Report: AMD's competitor to Nvidia DLSS, FidelityFX Super Resolution, to launch in Spring

[Random_Person1234]

Summary

AMD's response to DLSS, FidelityFX Super Resolution, will launch in Spring according to a report. This may come during a major Adrenaline driver release around March. The source for this report says that Infinity Cache would be beneficial for FidelityFX Super Resolution.

Quotes

Quote

AMD's much-awaited response to NVIDIA DLSS which will be known as FidelityFX Super Resolution is expected to arrive this Spring, reports Prohardvare. The solution will make use of AI-assisted ML (Machine Learning) to help boost frame rates in games running on the latest RDNA 2 GPU powered Radeon RX 6000 series graphics cards.

Quote

The source reports that Infinity Cache would also be quite beneficial for FidelityFX Super Resolution while the technology can be enabled from within the game where it's supported, just like DLSS. On the other hand, the new AMD Radeon Boost feature will help solve some issues related to dynamic scaling where the feature would scale the games to a lower resolution to boost performance even when there's not a lot of action involved. So an update would help solve some of the issues on that end too. Overall, it looks like we are due for a major Radeon Software release soon so stay tuned for more info.

My Thoughts

I'm interested to see how good this competes with DLSS. I'm also wondering whether this will significantly improve AMD's ray-tracing performance compared to competing cards from Nvidia.

Sources

https://wccftech.com/amd-fidelityfx-super-resolution-launching-in-spring-to-tackle-nvidia-dlss/

[SkilledRebuilds]

Spoiler

POTM with or without Context.?

Yay,.. because I don't see the 6800 series as worth it without it.

[Middcore]

14 minutes ago, SkilledRebuilds said:

  Reveal hidden contents

POTM with or without Context.?

Yay,.. because I don't see the 6800 series as worth it without it.

 

In many ways I see DLSS as the more impressive and interesting feature than ray tracing.

 

FidelityFX SR almost certainly will not be as good as DLSS at launch. DLSS itself was pretty shite in its first implementations back in 2019.

 

However, unless it starts to become commonplace for games to actually support these features, they both fall squarely into the category of "perks". Wikipedia lists a grand total of 43 games that support DLSS, over two years on... but that list includes 10 titles that claim they'll support DLSS at some future time but have a date for actually implementing it of "TBA", and several games which haven't been released yet. Some are also games I have literally never heard of. 

 

I'd like to think that FidelityFX SR will see more widespread support since historically AMD has been more into open standards whereas Nvidia tried to sell based on exclusive features, but AMD themselves seem to be moving more in the latter direction lately. 

[Middcore]

9 minutes ago, Middcore said:

However, unless it starts to become commonplace for games to actually support these features, they both fall squarely into the category of "perks". Wikipedia lists a grand total of 43 games that support DLSS, over two years on... but that list includes 10 titles that claim they'll support DLSS at some future time but have a date for actually implementing it of "TBA", and several games which haven't been released yet. Some are also games I have literally never heard of. 

 

I should have also mentioned one of the games on the list that's actually released and supports DLSS is Anthem. 

[Middcore]

2 minutes ago, WikiForce said:

finally we have it, now they will kick Nvidia's ass. Hopefully they also bring navi refresh with G6X memory.

 

In March? When they're selling every card they make already and TSMC is at the limits of its capacity? 

 

 

[Dreckssackblase]

I really hope AMD comes up with a clever nickname for it because FidelityFX Super Resolution or FFXSR are a mouthful. 

[leadeater]

49 minutes ago, Dreckssackblase said:

I really hope AMD comes up with a clever nickname for it because FidelityFX Super Resolution or FFXSR are a mouthful. 

And then there is Square Enix with FFXIV:ARR lol

[D13H4RD]

8 hours ago, WikiForce said:

finally we have it, now they will kick Nvidia's ass. Hopefully they also bring navi refresh with G6X memory.

