AMD shares more information about FSR 2.0 technology

0


AMD has provided new data on AMD FidelityFX Super Resolution (FSR) 2.0, the company’s widely accepted open-source cross-platform scaling technology.

FidelityFX Super Resolution 2.0 is a time-scaling solution coming Q2 2022 that delivers incredible image quality and increases frame rates in supported games using previous frame data to provide image quality similar to or better than native image quality at all resolutions.

You can find more details about AMD‘s dev session in the following blog, while key details from today’s session include:

  1. High quality scaling without machine learning: Although ML is a way to solve problems, it is not a prerequisite for obtaining good quality image scaling. AMD engineers have developed and optimized a hand-coded set of advanced algorithms to recognize and combine data from previous scenes, feeding a high-fidelity, upscaled image in an easily optimized way that can meet your a range of different scenarios. Importantly, not requiring dedicated ML hardware means more platforms can benefit from it and more gamers will be able to experience FSR 2.0.
  2. FSR 2.0 quality modes: FSR 2.0 will offer three main quality modes available in all games supporting the technology, and an optional mode for use by developers. These include ‘Quality’, ‘Balanced’ and ‘Performance’ modes, featured in today’s blog and updated screenshots. The optional fourth mode “Ultra Performance” is also available for developers who wish to offer the highest performance gain while maintaining image quality representative of native rendering.
  3. Hardware Support: Like FSR 1.0, FSR 2.0 maintains AMD’s open approach, supporting a wide range of AMD and competitor hardware, however, as a time-based solution, FSR 2.0 is more demanding, which means more hardware powerful and modern may be needed for higher scaling resolutions. More information about this can be found in the blog.
  4. How FSR 2.0 is integrated into games: Just like FSR 1.0, FSR 2.0 will be open source through the MIT license and will be available to developers as an intuitive and easy to use API that supports DirectX 12 and Vulkan. There will also be a plugin for Unreal Engine.
  5. More game support and FSR 2.0 on Xbox: In addition to DEATH LOOP, Speak will also include support for FSR 2.0 when the game launches in October 2022. FSR 2.0 will also be fully supported on Xbox and will be available in the Xbox GDK for registered developers to use in their games.

High-quality upscaling without machine learning?

When we announced the FSR 2.0 technology last week, we told you that it does not use machine learning (ML) in its scaling algorithm and therefore does not require any type of dedicated ML hardware in supported products and platforms. But since then we have seen that many of you have asked for more details on how we are able to achieve the results that FidelityFX Super Resolution 2.0 technology can deliver without ML.

Let’s add some context first – in any science, including software engineering, discoveries are made by analyzing data from experimentation, which yields mathematical models that can explain the results. These models can then be applied to new scenarios to make predictions. Generally speaking, machine learning (ML) is an incredibly useful set of tools and techniques that can make this process easier and faster. However, the results achieved by ML can sometimes not be the most optimal, lacking the spark of human imagination that can often lead to breakthroughs for complex problems.

Often ML-based real-time temporal upscalers use the learned model only to decide how to combine previous history samples to generate the upscaled image: there is usually no actual generation of new features from the recognition of shapes or objects in the scene. AMD engineers have leveraged their world-class expertise to research, develop and optimize a set of hand-coded advanced algorithms that map these relationships from the source and its historical data to improved resolutions.

The FidelityFX Super Resolution 2.0 analytical approach can offer significant advantages over ML solutions, such as better control to respond to a range of different scenarios and better ability to optimize. Importantly, not requiring dedicated ML hardware means more platforms can benefit from it and more gamers will be able to experience FSR 2.0.

FSR 2.0 quality modes

Like FSR 1.0, FSR 2.0 has different quality modes that allow you to customize the balance between image quality and performance to your preference. FSR 2.0 offers three main quality modes that you can expect to find in all games supporting the technology, and an optional mode that developers can use if they wish.

