Can new technology withstand DLSS?
AMD has introduced the Radeon RX 6X50 XT series with the release FidelityFX Super Resolution 2.0 technology debut, AMD FSR 2.0. This technology has similar objective to Nvidia Deep Learning Super Sampling (Nvidia DLSS) and the latest Intel Xe Super Sampling (Intel XeSS): Use image magnification techniques to enable lower resolution video. Retains as much detail as possible and therefore provides more performance with low loss of visual quality.
More than evolution, FSR 2.0 is a completely different technology than FSR 1.0
AMD FidelityFX Super Sampling 2.0 is AMD’s own “built from scratch” technology, so unlike version 1.0, it almost deserves its name. Let’s find out from AMD’s new strategy, results in terms of performance and graphics quality, as well as how the dispute with Nvidia DLSS is. Do we no longer need tensor cores and therefore RTX to do good scaling?
Definitely AMD DLSS
The first big change from AMD FSR 2.0 is that it now very much resembles Nvidia’s approach to DLSS, or rather DLSS 2.0. AMD FidelityFX Super Sampling 1.0 did the same as all the listed technologies but with different hardware. He took a picture, magnified it, and used the FidelityFX Constrast Adaptive Sharpening (FidelityFX CAS) to preserve image detail and definition.
The new approach is closer to DLSS
AMD FSR 2.0 now uses more information with a temporary filter. In other words: when you decide what each pixel will look like, FSR 2.0 takes into account information from previous shots. It will also use other information from the game engine Motion vectorsOr motion vectors that inform the technology of where objects are and where they are going, which will once again help change the image size to make the right choice.
If you’ve experienced deja vu, make no mistake, as you probably remember our article on DLSS version 2.0 and how well it worked on super sampling technology on the GeForce RTX side.
FSR 2.0 captures motion vector information, object depth information and colors, mixes it with previous frame data, and decides how to fill in the missing spaces to increase image resolution.
The FSR approach is very similar to DLSS in its modes. Essentially, you reduce the actual rendering of the pattern and you use more machine learning to fill in the gaps in high performance modes. Below is the progress of these modes, remember that they indicate the final resolution, or output resolution, in this table:
– Continues after advertising –
Comparison in DLSS vs FSR 2.0 video
FSR 1.0 vs FSR 2.0 and vs DLSS graphics
It’s time to look at the evolution against FSR 1.0 and what the dispute is with DLSS in terms of graphics. We compared 4K resolutions as well as Full HD with entry and intermediate cards from both Nvidia and AMD. For the first test, let’s look at the performance on the 4K stage:
Below we will analyze the quality progression of FSR in its performance, in balanced and quality modes on this stage. To make the difference clearer, we use 5x zoom to capture details:
The result of the quality mode is excellent. It’s obvious how it manages to retain a lot of detail, and thanks to the sharpening effect we get the feeling that the resolution is higher than actually 1080p images. When switching to performance mode, the image starts to deteriorate, especially showing more artifacts on moving objects, which becomes more apparent if you watch the comparison video. Despite the loss of quality, FSR 2.0 gameplay is still quite effective in performance mode, especially when you are not a “screen-nosed” with a 5x zoom effect.
Now let’s compare again, but compare FSR 2.0 to 1.0 and also Nvidia DLSS:
– Continues after advertising –
With the introduction of three technologies 4K performance mode, i.e. all rendering in Full HD and upgrading to 4K, the evolution of FSR 2.0 is awesome. The image has much more definition, retains more detail, and has fewer artifacts on the edges of moving elements. DLSS still has an advantage over objects with more defined edges, fewer artifacts, but the difference is already negligible, especially without the scaling we used in the comparison above.
Let’s move on to another scene, now exploring the differences between these technologies in their quality modes, ranging from 2560×1440 to 3840×2106.
We have two enlarged segments in this scene:
In this part of the comparison, static images do not justify the elements that are more apparent in the video. Both native 4Ks along with TAA and FSR 1.0 have serious problems flashing at the edges of objects. FSR 2.0 and DLSS reduce or completely correct most of them, providing a much more complete and better image than native 4K rendering. DLSS has the advantage and handles some templates better, such as grids attached to the roof of a building, but in this comparison, FSR 2.0 better handles character movement in the background. Here I recommend you to watch this scene again in comparison to the video.
