AMD FSR vs. Nvidia DLSS: Which Upscaler Reigns Supreme?

[Admin]

See how AMD's FSR and Nvidia's DLSS affect image quality and performance in four games that support both technologies.

AMD FSR vs. Nvidia DLSS: Which Upscaler Reigns Supreme? : Read more

 

[-Fran-]

This is not only the most important conclusion that needs to be drawn from this, but hopefully the only one: "Ultimately, games and game developers now have several choices available, and that's a good thing".

For all everyone can argue or fanboi or pit one tech against another, the truth of the matter is that if these technologies are pushed to compete with one another and they can be used widely (i.e: no vendor lock), then everyone wins; maybe even AMD, Intel and nVidia.

Personally, I still find them stupid as up-scaling means you just need a better card or lower details. But as with everything, this is just another voluntary toggle that's there for you to use if you find it works for you.

Regards.

 

[JTCPingas]

Tom's Hardware failed to mention that FSR is open source. That alone is a big win over DLSS.

 

[Alvar "Miles" Udell]

Something I can't wait to see is how Intel's XeSS compares with DLSS and FSR in terms of performance and quality, as well as game compatibility. Three different approaches mean the potential for a large performance delta between three cards of on paper equal power.

 

[hotaru251]

XeSS

wasnt that just a modified version of FSR?

 

[Blacksad999]

wasnt that just a modified version of FSR?

It's a similar upscaling method to DLSS, using AI to "intelligently" fill in the gaps, rather than the FSR method, which is Lanzcos with edge detection.

 

[Alvar "Miles" Udell]

wasnt that just a modified version of FSR?

Nope, and Intel said it will eventually be open sourced as well, but it will have two modes, one for cards with AI processors and one without. If it all works out it's possible AMD FSR will be quickly buried and forgotten, leaving AMD to be a follower yet again instead of a leader.

Here is Intel’s Spectacular Xe SuperSampling (XeSS) Demo In 4K Quality (wccftech.com)

XeSS Is Intel’s Answer to DLSS, and It Has One Big Advantage | Digital Trends

 

[JarredWaltonGPU]

Tom's Hardware failed to mention that FSR is open source. That alone is a big win over DLSS.

No, it's in there! (Or at least, it is now. ) Note that the linked DLSS and FSR articles definitely go into more detail about how each works, as the focus here was mostly on performance and image quality comparisons.

 

[Squidkid]

I don't know if my eyes are just still in good condition but fsr is sooo blurry it just looks like youve lowered the resolution and then used reshade.

It's not even a competition betweenthe two dlss is far superior with almost sometimes better than native image quality.

Fsr was just a gimmick to compete with Nvidia as amd knew they had lost with ray tracing it's like the no frills version.

I don't mind using dlss in games but I don't think I'd ever use and play a game with fsr just loses way to much quality would rather spend more money and buy a faster card.

 

[JarredWaltonGPU]

I don't know if my eyes are just still in good condition but fsr is sooo blurry it just looks like youve lowered the resolution and then used reshade.

It's not even a competition betweenthe two dlss is far superior with almost sometimes better than native image quality.

Fsr was just a gimmick to compete with Nvidia as amd knew they had lost with ray tracing it's like the no frills version.

I don't mind using dlss in games but I don't think I'd ever use and play a game with fsr just loses way to much quality would rather spend more money and buy a faster card.

DLSS definitely looks better in most of the games I've tried as well. The thing is, FSR works on pretty much everything. If you're on an old GTX 1060 and struggling to get above 60 fps, FSR can probably get you there. And for games where you only need an extra 5-10 fps, you can use the ultra quality mode and likely never really notice. But the problem is FSR (and DLSS) aren't generally used on older games, and only about a third maybe of new games. Which means you'll have to use lower resolutions and just do it the old fashioned way. I'm still hopeful FSR can improve in quality over time, or maybe XeSS will end up as a universal upscaling solution with quality matching DLSS while working on all modern GPUs. One can dream, right?

