The two companies ruling the GPU markets don’t simply follow the purpose of delivering more powerful GPUs with each generation. There is a lot in terms of technology that goes into manufacturing them. One of these is upscaling. AMD and Nvidia have their own technologies for upscaling. With the launch of Nvidia’s RTX 4000 series introducing us to DLSS 3.0 and AMD recently giving us a peek at FSR 3.0, what better time to discuss these technologies than now? Today, we compare these technologies and clear the verdict of Nvidia DLSS vs AMD FSR.
To compare these upscaling methods, we need to start with the basics and work our way up.
What Is Upscaling?
Basically, everything your monitor displays is made up of pixels. So, pixels are the units or blocks of digital images. Upscaling, or upconversion, is the process of stretching these pixels to fit a higher native resolution. For example, the resolution 1920×1080 contains 2,073,600 pixels while 1280×720 contains 921,600 pixels.
Now It sounds simple, but the problem is simply stretching pixels will always lead to a reduction in image quality. So if you open a video of 720P resolution on your 1080P monitor, you’ll see a decrease in video quality. In other words, a lack of pixels for a particular resolution will always be trouble. So how do Nvidia and AMD counter this?
AMD’s upscaling technology is called FSR or FidelityFX SuperResolution, while Nvidia’s technology for upscaling is called DLSS or Deep Learning Super Sampling.
Nvidia’s DLSS 1.0 was launched with the first generation of RTX graphics cards in 2019. Since then, this technology has seen a lot of progress, and Nvidia are all set to release DLSS 3.0 in November 2022. As the name suggests, Nvidia uses AI and deep learning for this technology, and it is a form of supersampling.
Now, what is supersampling? Well, it is a very advanced anti-aliasing method used to remove jagged edges produced from non-parallel pixels). Supersampling does this by rendering the image at a higher resolution and then downscaling that to fit the lower resolution. SSAA is normally not preferred as a method of anti-aliasing because it causes a huge performance hit. But in DLSS, the function is performed offline by AI, greatly improving performance. DLSS offers four modes:
- DLSS Quality is the most conscious in terms of quality, producing a super-sampled resolution closest to the target resolution.
- DLSS Performance prioritizes performance over quality. This typically means a resolution you’ll get a resolution that is twice the number of pixels of your target resolution. (e.g 1080P to 2160P)
- DLSS Performance Ultra is geared completely towards performance. Here, DLSS may have to render frames that are too detailed compared to the target resolution, and quality will likely be affected.
- DLSS Balance provides a great balance between performance and quality. Recommended for most users.
Moreover, DLSS has seen three iterations so far, which are mentioned below.
DLSS 1.0 did not receive a lot of warm reception, likely because the concept was new at the time, and DLSS was the least efficient of the three iterations. DLSS 1.0 started by feeding lower-quality image sequences to a neural network that is trained to produce higher-quality results. Aspects from these high-quality frames are then filled into the low-quality frames.
This would give the look of high resolutions with minor differences, while your GPU is only utilized in rendering the original low resolution. The information is fed to your GPU through driver updates. Thus, games could essentially be run at higher resolutions providing the performance of lower resolutions.
DLSS 1.0 saw the neural network training being carried out separately for each game, which was a big limitation.
DLSS 2.0 brought major performance improvements, including improvements in the “temporal feedback” technique. Temporal feedback is the information from motion vectors that predicts the motion of the next frame from the previous one.
Furthermore, there are improvements in the AI network thanks to improved tensor core architecture, which speeds up the entire learning process by two times.
DLSS 2.0 is also much more generalized because it introduces a single network for all the games instead of game-specific networks. This makes it easy to integrate DLSS into more games faster.
DLSS 2.0 also introduced us to four modes, Quality, Balance, Performance, and Performance Ultra. These have been discussed above.
Nvidia announced DLSS 3.0 during the GTC conference in September 2022. This new iteration promises up to two times better performance than DLSS 2.0.
DLSS 3.0 makes use of Nvidia’s new Optical Flow Accelerator, which produces new frames rather than just pixels. The optical flow accelerator feeds the neural network with information about the latest frame and the frame before that. This information includes the behavior and motion of pixels and creates an “optical flow field“. This allows the neural network to produce new frames with incredibly accurate detail. (Read More)
FSR or FidelityFX Super Resolution is AMD’s upscaling method, which is also a form of super-sampling. The process here is software-based, unlike DLSS, which relies on tensor cores.
