FidelityFX Super Resolution is AMD's answer to Nvidia's DLSS, set to debut this year

[nanoguy]

Big quote: AMD has confirmed that FSR "FidelityFX Super Resolution" is coming this year to rival Nvidia's DLSS as a more open solution. According to AMD's VP of Graphics, FSR is "one of the biggest software initiatives we have internally, because we know how important it is that if you want to turn on ray tracing, and you don't want to (...) your GPU get hit so hard."

Nvidia's DLSS (Deep Learning Super Sampling) has earned appreciation from gamers, and for good reason -- the second version of the technology produces some stunning results in terms of both image quality and incremental performance, especially in titles that have support for ray tracing. The DLSS ecosystem is growing with every passing month, which makes Nvidia's RTX graphics cards a compelling choice.

AMD has yet to bring an equivalent technology to Radeon GPUs, and while many have resorted to using a lower than native resolution in combination with Radeon Sharpening or FidelityFX CAS to achieve somewhat similar results, that's not an ideal nor equivalent solution. On the upside, those tweaks do not require involvement from game developers to add specific support for the super sampling technique and also works at all resolutions, unlike DLSS.

AMD has been working on FidelityFX Super Resolution (FSR), a technology that would work across several AMD Radeon GPUs including those found in the Xbox Series X/S and the PlayStation 5. Understandably, it's trying to get it right on both PC and consoles, so it wasn't ready in time for the launch of the Radeon RX 6000 series graphics cards, nor the new generation of gaming consoles.

During an interview with PC World, AMD's VP of the Graphics Business Unit, Scott Herkelman said that work on FidelityFX Super Resolution is "progressing very well internally." He also said that evaluating the quality of the upscaling algorithm will take a bit more time, but at the same time the company has made it a priority to debut FSR this year.

AMD typically puts some emphasis on making its technologies open and cross-platform, and FSR appears to be no exception. At the same time, it'd be wise for the company to learn from Nvidia's mistakes with DLSS, which wasn't successful until after a year after its debut and after a major overhaul to its approach.

Herkelman explained that AMD's FSR won't necessarily be based on machine learning, which should offer a glimmer of hope to a lot of gamers stuck with older Radeon cards. At least in theory, AMD is aiming to launch FSR for all compatible hardware in one go, so the new upscaling tech could very well be based on Microsoft's DirectML API. As exciting as all of this sounds, we'll have to wait and see how AMD's approach works in practice.

Permalink to story.

https://www.techspot.com/news/88974-fidelityfx-super-resolution-amd-answer-nvidia-dlss-set.html

 

[hahahanoobs]

"AMD typically puts some emphasis on making its technologies open and cross-platform"

When you're primarily a component company that makes sense, but it isn't always a good thing.

 

[Dimitriid]

From the tone and comments it feels like very late 2021. Which might not be *terrible* since well, most folk probably will wait out almost if not all of 2021 waiting for GPU prices to come down anyway, but at this point this feels more like "Don't forget about us!" since DLSS 2.0 is making news but they've got nothing solid to show, they should be on the "Here's a quick demo and the first few games to support the feature coming later this year" phase not the "Yeah we're committing to releasing I guess *something* this year" which sounds more like maybe 2 or 3 games, not even popular ones.

 

[yRaz]

they should be on the "Here's a quick demo and the first few games to support the feature coming later this year" phase not the "Yeah we're committing to releasing I guess *something* this year" which sounds more like maybe 2 or 3 games, not even popular ones.

On the upside, those tweaks do not require involvement from game developers to add specific support for the super sampling technique and also works at all resolutions, unlike DLSS.

read the article

 

[VariableSpike]

Really good to know this is coming, especially as DLSS has been a large part of the draw towards Nvidia cards - as long as it works as well as DLSS, then it should be great, especially as it doesn't require the game devs to enable it and make it work

 

[punctualturn]

Really good to know this is coming, especially as DLSS has been a large part of the draw towards Nvidia cards - as long as it works as well as DLSS, then it should be great, especially as it doesn't require the game devs to enable it and make it work

The sad part is, as good as dlss is, there are still few games that support it and thats where marketing comes in and makes everyone believe that every single game that ever existed already supports it.