Also a fun aside: Serious image upscaling and enhancement that doesn't result in more noticeable image quality losses is certainly possible. The AMD and Nvidia images used in the article were upscaled using BigJPG.com, and the results are quite good. It also took about a minute to get the final result, but the source images were only 720p for AMD and 510p for Nvidia. And if you go look now, you'll see the denoising results of the upscaling, but it's still better (IMO) than the upscaling you get with Photoshop. I really hope the next iteration of Photoshop adds a high-end AI trained upscaling algorithm! Deep Image is the one to beat right now. Give me that level of quality at 60 fps and I'd be super impressed. 🙃

 

[Sleepy_Hollowed]

Like others have mentioned, the AMD version has an edge just because it's Open Source and not only will it get better over time, but it's free to use for developers, which will help many smaller budget games use it instead of DLSS unless there's no tools to implement it easily.

 

[Zarax]

To be fair, FSR and DLSS have different aims and different lengths of development time.
Personally I would love NVIDIA to commit more to their vectorized textures research (which would make them resolution free) and AMD to deepen FSR using a stronger integration with gaming engines (possibly it would make it slightly less flexible but higher quality), both solutions have their strengths and weaknesses.

 

[BeedooX]

I don't mind using dlss in games but I don't think I'd ever use and play a game with fsr just loses way to much quality would rather spend more money and buy a faster card.

I've tried FSR in a couple of games now, and I couldn't really tell the difference between it and native - other than getting a boost in frame-rates.

I'll be honest with you, when I see a side-by-side static image between DLSS and FSR, the DLSS image always makes me feel like I've just put drops in my eyes and I can no longer focus.

I'm not saying DLSS isn't impressive in it's ability to retain detail, but often it could subjectively use some sharpening.

 

[VforV]

Remember people, supposedly DLSS 3.0 is in the works (to be launched with Lovelace, not 100% confirmed yet) and also we knew since the launch of FSR 1.0 that AMD was already working on a better version, so FSR 2.0 is also coming.

If anyone is expecting AMD to just drop dead and not compete anymore (both hardware and software), they are very naive.

DLSS has better quality in still images (arguably and depending on the game and resolution), while FSR looks better in motion and it's also sharper, which I and many others prefer.

For those aspects alone there is no clear cut winner, but for the fact that FSR is open and works on any GPU, for that part alone FSR is the winner. And for the fact that nvidia made DLSS a closed RTX premium tax only system and mocked their own GTX costumers, I say **** nvidia.

XeSS being open (at some point) is also nice. I can't wait for intel to take on nvidia, also something I did not think I'll ever say, but I am rooting for intel in that particular fight.

 

[JarredWaltonGPU]

DLSS has better quality in still images (arguably and depending on the game and resolution), while FSR looks better in motion and it's also sharper, which I and many others prefer.

I disagree very much with the second part of your statement, because it absolutely depends on the game and settings available. Deathloop looks awful with FSR at any setting beyond ultra quality, and even ultra quality looks a bit soft. In motion, though, I'm hard pressed to see much of a difference between any of the modes on DLSS or FSR -- at least in fast motion and actually playing a game like Back 4 Blood. Maybe "in motion" in Myst, you might be able to determine which mode is running. Also, testing at 4K (which is what I used for this piece) is really a best-case scenario. DLSS and FSR at 1080p tend to have far more noticeable artifacting. There's basically a minimum amount of detail you need to get a good upscaling result, and when you start at less than 1080p, there's simply a lack of data.

Let me also note that CAS (FidelityFX Contrast Aware Sharpening) tends to look quite good to me. Unfortunately, for a lot of games with DLSS and/or FSR support, enabling DLSS/FSR disables CAS, which reduces the sharpness you'd normally get. In those games, you can often get a more pleasing overall look with temporal upscaling (like 90% resolution) and CAS than you'd get with FSR/DLSS and no CAS.

