Huge forward shift, FSR 4 calls, and sometimes even overcomes DLSS CNN Model

Together with the release of the Radeon RX 9070 and 9070 XT graphic cards, which we will devote to a separate article, AMD also brought a new version of FSR technology, in particular, in its fourth large version. And as soon as it turned out in various tests, improvements compared to FSR 3.1 are completely massive.

Comparison with FSR 3.1 brought a digital foundation where Alex Battalia checked Ratchet & Clank: Rift Apart and Horizon, prohibited by the West. For example, this indicates the quality of particles effects and, in particular, to the surface of the waterfall, where FSR 3.1 is completely lost from ghosts and artifacts, while FSR 4 is very stable.

The same applies to the transparent surface of a small lake, which with FSR 3.1 looks very blurry, while FSR 4 offers a clean and clearer surface. The fact that FSR 4 is more acute and can reconstruct more details can be seen well on the models of characters and their clothes.

Nevertheless, the greatest progress occurred in the overall stability of the image. Already in the static chamber, FSR 3.1 contains much more pseudonyms, but when moving, there is an even more noticeable difference. As soon as the scarlet begins to rush its spear on the horizon, the picture with FSR 3.1 is completely collapsed and full of pixelization. FSR 4? Very stable and without serious problems.

As for the comparison with DLSS, here he offered a very detailed video channel without spare ones, where Tim Shisser focused on resolving 1440p. While FSR 3.1 was far beyond the NVIDIA technology, with FSR 4 AMD it switches to a completely different level. Not only does he call Dlss CNN Model, but even overcome it in some respects. For example, FSR 4 can offer a less blurry and more clear image than DLSS with the CNN model. Often it is better in the reconstruction of particles or grass effects.

In the field of anti-alisters, this is a little weaker. FSR 4 can be a little better than DLSS CNN, but in a number of games it can be less stable. In addition, the technology has a relatively strange problem when the balance mode is more stable than the quality mode.

Of course, the comparison of FSR 4 and DLSS 4, IE DLSS, with the new transformer model, is somewhat different. In some relations FSR 4, such as transparency, hair or rain, can be relatively close to DLSS 4, but in other aspects the transformer model is a relatively clear winner. Despite this, this is a huge step forward for AMD.

As for performance, FSR 4 is slightly more demanding than FSR 3.1, and you will not get so much additional power. In principle, this is somewhere at the level of the DLSS transformer, mostly a little slower. DLSS CNN is much faster.

In conclusion, FSR 4 uses machine learning and due to the need to accelerate an 8-bit floating point, which is not available in the old series, and therefore FSR 4 works only on new RDNA 4 architectural cards, Radeon RX 9000.

At the same time, the number of games in which you can enable FSR 4 is quite limited. Although the technology can be used in all names offered by FSR 3.1, it does not work for all older FSR versions, and integration will be necessary directly from the developers.

Regular news review