Conclusion
First of all, we'd like to congratulate AMD on their new release, which not only positively impacts gamers with Radeon cards, but also people with GeForce and Intel Arc hardware. The ability to run FSR Frame Generation independently of the actual upscaler is huge. It means that FSR Frame Generation has the potential to become the de facto frame generation solution for all gamers, because it works on all hardware—AMD, NVIDIA and Intel. Especially for gamers with older NVIDIA GeForce cards this is a godsend, because NVIDIA's own DLSS 3 Frame Generation technology requires a GeForce 40 card.
To alleviate any concerns over the GPU hardware used, we tested FSR 3.1 upscaling and Frame Generation not only a GeForce RTX 4080 GPU, but also a GeForce RTX 3060 and Radeon RX 7900 XT, to see how FSR 3.1 upscaling and Frame Generation would perform on different GPU architectures—the results were identical.
Image Quality
Unfortunately, there are some major issues with the FSR 3.1 upscaling implementation in Horizon Forbidden West. The first problem is that FSR 3.1 upscaling has significantly worse ghosting and smearing artifacts on small flying objects, such as snow, dust or flying birds in the distance when compared to FSR 2.2, across all resolutions. To be fair, the ghosting and smearing artifacts were also an issue with FSR 2.2, but only at lower output resolutions such as 1080p and 1440p, and to a much lesser degree—with FSR 3.1 they are visible even at 4K FSR 3.1 "Quality" mode. The amount of excessive ghosting and smearing is very similar to what we observed with XeSS 1.2 in our initial launch day review, but the problem was completely fixed when the game was updated to use XeSS 1.3, suggesting this is a problem with the game engine, not with FSR itself. We hope that a similar fix releases soon to benefit those wanting to use FSR 3.1. Another issue is related to how the sharpening filter works. FSR 3.1 now uses different amounts of sharpening, even when the sharpening slider is set to an equal value for all upscaling and antialiasing solutions. Compared to FSR 2.2 for example, the FSR 3.1 image often looks softer at the same slider settings.
Pixelation in Motion
In our previous reviews of FSR 2.2/3.0 implementations we've often seen major issues with pixelation in motion, shimmering, or flickering in vegetation—those artifacts were very common across multiple games. That's why we closely examined this during our FSR 3.1 testing, expecting major improvements. We can confirm that the amount of pixelation in motion is lowered with the FSR 3.1 update, even at low resolutions such as 1080p. However, the actual amount of visible pixelation in motion will vary depending on how much motion is happening on screen. For example: when simply running around at a standard speed, not sprinting, there's noticeably less pixelation compared to the FSR 2.2 image, but as soon as you start sprinting, FSR 3.1 is only slightly better. Essentially, this means that slower-paced sequences of the game will receive some improvement in image stability, while more action-packed and fast-paced sequences will look almost the same as before, which is a bit disappointing.
Particle Effect Stability
Particle effect rendering and their temporal stability is another important aspect of image quality. Unfortunately, there is only a very small difference between FSR 2.2 and FSR 3.1 when it comes to particle effect details, such as neon-like signs, smoke and fire effects, waterfalls, or sea waves. The FSR 3.1 image still produces noticeable degradation in those particle effects, resulting in a blurrier image in motion. These temporal instabilities are especially visible in waterfalls and sea waves, and when comparing FSR 3.1 to DLSS, or even XeSS 1.3 in its DP4a mode, the differences are still very significant in favor of both DLSS and XeSS 1.3. In our side-by-side comparison video we specifically contrast particle effect rendering between the competing upscalers, where issues with FSR 3.1 are visible even without zooming in on the image.
Frame Generation
On the other hand, the image quality of FSR 3.1 Frame Generation is excellent. In Horizon Forbidden West, when using DLSS as the base image for both Frame Generation solutions, we didn't see any major differences in image quality between AMD's and NVIDIA's Frame Generation solutions, which is a very good thing. The only exception is a slightly softer overall image in motion with FSR 3.1 Frame Generation, specifically at 1080p resolution.
Now all we need is for adoption rates in games to increase significantly, which should be much easier now. The FSR 3.1 update expands the supported hardware base for frame generation close to 100%, because FSR 3.1 will run on all cards from NVIDIA, AMD and Intel (except for some very old ones). For developers this should justify the resources spent on adding FSR Frame Generation support.
With the ability to enable FSR Frame Generation with any antialiasing or upscaling solution, including DLSS or DLAA, we can now properly test AMD's Frame Generation image quality without artifacts introduced by FSR upscaling, directly comparing it to NVIDIA's Frame Generation.
Performance
Regarding performance, the FSR 3.1 upscaling implementation does not offer further improvements in performance compared to FSR 2.2, essentially offering the same performance gains as DLSS (which is good). In Horizon Forbidden West, the upscaling performance increase is around 25% in "Quality" mode compared to native rendering at 4K resolution, and around 20% at 1440p and 1080p resolutions. XeSS 1.3 has slightly different performance gains due to alternative resolution scaling values in its presets. With DLSS Super Resolution or FSR 3.1 in "Quality" mode and Frame Generation enabled, you can expect almost doubled performance across all resolutions compared to native rendering.