 

[escksu]

Remember people, supposedly DLSS 3.0 is in the works (to be launched with Lovelace, not 100% confirmed yet) and also we knew since the launch of FSR 1.0 that AMD was already working on a better version, so FSR 2.0 is also coming.

If anyone is expecting AMD to just drop dead and not compete anymore (both hardware and software), they are very naive.

DLSS has better quality in still images (arguably and depending on the game and resolution), while FSR looks better in motion and it's also sharper, which I and many others prefer.

For those aspects alone there is no clear cut winner, but for the fact that FSR is open and works on any GPU, for that part alone FSR is the winner. And for the fact that nvidia made DLSS a closed RTX premium tax only system and mocked their own GTX costumers, I say **** nvidia.

XeSS being open (at some point) is also nice. I can't wait for intel to take on nvidia, also something I did not think I'll ever say, but I am rooting for intel in that particular fight.

AMD can't patent FSR even if they want to. Lanzcos upscaling and edge sharpening isnt new technology thats created by AMD, it has been around for ages. Furthermore, the method of upscaling a still image (a frame created) isntt new technology as well. Your monitor/TV already has built-in upscaling technlogy so that images lower than the native resolution will be able to fill up the whole screen. Media player software does this as well.

 

[renz496]

Like others have mentioned, the AMD version has an edge just because it's Open Source and not only will it get better over time, but it's free to use for developers, which will help many smaller budget games use it instead of DLSS unless there's no tools to implement it easily.

DLSS get better due to it's ML nature. Not sure how FSR will get better unless AMD add ML features to it. Else AMD might have to come up with completely new tech to counter new iteration of DLSS. Remember originally AMD try to counter DLSS with RIS. then they have to make FSR to counter DLSS 2.0.

Older DLSS might be harder to implement but nvidia try to make that less a hassle when they already train DLSS with majority of game engine out there.

 

[renz496]

For those aspects alone there is no clear cut winner, but for the fact that FSR is open and works on any GPU, for that part alone FSR is the winner. And for the fact that nvidia made DLSS a closed RTX premium tax only system and mocked their own GTX costumers, I say **** nvidia.

To be fair DLSS originally was meant to augment ray tracing titles to counter the performance hit from enabling RT. It wasn't supposed to come to non RT title. In RT tittle those GTX already extremely slow now you want to burden the shaders more with AI processing? And whole focus is to take advantage of tensor core inside nvidia new architecture at the time. Dealing with AI already makes it complicated. Probably why nvidia did not want to advance the development of AI upscaling using gpu shader. Just look how RTX voice processing end up being on GTX1080 done by gamer nexus. Even on AMD side we saw they try to avoid using AI. they rather create FSR rather than augment their original RIS with AI to counter DLSS 2.

 

[VforV]

I disagree very much with the second part of your statement, because it absolutely depends on the game and settings available. Deathloop looks awful with FSR at any setting beyond ultra quality, and even ultra quality looks a bit soft. In motion, though, I'm hard pressed to see much of a difference between any of the modes on DLSS or FSR -- at least in fast motion and actually playing a game like Back 4 Blood. Maybe "in motion" in Myst, you might be able to determine which mode is running. Also, testing at 4K (which is what I used for this piece) is really a best-case scenario. DLSS and FSR at 1080p tend to have far more noticeable artifacting. There's basically a minimum amount of detail you need to get a good upscaling result, and when you start at less than 1080p, there's simply a lack of data.

Let me also note that CAS (FidelityFX Contrast Aware Sharpening) tends to look quite good to me. Unfortunately, for a lot of games with DLSS and/or FSR support, enabling DLSS/FSR disables CAS, which reduces the sharpness you'd normally get. In those games, you can often get a more pleasing overall look with temporal upscaling (like 90% resolution) and CAS than you'd get with FSR/DLSS and no CAS.

FidelityFX CAS is indeed disabled when FSR is used, but if I'm not mistaken you can still use the sharpen feature in Radeon Software, right? I'm not playing any FSR games now to test this.
And if not there is always Reshade sharpen, which I know I used with FSR and it does look even better.