Which means, as of right now, you are looking at future games hence making this not that important.

What’s important right now is drivers stability, cards availability and hopefully, a return to more sane msrp prices, which thanks to nvidia, the market was raped by them when they launched the rtx20 series and knew that amd wouldn’t have anything for a whole year.

 

[Watzupken]

From the tone and comments it feels like very late 2021. Which might not be *terrible* since well, most folk probably will wait out almost if not all of 2021 waiting for GPU prices to come down anyway, but at this point this feels more like "Don't forget about us!" since DLSS 2.0 is making news but they've got nothing solid to show, they should be on the "Here's a quick demo and the first few games to support the feature coming later this year" phase not the "Yeah we're committing to releasing I guess *something* this year" which sounds more like maybe 2 or 3 games, not even popular ones.

I don't know. If it is end of the year, it is terrible to me. This was first announced with the release of the Navi 21 lineup last year. By the end of 2021, that will make it a year or more after the release of RX 6000 series which I feel will be very late in the product cycle before we see RDNA3 in 2022. While open and easy adoption is critical to its success, I feel so is time to market. AMD should not be taking their sweet time to deliver a critical piece of software enhancement, especially when running RT on their hardware takes a bigger performance hit than competition.

 

[Watzupken]

Really good to know this is coming, especially as DLSS has been a large part of the draw towards Nvidia cards - as long as it works as well as DLSS, then it should be great, especially as it doesn't require the game devs to enable it and make it work

I don't think any upscaling can be done magically without any dev work. In the case of DLSS, the adoption is easier now, but it took a long time to get there considering Nvidia introduced DLSS 2.0 quite some time back. And it is available as a plugin for a specific version of Ureal Engine 4 as well if I am not mistaken. So if there is a significant change in the codes with a new version, I wonder how much effort is required to fix it.

 

[Jon Tseng]

read the article

yRaz were you implying by your comment that FidelityFX Super Resolution doesn't require dev involvement and thus will be used for more than 2-3 games?

If so I do not think this is the case. Your second quote refers to existing Radeon Sharpening and FidelityFX CAS technologies not requiring specific game dev support, not to the upcoming Super Resolution tech.

 

[Kosmoz]

I don't know. If it is end of the year, it is terrible to me. This was first announced with the release of the Navi 21 lineup last year. By the end of 2021, that will make it a year or more after the release of RX 6000 series which I feel will be very late in the product cycle before we see RDNA3 in 2022. While open and easy adoption is critical to its success, I feel so is time to market. AMD should not be taking their sweet time to deliver a critical piece of software enhancement, especially when running RT on their hardware takes a bigger performance hit than competition.

Look at it this way, sure we would all like to have this NOW, but nvidia launched RTX and DLSS 1.0 and ware almost non existent for about 2 years... think about that. Would you like to be lied by AMD, like nvidia did and launch the tech tomorrow and have no games to work with it? Because that's what nvidia did...

So AMD basically even though it's late to the party, they at least don't want to launch with a lie, that its working when its not. Also making this open and across all platforms and games is something very very impressive and I really hope they can deliver... it would be a much better feat than nvidia DLSS. Imagine the work for this compared to a few DLSS games.

And the funny part is that nvidia is the bigger one, they would have more resources to make an universal solution, but NO we know they like their closed boxes just so they can milk the cows dry...

I would not be surprised if after AMD launches FSR (that's the official abbreviation now, from AMD), it will force nvidia to make their DLSS more open and broad, like AMD forced nvidia change with Freesync vs Gsync.

 

[Jpe1701]

If they make this work with the 5700xt and it works as well as dlss then I won't even need to upgrade for a while.

 

[Puiu]

If they make this work with the 5700xt and it works as well as dlss then I won't even need to upgrade for a while.

it will most likely be enabled on only rdna2 at the beginning.