In this process, a low-resolution image is upscaled and undergoes edge reconstruction. This image is then sharpened to improve quality further. The level of quality depends on the mode of FSR you use. Here, the quality goes in descending order:
- Ultra Quality (1.3x scaling)
- Quality (1.5x scaling)
- Balanced (1.7x scaling)
- Performance (2.0x scaling)
Although FSR is considered a direct competitor to DLSS, AMD’s first iteration of FSR launched in June 2021, over two years after DLSS 1.0. In the latest news, AMD announced the launch of FSR 3.0.
FSR 2.0 sees an improvement in performance along with improved image quality. In FSR 2.0, AMD moved from spatial upscaling to temporal upscaling, which is the same method Nvidia has been using for DLSS. In temporal upscaling, data from the previous frames are used to produce a new frame of much higher quality. Motion vectors are also being used here, the same as DLSS, which accurately predicts motion for the next frame. The image is also further improved by using custom anti-aliasing.
AMD announced FSR 3.0 during the premiere of the RX 7000 series of graphics cards, with official support coming in 2023. In the third iteration of FSR, AMD is combining their super-resolution technology with a brand new technology called “fluid motion frame” technology. AMD has not mentioned any details about this technology but has promised up to a 2x improvement in FPS over FSR 2.0 in 4K gaming. (Read More)
Nvidia DLSS Vs AMD FSR
With all the details out of the way, let’s see how these two technologies differ in quality and performance. Note that we’re comparing DLSS 2.0 to FSR 2.0. Before we start, a quick overview of the difference between them:
|Nvidia DLSS||AMD FSR|
|Hardware-based super sampling method powered by Nvidia Tensor cores||Software-based upscaling technology. Not hardware dependent|
|Compatible only with RTX 2000 series and above||Compatible with a variety of AMD and Nvidia GPUs, as well as PS5 and XBOX X/S.|
|Not open-source.||Open-source, can be integrated easily into Unreal Engine 5 and Unity|
Now, unfortunately, the options are quite limited for us to compare DLSS and FSR for an apples-to-apples comparison, but Deathloop provides us some respite. Here, thanks to data provided by Hardware Unboxed, we can compare the quality and performance of the two technologies in different modes:
Tip: Click on the images if you wish to zoom them in and analyze them further. (then CTRL + Scroll Up to zoom in, CTRL + Scroll Down to zoom out)
Quality Comparison: 4K
Let’s compare the results of Nvidia DLSS vs AMD FSR in “Quality” mode, which provides the best quality for both technologies. Here, with the images zoomed out, there’s little to no quality differences to be noted. But with a 300% zoom, it can be noticed that FSR 2.0 takes a slight edge in terms of sharpness. Furthermore, a somewhat increased level of detail can be noticed on the rocky platform above.
Quality Comparison: 2K
Now, let’s move on to the quality comparison in 2K. We’re continuing with the “quality” mode settings on both FSR and DLSS:
No quality difference is to be seen. Let’s see a zoomed-in result for better details.
Again, there’s really not much of a difference to be noted. But with some careful analysis, it can be said that the sharpness of shadows is slightly better for DLSS, while texture detail is slightly better on FSR.
Quality Comparison: 1080P
In 1080P on Quality mode, there are a few things to be noted here. At first sight, you can see DLSS provides much better sharpness, especially where there are shadows. Color saturation is slightly toward FSR’s side, though.
If you look way into the background behind this building, you can see better texture quality in FSR, whereas the background is rather foggy and shows less detail in DLSS.
Finally, some flickering can be seen in the video on the red text in FSR’s side, but DLSS shows no flickering whatsoever.
To sum up, FSR 2.0 and DLSS 2.0 are somewhat even in the quality comparison. FSR shows slightly better results in 4K. In 2K, both the technologies show very similar results, DLSS shows slightly better shadow sharpness, and FSR shows slightly more detail in textures. In 1080P, there are a few variables. DLSS shows improved sharpness in the foreground, while FSR shows better texture quality in the far background.
The results cannot be generalized for all the upscaling tests, but it’s still safe to say that FSR produces slightly better texture detail, while DLSS shows slightly better sharpness.
It’s time to carry out the performance comparisons to see how the two technologies compare. We’ll be using two GPUs: Nvidia’s RTX 3060 Ti and AMD’s RX 6700 XT. The 3060 Ti will allow us to compare results across FSR and DLSS directly, while the 6700XT will be tested and compared separately.
4K Performance: 3060 Ti
Starting off with the 4K performance benchmark on the 3060 Ti, we see an average of 41 FPS and %1 low of 32 FPS on native 4K. DLSS 2.0, while bettering FSR on every mode in terms of average FPS, actually falls behind in terms of 99th percentile FPS. This means that the frame time is less for FSR compared to DLSS. So, despite the slightly lower average FPS, AMD is the winner here.