To be fair DLSS originally was meant to augment ray tracing titles to counter the performance hit from enabling RT. It wasn't supposed to come to non RT title. In RT tittle those GTX already extremely slow now you want to burden the shaders more with AI processing? And whole focus is to take advantage of tensor core inside nvidia new architecture at the time. Dealing with AI already makes it complicated. Probably why nvidia did not want to advance the development of AI upscaling using gpu shader. Just look how RTX voice processing end up being on GTX1080 done by gamer nexus. Even on AMD side we saw they try to avoid using AI. they rather create FSR rather than augment their original RIS with AI to counter DLSS 2.

BINGO! They don't use AI so all GPUs can benefit. That's a plus for me, not a negative. And what you see as a plus for me is a negative.

Nvidia does not care about anything else other than the most obscene amounts of money they can make, so they use AI and "force" it's custumer base to upgrade to RTX, while AMD gave FSR to everyone for free. Well this former nvidia customer (GTX 1080 and 1060) says: I'm not your object of mockery nvidia!
I rather pay intel for their GPUs than nvidia again.

I'm done with them, don't care about their tech or their X% extra performance in some games, because in others they lose vs AMD. It's not like they win at everything and every game and every scenario by +50% difference. I care more about my integrity and principles.

 

[renz496]

BINGO! They don't use AI so all GPUs can benefit. That's a plus for me, not a negative. And what you see as a plus for me is a negative.

I did not see AMD not using AI as a negative. But rather what AMD try to do with FSR has been done by others in the past. They just did not really put a fancy names on it. Epic games for example decided not to support FSR Aon UE5 because they think their TAAU will be much better. An improvement from their existing TAA on UE4. But they still allow DLSS to be added to their engine because DLSS try to do something different than what they try to do with TAAU.

But going forward having some sort of tensor cores inside the gpu might end up being the standard. Nvidia already do it. Intel also doing it. All they need to do is to convince MS to make it into a direct x standard that everyone can use.

Right now we have three different tech competing. DLSS, XeSS and FSR. even if they some of them were open source as long as the tech are associated to certain company adoption will be mixed in the bag. Open source does not mean the tech will be accepted by everyone. Just look at FC6. Ubisoft specifically mention that console version of the game will not going to use FSR because implementing FSR is part of their partnership with AMD on PC. It is open source and easy to implement. What kind of excuse Ubi have for not implement them on console? And i remember what happen before with Battlemage TressFX implementation. There were problems on nvidia card. Rather than fixing the problem they simply disable TressFX entirely for nvidia cards. When asked about it they said tressfx implementation are done because of their marketing partnership with AMD. but in reality they probably don't really care if the tech will work well for everyone. For some publisher they probably not going to use the tech as long as it is being associated with IHV A, B or C unless there are somekind of sponsorship even if they were easy to do. Because doing it freely will give those IHV free marketing.

 

[VforV]

I did not see AMD not using AI as a negative. But rather what AMD try to do with FSR has been done by others in the past. They just did not really put a fancy names on it. Epic games for example decided not to support FSR Aon UE5 because they think their TAAU will be much better. An improvement from their existing TAA on UE4. But they still allow DLSS to be added to their engine because DLSS try to do something different than what they try to do with TAAU.

But going forward having some sort of tensor cores inside the gpu might end up being the standard. Nvidia already do it. Intel also doing it. All they need to do is to convince MS to make it into a direct x standard that everyone can use.