 

[Fulljack]

Can't wait until someone made it working on Nvidia GPU. would be great for anyone that are still using older graphic cards like Pascal or Maxwell.

 

[Puiu]

Let's reserve our praises/criticism until we actually see it implemented in a few games.

 

[Dimitriid]

Look at it this way, sure we would all like to have this NOW, but nvidia launched RTX and DLSS 1.0 and ware almost non existent for about 2 years... think about that. Would you like to be lied by AMD, like nvidia did and launch the tech tomorrow and have no games to work with it? Because that's what nvidia did...

So AMD basically even though it's late to the party, they at least don't want to launch with a lie, that its working when its not. Also making this open and across all platforms and games is something very very impressive and I really hope they can deliver... it would be a much better feat than nvidia DLSS. Imagine the work for this compared to a few DLSS games.

And the funny part is that nvidia is the bigger one, they would have more resources to make an universal solution, but NO we know they like their closed boxes just so they can milk the cows dry...

I would not be surprised if after AMD launches FSR (that's the official abbreviation now, from AMD), it will force nvidia to make their DLSS more open and broad, like AMD forced nvidia change with Freesync vs Gsync.

I'm going to do a sort of baseless speculation here and assume Nvidia took this long because this is really difficult to get right.

Thinking about the timeline, we know that the feature Nvidia really wanted to push wasn't DLSS 1.0 but Ray tracing. Something that they had been working on for a long while but kept running into the same issue: GPUs were not powerful enough. So developing DLSS was more like a happy byproduct of trying really hard to make Raytracing somewhat plausible. And combined with the relative stagnant of resolution on the PC (Due to favoring higher refresh rates instead of higher resolutions) means that it could sort of work.

Similarly, we've got DLSS 2.0 because Nvidia really wanted to get their second go at Ray tracing right, and even with the Ampere performance boost it still wasn't enough but DLSS 2.0 now being far more viable means that yes actually, upscaling a lower resolution image to save a ton of performance lets you finally enable ray tracing without as big a trade off now. It's still noticeable, but looks far better.

So in my estimation is since AMD just hasn't been developing Ray Tracing and Super Resolution nearly as long I think at best we'll get DLSS 1.0 kind of quality or worst, likely worst since I think Nvidia just has more R&D resources than AMD to dedicate to just getting this one thing right vs AMD that's busy well, beating intel successfully which is wonderful but something's gotta give and we all feel that something will eventually be Radeon so we realistically shouldn't expect great improvements on software side and DLSS 2.0 levels of quality probably *would* require lots of R&D dedicated to it.

Last reason is that AMD might not want something that's just way better than DLSS 2.0 and works with as many or more games right off the bat, quite simply because that will eat the high end market cards quite a bit: There's a reason why Nvidia never did a 2050 and while I still sort of doubt they will do a 3050 this time around and probably will just name it 1750 and remove DLSS 2.0 and Ray tracing capabilities: can't have the 150 dollar cards being able to look almost as good as the 3070 by just enabling DLSS 2.0

 

[Scrye74]

I have started my own graphic card company. My current offering is better than anything even conceived of by Intel, Nvidia or AMD. It's faster, uses less power and MSRP cost is a fraction of the others. It will even teach your kids to read, bring unicorns into existence and solve world hunger.

You can't buy one right now though.

 

[Xex360]

They need to make it game agnostic, else it will end like DLSS which is basically non existent unless you play those 2 games that have it.
Instead of wasting silicon on useless technologies like these, just use the space to put more cores and increase performance of the GPU which can be used by all games new and old.

 

[punctualturn]

They need to make it game agnostic, else it will end like DLSS which is basically non existent unless you play those 2 games that have it.
Instead of wasting silicon on useless technologies like these, just use the space to put more cores and increase performance of the GPU which can be used by all games new and old.

They already stated that it will be available to everyone that is interested.

This is why I like AMD, for whatever reasons, they always do this and its a valid effort, unlike the d!cks at nvidia, everything is propetary and a lock-in trap for the customers.