Unlikely. Not only does NVIDIA have dibs on G6X but AMD already has their own issues on supply.

[Master Disaster]

And it'll be another 12 months (at least) before most normal gamers are able to get their hands on a card that supports the feature.

 

I really cannot understand why they don't add support for these new techs (Smart Cache & Super Resolution) to older gen cards that people actually have right now.

[ZuppaSalata]

damn, I hoped it would also work on RX 5000 series

[RejZoR]

Until they figure out how to make DLSS and Super Resolution game agnostic, they are pretty useless even though they do create good visual and performance results.

 

Reason I say this is that while a lot of games have DLSS support NOW on release, there is a lot of games that would benefit from it, but don't support it. I mean, there are people who want to run games at 4K, 6K and 8K and even if games are older, they often run really poorly at such resolutions. And they'll never get DLSS because they were released 2 or 3 years ago. At those resolutions, even RTX 3090 might not run them at really high framerates, but such game agnostic enhancer could be useful.

 

AMD is actually closer to that with Radeon Boost. Because worst drop in performance is when you try to aim around. And they are doing that in a game agnostic way. They are making it even smarter, but I wonder how game agnostic that will be given the examples in Borderlands 3 were mentioned to have built-in functionality for that. So, again game specific like DLSS.

[LAwLz]

13 minutes ago, RejZoR said:

Until they figure out how to make DLSS and Super Resolution game agnostic, they are pretty useless even though they do create good visual and performance results.

Can't be done using their current technique, and I don't really see a way to make it work in any other way either.

 

15 minutes ago, RejZoR said:

Reason I say this is that while a lot of games have DLSS support NOW on release, there is a lot of games that would benefit from it, but don't support it. I mean, there are people who want to run games at 4K, 6K and 8K and even if games are older, they often run really poorly at such resolutions. And they'll never get DLSS because they were released 2 or 3 years ago. At those resolutions, even RTX 3090 might not run them at really high framerates, but such game agnostic enhancer could be useful.

I don't think this is an issue because you have to start somewhere. I mean, isn't that like saying we should never have moved on from DirectX 9 because "sure we can release DirectX 10, 11 and 12 if we want but current games won't get updated so why even bother."?

You have to start somewhere, and as time goes on the library of games that support newer features will grow. Hopefully to the point where pretty much all games support it. It's the same with any new feature that gets released in tech. Why release a 4 core processor when old games like Diablo 1 only support 1 core?

 

18 minutes ago, RejZoR said:

AMD is actually closer to that with Radeon Boost. Because worst drop in performance is when you try to aim around. And they are doing that in a game agnostic way. They are making it even smarter, but I wonder how game agnostic that will be given the examples in Borderlands 3 were mentioned to have built-in functionality for that. So, again game specific like DLSS.

Radeon Boost and FidelityFX are apples and oranges.

One reduces image quality to (hopefully) boost performance.

The other increases image quality (hopefully) at no loss of performance.

[Jet_ski]

I expect AMD’s FidelityFX to be on par or close to Nvidia’s DLSS. Mainly because most of the deep learning architectures are created by researchers who want them to be open and public. It’s only a matter of execution for AMD.

 

Come one AMD, give us a freakin CUDA alternative!

[RejZoR]

8 minutes ago, LAwLz said:

Can't be done using their current technique, and I don't really see a way to make it work in any other way either.

 

I don't think this is an issue because you have to start somewhere. I mean, isn't that like saying we should never have moved on from DirectX 9 because "sure we can release DirectX 10, 11 and 12 if we want but current games won't get updated so why even bother."?

You have to start somewhere, and as time goes on the library of games that support newer features will grow. Hopefully to the point where pretty much all games support it. It's the same with any new feature that gets released in tech. Why release a 4 core processor when old games like Diablo 1 only support 1 core?

 

Radeon Boost and FidelityFX are apples and oranges.

One reduces image quality to (hopefully) boost performance.