4K Performance: 6700 XT
The 6700XT being a weaker card falls behind in overall performance compared to the 3060 Ti. The good thing is that %1 lows haven’t seen a huge drop, and FSR still provides a huge boost in FPS over native 4K.
1440P Performance: 3060 Ti
In native 1440P, the 3060 Ti gives us 81 FPS on average, with a %1 low of 53 FPS. This average FPS is increased by up to 126 through FSR, and 127 FPS through DLSS. Overall, we see major boosts here for both DLSS and FSR, but DLSS takes a slight edge in both average FPS and %1 lows.
1440P Performance: 6700 XT
In 1440P, 6700XT also provides a huge boost. The 6700XT manages 69 FPS on average in native 1440P on the ultra preset. In FSR 2.0’s best case scenario of quality, the average FPS takes a leap to 99, while the performance mode increases it further by 20 FPS to 119 FPS.
1080P Performance: 3060 Ti
In 1080P gaming, both DLSS and FSR are somewhat limited in the sense that they can’t push FPS beyond a certain limit (~128 FPS). This is not a big problem, as we’re already seeing hitting 115 FPS on native 1080P, with FSR and DLSS providing another FPS boost with their upscaling capabilities.
1080P Performance: 6700 XT
1080P upscaling on the 6700 XT is an interesting case. Here, on the native 1080P resolution, we averaged 93 FPS. With FSR 2.0 on “quality” mode, the FPS jumps to 128, which was also the case with the 3060 Ti. But the FPS goes further up on balanced and performance modes, to 137 FPS and 143 FPS, respectively. So in 1080P upscaling, 6700XT gives a better boost in performance compared to 3060 Ti for FSR.
The performance results of Nvidia DLSS vs AMD FSR are also quite comparable. The only thing that can be said for sure is that both technologies provide huge boosts over native resolutions. The 6700 XT falls behind overall, but that’s because it’s a weaker card in general compared to the 3060 Ti.
The only major differentiating result is when you compare the 1080P performance of FSR on 6700XT against FSR on 3060 Ti. Here, AMD’s own card shows much better performance uplifts over the native resolution. This may hint that FSR may be suited slightly more toward AMD’s own hardware.
Compatibility And Availability
One important factor which really gives us a clear difference between Nvidia DLSS vs AMD FSR, is compatibility.
Nvidia DLSS is only compatible with their RTX GPUs. In fact, the upcoming DLSS 3.0 will only be available on their Ada RTX 4000 cards. AMD provides a broader horizon of compatibility, with FSR support available for cards as old as the RX 500 series, as well as Nvidia’s GTX 10-series and above. It’s also supported on the PS5 and XBOX X/S, both of which feature RDNA-2 GPUs.
The source code for AMD FSR is available on AMD’s website, which makes it much easier to integrate into games. DLSS integration, on the other hand, is slightly difficult, supposedly not possible without Nvidia’s cooperation.
In terms of availability, Nvidia’s claim is that DLSS is available for over 200 games (complete list), while AMD’s claim for FSR is 110+ games (complete list). So even though FSR is open-source, DLSS is available on a much wider variety of titles.
Final Thoughts: Which Technology Is Better?
Our comparison on Nvidia DLSS vs AMD FSR has only proved that this comparison shouldn’t be such a burning question among AMD and Nvidia fans. Both technologies will provide you massive performance uplifts while providing nearly the same or even better quality than native resolutions.
The only difference to note at present is that DLSS is available on a wider variety of titles, even though FSR is the more compatible technology and is open-source. FSR is supported on a larger variety of hardware, though, so it’s a mixed bag yet again.
If we compared AMD’s first iteration to Nvidia’s first iteration, the result would probably have favored Nvidia. But both Nvidia and AMD have improved their respective technologies by leaps and bounds since their initial release. Both have promised even further improvements in the forms of FSR 3.0 and DLSS 3.0. So, in a few months time, we might see the results favoring one team more than the other. But as things stand at present, both AMD and Nvidia are quite even in this competition.
Nvidia DLSS and AMD FSR are both really impressive technologies, showing little to no differences in performance or quality. So it’s hard to say that one technology is better than the other.
DLSS’s major problem that really sets it apart from FSR is that it’s dependent on Nvidia’s Tensor Cores, which are only found on their RTX GPUs. DLSS 3.0 will ONLY be available on the RTX 4000 cards.
Nvidia DLSS is available on over 200 titles, while AMD’s FSR technology can be found on over 110 titles.
Thank you! Please share your positive feedback. 🔋
How could we improve this post? Please Help us. 😔