 

[Kosmoz]

I'm going to do a sort of baseless speculation here and assume Nvidia took this long because this is really difficult to get right.

Thinking about the timeline, we know that the feature Nvidia really wanted to push wasn't DLSS 1.0 but Ray tracing. Something that they had been working on for a long while but kept running into the same issue: GPUs were not powerful enough. So developing DLSS was more like a happy byproduct of trying really hard to make Raytracing somewhat plausible. And combined with the relative stagnant of resolution on the PC (Due to favoring higher refresh rates instead of higher resolutions) means that it could sort of work.

Similarly, we've got DLSS 2.0 because Nvidia really wanted to get their second go at Ray tracing right, and even with the Ampere performance boost it still wasn't enough but DLSS 2.0 now being far more viable means that yes actually, upscaling a lower resolution image to save a ton of performance lets you finally enable ray tracing without as big a trade off now. It's still noticeable, but looks far better.

So in my estimation is since AMD just hasn't been developing Ray Tracing and Super Resolution nearly as long I think at best we'll get DLSS 1.0 kind of quality or worst, likely worst since I think Nvidia just has more R&D resources than AMD to dedicate to just getting this one thing right vs AMD that's busy well, beating intel successfully which is wonderful but something's gotta give and we all feel that something will eventually be Radeon so we realistically shouldn't expect great improvements on software side and DLSS 2.0 levels of quality probably *would* require lots of R&D dedicated to it.

Last reason is that AMD might not want something that's just way better than DLSS 2.0 and works with as many or more games right off the bat, quite simply because that will eat the high end market cards quite a bit: There's a reason why Nvidia never did a 2050 and while I still sort of doubt they will do a 3050 this time around and probably will just name it 1750 and remove DLSS 2.0 and Ray tracing capabilities: can't have the 150 dollar cards being able to look almost as good as the 3070 by just enabling DLSS 2.0

I mostly agree, except I'm more optimistic and based on the few leaks and little official info we have about FSR, AMD aims to have performance and quality closer to DLSS 2.0 than to 1.0. So I'd say we can at least expect DLSS 1.5 performance and quality, while also being open and widespread.

Also because of that it takes AMD longer too, just like it did with nvidia, only open and widespread on all devices is a much bigger undertaking, even harder to accomplish than what nvidia did.

They already stated that it will be available to everyone that is interested.

This is why I like AMD, for whatever reasons, they always do this and its a valid effort, unlike the d!cks at nvidia, everything is propetary and a lock-in trap for the customers.

Exactly why, I can't be too harsh with AMD when they make mistakes. No company is a saint, but between nvidia (or intel) and AMD I knwo the lesser evil is AMD. Every once in a while they do something open/free, that benefits everyone. So they have that +1 from me, until they stop doing that, if that ever happens.

 

[Roboyt0]

read the article

Pffft. Only plebs actually read the article in its entirety before solidifying an opinion and then taking the effort to make a comment. /s

 

[Roboyt0]

I don't know. If it is end of the year, it is terrible to me. This was first announced with the release of the Navi 21 lineup last year. By the end of 2021, that will make it a year or more after the release of RX 6000 series which I feel will be very late in the product cycle before we see RDNA3 in 2022. While open and easy adoption is critical to its success, I feel so is time to market. AMD should not be taking their sweet time to deliver a critical piece of software enhancement, especially when running RT on their hardware takes a bigger performance hit than competition.

Until it doesn't work properly for any (amount) of reason(s); even if it was only under rare circumstances. Then you have people getting on their soap boxes and shouting from proverbial mountain tops about how AMD/Radeon released something else that is 'broken' and is 'just like their drivers'.

They're working on implementing this for PC, Xbox and Playstation as well as having some sort of functionality for a MUCH larger variation of their products than Nvidia.

Everything is easy when you're sitting on the sidelines spectating.