Right now we have three different tech competing. DLSS, XeSS and FSR. even if they some of them were open source as long as the tech are associated to certain company adoption will be mixed in the bag. Open source does not mean the tech will be accepted by everyone. Just look at FC6. Ubisoft specifically mention that console version of the game will not going to use FSR because implementing FSR is part of their partnership with AMD on PC. It is open source and easy to implement. What kind of excuse Ubi have for not implement them on console? And i remember what happen before with Battlemage TressFX implementation. There were problems on nvidia card. Rather than fixing the problem they simply disable TressFX entirely for nvidia cards. When asked about it they said tressfx implementation are done because of their marketing partnership with AMD. but in reality they probably don't really care if the tech will work well for everyone. For some publisher they probably not going to use the tech as long as it is being associated with IHV A, B or C unless there are somekind of sponsorship even if they were easy to do. Because doing it freely will give those IHV free marketing.

I agree on one thing, in 3-5 years time these issues we have now about DLSS not supporting older gens or other hardware and FSR not using hardware acceleration - all these points will probably be moot at that time (no one will care about GTX anymore then).

I think we will get at a very uniform level of tech quality, performance and usefulness - in a way that either all 3 (DLSS, FSR, XeSS) support everything and are very close to one another in all aspects, or 1 of them will be the absolute king (which I highly doubt it will happen).

Either way this weird/mixed situation is only temporary, but for now I support FSR or even XeSS, and I really don't care about DLSS at all.

P.S. Also I do agree that some game engines (just a few) do have a better version of upscaling tech than FSR 1.0, like UE5 and Insomniac's engine, for example. I expect FSR 2.0 to fix a lot if issues that the 1.0 version has now, though.

 

[LokkenJP]

Something I still fail to understand is why people keep trying to compare DLSS vs FSR.

It's like comparing a F1 car with a street legal Volkswagen.

Of course DLSS is better. It SHOULD be better, any time and under any circumstance. It relies on somewhat expensive hardware (RTX only with Tensor Cores), and on heavily trained neural networks (a work that it's also needed, but hopefully it's done, at least for now, by NVidia themselves).

FSR, on the other hand, is an universal technology, open source and one that doesn't need any special hardware to run. Of course it's less advanced by it's own nature, and it SHOULD be slightly worse on any likely scenario.

Is like asking if a F1 is a better car than a Volkswagen Arteon. A F1 is faster, yes, and much more maneuverable. That doesn't mean that everyone should rush to their car dealers to get a F1 single seater.

The question is not if DLSS is better or not than FSR. It is better, by it's own design.

The real question is if FSR is good enough to be used on all those hardware that is not able to run DLSS (AMD, Intel, and older nVidia own GPUs), and if it's relative quality and universal availability will make up in the long run for the superior, yet more expensive DLSS alternative.
This last question, only time will tell.

Maybe the two can coexist; but I've always preferred open standards than propiertary solutions, and AMD solution wins here hands down. FSR needs to prove that it is good enough for it to become dominant, though. Even if everyone and their mother should be aware that, on any case, it won't be able to match DLSS quality most of the time (or any time for that matter).

For now, the results are promising though. FSR is, right now on it's first iteration, quite better than DLSS was on 1.0, even being designed under a "worse" technology base. It works quite nice on most if not all recent non-RTX GPUs out there. And it should improve even more with time...

 

[husker]

"That doesn't mean that everyone should rush to their car dealers to get a F1 single seater. "
-LokkenJP

Now you tell me!
Seriously though, great post & I agree 100%

 

[Pytheus]

I would expect DLSS to die in the future. Anyone who has been around long enough knows it's not the better technology that wins, it's the one that reaches the most customers. An open source product that can run on anything is more desirable that a proprietary feature on a limited number of GPUs.

 

[renz496]

I would expect DLSS to die in the future. Anyone who has been around long enough knows it's not the better technology that wins, it's the one that reaches the most customers. An open source product that can run on anything is more desirable that a proprietary feature on a limited number of GPUs.

Except nvidia is very serious when it comes to pushing their tech. That's why with DLSS nvidia do it a bit different. They try to make it as easy as possible for game developer to implement them and they work with as many game engine developer to make DLSS being part of the game engine from the get go. For game developer that use those engine all they need to do is enable the plugin and it will work automatically for them. The heavy lifting on DLSS were all done by nvidia.