowering the rendering percentage from the settings as many games offer.

If we compared with NVidia DLSS we can say that it stays between DLSS v1 and DLSS v2, something quite positive considering that we are only dealing with the first version of AMD FSR and that it does not require exclusive hardware.

Benchs with integrated GPUs

We test the AMD FSR system with integrated GPUs, undoubtedly this is a very important resource to take into account to get those extra frames per second and have a better gaming experience. As it happens in dedicated GPUs we can see how the performance scales correctly, we test with 1080p output resolution which is the usual in this type of GPU. In the results we can see that we do not have such substantial improvements over dedicated GPUs, but it is still a very good alternative to get more performance and better image quality on screen.

Conclusions

As we have seen, AMD FidelityFX Super Resolution (FSR) is a very good alternative to get more performance from our games without compromising image quality. In games that offer this alternative it is certainly worth using it if we need a few extra frames to have a smoother experience. In Quality modes the loss of image quality is not critical, in Balance and Performance modes the loss of image quality is noticeable, but it is an alternative if we need to squeeze the performance of the hardware.

Thanks to the open philosophy of AMD FSR we can use this system in all types of hardware, something that will undoubtedly call more developers to adopt it. For now the implementation of this system is not as widespread as it should be, but we have more and more projects integrating this technology. Compared with NVidia DLSS the results are somewhat worse especially if we compare them with DLSS v2, but we must take into account that it is a filter without AI and does not impose specific hardware.

After testing AMD FidelityFX Super Resolution we have to say that we find it a very interesting and positive solution. In these times of uncontrolled hardware prices it is certainly appreciated being able to get more performance out of our current hardware without the need for a large investment.