 

[Dimitriid]

I mostly agree, except I'm more optimistic and based on the few leaks and little official info we have about FSR, AMD aims to have performance and quality closer to DLSS 2.0 than to 1.0. So I'd say we can at least expect DLSS 1.5 performance and quality, while also being open and widespread.

Also because of that it takes AMD longer too, just like it did with nvidia, only open and widespread on all devices is a much bigger undertaking, even harder to accomplish than what nvidia did.


Exactly why, I can't be too harsh with AMD when they make mistakes. No company is a saint, but between nvidia (or intel) and AMD I knwo the lesser evil is AMD. Every once in a while they do something open/free, that benefits everyone. So they have that +1 from me, until they stop doing that, if that ever happens.

This is a reasonable expectation AMD could get it right faster than Nvidia did, however there's one key factor here to also consider: Nvidia can and probably will just pay a bunch of publishers to get far more games supported on their proprietary implementation.

So is not that I have no confidence in AMD, I am just saying that AMD really needs to go above and beyond and greatly surpass Nvidia in order for their tech to be attractive enough for publishers to just turn down Nvidia bribes.

 

[punctualturn]

This is a reasonable expectation AMD could get it right faster than Nvidia did, however there's one key factor here to also consider: Nvidia can and probably will just pay a bunch of publishers to get far more games supported on their proprietary implementation.

So is not that I have no confidence in AMD, I am just saying that AMD really needs to go above and beyond and greatly surpass Nvidia in order for their tech to be attractive enough for publishers to just turn down Nvidia bribes.

Well, we can help, by ignoring all of those dirty nvidia shenanigans that you described.

 

[Shadowboxer]

This feels too late to me, they haven’t got any demos, no promised list of games, firm dates or anything, I’m guessing they don’t have much to show to press yet. But creating the technology is not even half the battle. Developers need to adopt it, studios need to release games for it. It’s taken Nvidia with their enormous market share and wealth years to get to the point we are at now.

Hardly anyone is buying AMD cards these days, so why would developers spend the time and money implementing this new super resolution solution? It’s going to be a long uphill battle and I have a feeling DLSS isn’t going to stop at 2.0.

Oh and of course it’s open source. AMD want to create this software then let everyone else work on it because they don’t have the expertise to own a project like this. They can’t even provide basic customer support for its CPUs, instructing users to go to Reddit! Believe me, no executive likes to spend large amounts of R&D to create a technology and float it out there for free.

Of course at this point it’s all just really obvious that AMD doesn’t give a dam about Radeon. They use the brand to win lucrative contracts with Sony and MS. They don’t give a dam about the tiny percentage of their customers buying their dedicated GPUs.

 

[TimmyP]

Every single post in this article is from an AMD apologist, who STILL doesn't understand this technology.

Image reconstruction requires much more than up-sampling, and its not gonna come through drivers, only decoding\accelerating hardware. You fools fault NVIDIA for "releasing it too soon" and now they are on top, why? Because DLSS 1.0 did what it needed to do: provide Nvidia hard data with regards to what parts of NON SPECIFIC GAME DATA need to be decoded. That was the whole game; YOU CANT DO THIS MASS TESTING INTERNALLY.

Now, DLSS is automated without any human training required AND its accelerated by tensor cores (vs 1 which was NOT). It literally adds more image data in a same resolution scenario; essentially a 1080p screen looks better than it could without DLSS at that same resolution, while providing a better than TAA supersample, FOR FREE because the image data its working with is already supersampled to 16k-32k (I think up to 32k?).

THEN, you have to high refresh monitors which are the true benefactors to this tech. Everyone talks about 4k this and that, yet DLSS is the difference between 120hz and 165hz. Which is absolutely enormous, especially with VRR\2.1.

You people sound like FOSS-ONLY proponents; this tech MUST be outsourced for it to be widespread and "good"... HA. Did you ever consider that Nvidia has had a clear vision of what they are doing with this tech? Because clearly they have.

Sorry you chose AMD this time around (if you could) but without a REAL answer to machine learning image reconstruction, you made the wrong one.