The other increases image quality (hopefully) at no loss of performance.

That was not my point. At all.

[leadeater]

5 minutes ago, Jet_ski said:

It’s only a matter of execution for AMD.

Sadly AMD's timeline of execution by measure of quality and completeness compares to Nvidia's is longer. I'm actually not hoping for much better than DLSS 1.0 for their first go around.

[Vishera]

The problem is that NVIDIA already laid out the framework for the adoption of RTX,AMD coming late to the game can kill the adoption of it's Ray Tracing technology.

Also,NVIDIA is not the only competitor,Unigine has a really good Ray Tracing technology,Cry Engine on the other hand doesn't have a good Ray Tracing technology.

 

Unigine Ray Tracing:

 

 

[Vishera]

1 minute ago, leadeater said:

Sadly AMD's timeline of execution by measure of quality and completeness compares to Nvidia's is longer. I'm actually not hoping for much better than DLSS 1.0 for their first go around.

Don't forget that NVIDIA is a giant in the AI industry,AMD has no chance to compete with DLSS.

[leadeater]

8 minutes ago, Vishera said:

Don't forget that NVIDIA is a giant in the AI industry,AMD has no chance to compete with DLSS.

Well at least in favor for AMD is a lot of the industry is open research and development so it's pretty level playing fielding in that aspect. So long as AMD manages to come up with a good implementation strategy they should still do well enough, where I think they'll be massively behind Nvidia is in the size of the development team and access to resources to continually refine and improve.

 

Having more people and more resources only gains you so much before that benefit diminishes with more of it, so at least for now I think the biggest advantage Nvidia has was being first and by a long time too.

[LAwLz]

26 minutes ago, RejZoR said:

That was not my point. At all.

Then what was your point?

[Jet_ski]

52 minutes ago, gabrielcarvfer said:

HIP? https://github.com/ROCm-Developer-Tools/HIPIFY

I hadn’t tried this. I’ve been told by people smarter than me that it’s way more headache than it’s worth. But unless AMD used Nvidia GPUs to train their FFX, they must have built at least the foundations for their alternative to CUDA.

[Doobeedoo]

This will be interesting to see, it will take time to improve upon it self, but hey. Running latest and new games maxed out at 4K to have say 60-120fps is quite a task, for now.

[leadeater]

2 hours ago, Jet_ski said:

I hadn’t tried this. I’ve been told by people smarter than me that it’s way more headache than it’s worth. But unless AMD used Nvidia GPUs to train their FFX, they must have built at least the foundations for their alternative to CUDA.

The most base level alternative is OpenCL and has been around for a long time. The problem isn't that there isn't an alternative it's that compared to CUDA they suck to use and everyone already knows CUDA or can more easily learn CUDA.

[Fnige]

10 hours ago, LAwLz said:

Can't be done using their current technique, and I don't really see a way to make it work in any other way either.

Idea: Make resolutions lower in certain areas where there isnt much difference in colour.

 

Have 2 settings for this. One being for colour difference before it makes a new pixel and another for the pixel size.

 

It would work well for 2D games though idk about 3D games

[thechinchinsong]

9 hours ago, leadeater said:

The most base level alternative is OpenCL and has been around for a long time. The problem isn't that there isn't an alternative it's that compared to CUDA they suck to use and everyone already knows CUDA or can more easily learn CUDA.

Building an good, viable competitor to CUDA is like building a good viable competitor against x86. It took years of development by the largest companies in the world, thorough planning, good execution, an amazing software ecosystem, and a large, loyal, profitable fanbase. It's very grim since the whole industry practically runs CUDA as a standard. Why go with something that is new, unproven, and not even as performant?

[leadeater]

3 hours ago, SlimyPython said:

Idea: Make resolutions lower in certain areas where there isnt much difference in colour.

Games are starting to do this, forget the proper name for it. LOD is reduced for less important objects and areas of the screen. I'm sure someone will chip in with the correct name.