We’ve adjusted the modes slightly from FSR 1.0, with the highest quality setting now being “Quality” mode, which aligns with modes available in other popular time-scaling technologies. The other two main modes are “Balanced” and “Performance”, and you can see the details of each mode in the table below.







FSR 2.0
Quality Mode
The description Scale factor Contribution
resolution
Exit
resolution
Quality “Quality” The mode provides similar or better image quality than the original with a significant performance gain expected.

1.5x by size

(2.25x area scale)
(67% screen resolution)

1280×720
1706×960

2293×960
2560×1440
1920×1080
2560×1440
3440×1440

3840×2160
Balance “Balance” offers an ideal compromise between image quality and projected performance gains.

1.7x by size

(2.89x area scale)
(59% screen resolution)

1129×635
1506×847
2024×847
2259×1270
1920×1080
2560×1440

3440×1440
3840×2160
Performance

“Performance” The mode provides image quality similar to native image quality with a major performance gain intended.

2.0x by size

(4x area scale)
(50% screen resolution)

960×540
1280×720

1720×720
1920×1080
1920×1080
2560×1440

3440×1440
3840×2160

The optional fourth mode “Ultra Performance” is also available for developers who wish to include a mode designed to provide the ultimate in performance gains while maintaining image quality representative of native rendering.

FSR 2.0 also supports dynamic resolution scaling, where the input resolution from which to scale is determined (and dynamically adjusted) by the minimum frame rate you want always achieve in-game. Developers also have the option to apply RCAS sharpening to the output as well.

Download 4K PNG

As FSR 2.0 required all three data points in the diagram above at render resolution – depth, motion vectors, and color – developers will find it easier to integrate FSR 2.0 into games that already have a bet render path on a time scale. With that in mind, while FSR 2.0 is still easy for developers to add to their game like FSR 1.0, onboarding time estimates can vary – it can be as little as a few days for games that have already have the necessary time-scaling data in place. However, for games without motion vectors or without support for decoupled display and rendering resolutions, integration may take longer.

Just like FSR 1.0, FSR 2.0 will be open source via the MIT license and will be available to developers as an intuitive and easy to use API (with source provided via a library) that supports DirectX® 12 and Vulkan ®, and there will also be a plug-in for Unreal® Engine.


Performance and hardware support

FidelityFX Super Resolution 2.0 is designed to increase frame rates in supported games; however, it is fair to say that as FSR 2.0 is an advanced temporal scaling solution, it will be more demanding on graphics cards than a spatial scaling solution like FSR 1.0.

This means that although FSR 2.0 supports a wide range of hardware, both from AMD and some of our competitors’ hardware, there are some caveats gamers should be aware of compared to FSR 1.0. To begin with, however, we want to say that as an open-source, cross-platform solution, we place no constraints on FidelityFX Super Resolution 2.0 compatibility, and what we’re talking about here are recommendations for the best experience using the technology.

Noting the above, for the best FSR 2.0 experience, we have some recommendations for the starting level of graphics card for the various target enhanced display resolutions. One thing to keep in mind, though, is that your gaming experience may vary with these hardware suggestions. Depending on your specific system specifications, the system requirements of individual games that support FSR 2.0, and your target resolution, you may still be able to get a good upscaling experience on lesser GPUs. performing or older than those listed below.

Optimal starting level hardware FidelityFX Super Resolution 2.0







Target scaling resolution AMD graphics cards NVIDIA® graphics cards
4K

Radeon™ RX 6700 XT
Radeon™ RX 5700
(and above)

GeForce RTX™ 3070
GeForce RTX™ 2070
(and above)

1440P

Radeon™ RX 6600
Radeon™ RX 5600
Radeon™ RX Vega Series
(and above)

GeForce RTX™ 3060
GeForce RTX™ 2060
GeForce® GTX 1080
(and above)

1080P

Radeon™ RX 6500 XT
Radeon™ RX590
(and above)

GeForce® GTX 16 Series
GeForce® GTX 1070
(and above)

*Recommendations may change.





Share.

Comments are closed.