During the Computex event in 2021, AMD finally announced what a lot of gamers have been waiting for. A response to Nvidia’s DLSS (Deep Learning Super Sampling) technology. This solution is called FSR or short for FidelityFX Super Resolution.
At the time of writing, FSR still isn’t officially released as it will on June 22nd. But, we already have enough information to understand its purpose and how it works.
Table of ContentsShow
What Is DLSS?
Before we can delve into AMD’s FidelityFX, we should first explain the original solution on the market, Nvidia’s DLSS.
DLSS has been in the making for a long time and it is a bit more complex than FSR since it relies on Tensor Cores found on RTX branded GPUs. Currently, just the 2000 and 3000 series.
With the power of the hardware and deep learning, DLSS does an upscale of lower-resolution images and outputs images of much higher resolutions. This results in significantly better performance while offering the same level of image quality or at least, minimally reduced quality.
At first, DLSS didn’t seem all that impressive because the first iterations of this technology proved to lose too much quality. The extra FPS were welcome, but users were left with blurry and smeared images as seen in Control and Battlefield V.
The changes that came with the release of DLSS 2.0 couldn’t be ignored. The performance uplift was the same, but in some situations, games even looked better than when played on native resolution.
These days, DLSS can be found in a lot of competitive and popular single games including Call of Duty Warzone and Black Ops Cold War, Cyberpunk 2077, Death Stranding, Watch Dogs: Legion, and several others.
Currently, the problem with DLSS is that needs to be included per game by developers. The implementation is still a slow process and that probably will not change any time soon since it is a solution that relies specifically on RTX hardware.
Fortunately, AMD’s FSR might change that since it is a software-based solution.
What Is FidelityFX Super Resolution?
Unlike DLSS, FSR does not rely on hardware which provides advantages and disadvantages to the table.
Even though FSR still has not been released and tested, but we can assume that the main problem with FSR is going to with image quality. Nvidia utilizes deep learning (through the Tensor Cores) to fill what the image is lacking since it is set at a lower resolution.
AMD’s method is a bit different that uses non-linear and linear upscaling. Taking the original (lower resolution) image, sending it through a network of those two types of upscaling, gathering the most important (highest quality) information of the image, and then combining it to create a pixel grid.
This grid of pixels is then further processed to deliver a higher-quality image.
This is a pretty simple explanation/understanding, but it is exactly that at its core.
AMD believes that this is a superior method to deep learning solutions, but since this is still a very young technology, we assume that it is not going to deliver a higher level of image quality. Of course, we have yet to see results, but considering Nvidia’s investment into DLSS and constant updates to their technology.
It’s going to take some time for AMD to reach the levels of DLSS 2.1.
However, it’s not all bad. Since this method does not rely on deep-learning and hardware such as Tensor Cores in RTX GPUs, FidelityFX Super Resolution can be used with a number of GPU series including RX 6000, RX 5000, RX 500, and even GTX 1000 series.
It’s entirely possible that these are only the officially tested GPUs. The technology might work on even older cards and newer RTX series. That is yet to be tested, but if it does work, it’s going to be a huge win for everyone, especially for AMD.
This is the kind of competition we like to see.
This also leaves a lot of room for improvement for FSR. Since it is completely software-based, future GPUs can easily be adjusted to work with this new technology with just a simple driver update.
Additionally, as FSR improves over the years, the in-game implementation might get much simpler for game developers. In other words, we’ll start seeing it in almost all games if it is so easy to implement.
Performance Gains From FSR
During the Computex event of 2021, we got a glimpse of what FidelityFX Super Resolution can do.
We know that these types of examples are always best-case scenarios picked by the company, but it’s still impressive from what we’ve seen.
Within the presentation, AMD explained that FSR offers 4 types of quality: Ultra, Quality, Balanced, and Performance.
In their showcased game, Godfall, we saw a 59% increase in FPS at 4K with an RX 6800XT while using the Ultra (best quality) setting. That is pretty impressive. Additionally, the Quality setting offered 2x the performance while Balanced and Performance options went well over 2x. In fact, FSR performance delivered 3 times the FPS.
You know there is a butheresomewhere.More FPS is always great, but not at a significant cost of image quality. Anything below the Ultra FSR setting started to show a huge loss in quality. Balanced and Performance looked significantly blurry and less detailed.
FSR’s four different quality modes
Although that is expected as DLSS is also blurry while using the Performance setting.
And sure, a 59% increase sounds great, but the question is: Will this same result be present in other games?
It is entirely possible that this specific game (Godfall) works great with FSR. The results might be entirely different on a GTX 1000 card, or resolution less than 4K. We have too little information to know any of this, so we’ll see what the final benchmarks will show us.
Make sure to come back to check out exactly what FSR will bring to the table on June 22nd.