Now let’s see what FSR 2.0 technology looks like at lower resolutions. In this case, we go with the Radeon RX 6600, which handles the game in Ultra, RT On and Full HD resolutions.
Again flicker is an element that will only be visible in video comparison, but I have been emphasizing that it was obvious how much FSR 1.0 bothers, even in quality mode, this problem. The performance mode of FSR 2.0 also had this problem, in addition to losing the definition to the same degree as it does in FSR 1.0. In short: FSR 2.0 quality mode is ideal for 1080p. The graphic result is very good and although the performance gain is modest, it is more interesting than natively running and using TAA.
This turns out to be the inevitable effect of working at very low resolutions. Like DLSS, FSR 2.0 suffers when the final resolution is too low. In quality mode, it displays HD to Full HD delivery, while in performance mode we have a base resolution of 890×540, which is little information for increased reconstruction.
FSR 2.0 performance
We compared the performance of FSR 2.0 to version 1.0 technology as well as the DLSS embedded in Deathloop with version 2.3 of Nvidia technology.
Entrance and old signs
When comparing old cards, we see that the popular Radeon RX 580 and GTX 1060 6GB are taking great advantage in performance, especially Radeon. Switching to the latest cards like GTX 1650 GDDR6 and Radeon RX 6500 XT, we get two opposite results.
The GeForce GTX 1650 GDDR6 benefited greatly from FSR 2.0, with a 20% increase, while the Radeon RX 6500 XT virtually stopped with rather poor shooting speeds. In fact, this card performed all the tests poorly, which showed that the limitations of this model – which we have already faced in the past – may have had a negative impact on this test.
AMD FSR 1.0 still delivers greater performance, but as we show by comparison, it is not at the same level as the other two technologies.
Starting with Full HD cards like the RTX 3050 and Radeon RX 6600, we have good results with FSR 2.0 quality and interestingly, GeForces has taken full advantage of it. The RTX 3050 has increased its performance by almost 25%, by 15%, which is still good, but much less relevant. DLSS was slightly better than FSR 2.0, while FSR 1.0 and its easy implementation remain the best test results.
Finally, we looked at the 4K RTX 3060, a card that does not have the performance to handle a game with this resolution. With all the technology being given an honorable incentive, FSR 2.0 already brings in 30% gain and averages over 50 fps. This shows that only use FSR in Performance mode, which still has a good level of quality, or change the preset from Ultra to higher to make 4K gameplay viable on the RTX 3060. But the advantage of the tensor is even greater. Cores and enable DLSS, which will bring the game up to 60fps + in high quality.
AMD FSR 2.0 still has minor drawbacks compared to DLSS 2.3 But it will not even need to tie or beat the Nvidia resource. You just got close enough Demonstrate themselves as a viable alternative for Radeon owners. And he did.
AMD FSR 2.0 didn’t even need to beat DLSS, we just got close enough. And he got it
Version 2.0 of the FSR is comparable in quality to DLSS, and had it not been for our magnified reviews, it would have been difficult to tell the difference. When the gameplay is going on, it is unlikely that anyone will see the difference if you do not hold the pixels under a microscope and observe them. This is great news not only for AMD cardholders but also for GeForces owners without DLSS support.
But all this is not a breakthrough and AMD FSR 2.0 needs more developer intervention. Fortunately, DLSS 2.3 paved the way for this type of temporary filter with increased reconstruction, so it might be faster to get it than in the early days of DLSS. But this is still to be seen in practice.
DLSS cardholders should continue to use the tensor cores of their GeForce RTXs. DLSS is faster than the AMD FSR, delivers more frames, and has slightly better graphics output. Next, the AMD FSR is an excellent option and should be used in 4K, even in performance mode, while Full HD is ideal for use in quality mode. If this does not provide enough footage to play, the way is to go with FSR 1.0, which is even more interesting than not using the feature, in many cases it might be better to stay in native resolution in these cases.
Source: AMD Community, GPU Open