AMD Reveals More FidelityFX Super Resolution 2.0 Technical Details

[Uskompuf]

AMD has recently shared some more technical information on their recently announced FidelityFX Super Resolution successor ahead of their GDC presentation. The GPUOpen developer website has been updated with new details regarding the various quality modes and implementation strategies available for FSR 2.0. AMD has confirmed that FSR 2.0 will work with graphics cards from AMD and NVIDIA going back to the Pascal GTX 10 series in addition to the Xbox Series X/S.

The technology is similar to NVIDIA DLSS 2.0 in that they both use temporal upscaling however FSR 2.0 does not rely on Machine Learning which makes the technology easier to integrate and allows more devices to be supported. FSR 2.0 will feature four quality options with Quality, Balanced, and Performance in addition to the optional Ultra-Performance mode that developers can choose to add. AMD has also included support for Dynamic Resolution Scaling where the resolution can be automatically scaled to meet a minimum target frame rate.





Implementation

The FSR 2.0 technology can easily be added to games that already supports NVIDIA DLSS 2.0 with AMD advising that implementation should take less than 3 days for such cases. The FSR 2.0 plugin for Unreal Engine 4/5 can also be used to quickly add support to games while a full implementation for games without any support for decoupled display or motion vectors can take upwards of 4 weeks.



Supported Games

The first two games to feature FSR 2.0 will be DEATHLOOP from Arkane Lyon and Forspoken by Luminous Productions. AMD has provided some updated comparison screenshots for the various FSR 2.0 quality modes in DEATHLOOP along with a preview video. The original uncompressed comparison images can be download here (155 MB).

Performance

The first performance figures for the Quality and Performance modes have also been provided by AMD for 4K, 1440p, and 1080p applications with various graphics cards. The minimum recommended hardware for 4K is the Radeon RX 5700/GeForce RTX 2070, while for 1440p it is the Radeon RX 5600/GeForce GTX 1080 and for 1080p it is the Radeon RX 590/GeForce GTX 1070.

Performance Mode

• Includes auto-exposure, no sharpening.

Quality Mode

• Includes auto-exposure, no sharpening.

Availability

AMD will make FSR 2.0 fully open-source in the near future under the same MIT license as the original FSR meaning that developers will be free to modify and redistribute the software commercially. The first PC games featuring FSR 2.0 will be released in Q2 2022 while Xbox Series X/S games will come at a later date. The AMD FidelityFX Super Resolution 2.0 GDC presentation video that will be released later today will include further technical details for the technology.

View at TechPowerUp Main Site | Source

 

[Thimblewad]

Awesome!

 

[Deleted member 185088]

Great news, even better with xbox series getting the technology, hopefully it'll accelerate it adoption, and allow games on xbox to all have a 60fps mode with good picture quality on a 4k TV.

 

[mama]

Looking forward to the comparison with DLSS 2.0.

 

[watzupken]

The benefit of FSR 1.0 is really because it is easy to implement. I am not too sure that still is the case with FSR 2.0. I feel game developers will have a hard time if they had to optimise their games for DLSS, FSR and XeSS. Perhaps FSR 2.0 may solve this problem if it is the easiest to implement, but then again, most games are sponsored by one of the 2 GPU makers now, and 3 going forward, so it is highly likely that there will be no 1 fixed standard.

 

[Durvelle27]

I’m exited to see what this allows the Xbox to achieve

 

[wolf]

The benefit of FSR 1.0 is really because it is easy to implement. I am not too sure that still is the case with FSR 2.0. I feel game developers will have a hard time if they had to optimise their games for DLSS, FSR and XeSS. Perhaps FSR 2.0 may solve this problem if it is the easiest to implement, but then again, most games are sponsored by one of the 2 GPU makers now, and 3 going forward, so it is highly likely that there will be no 1 fixed standard.

I would hope that if they can and have/will implement DLSS, they would also do FSR 2.0, and vice versa, and same with XeSS. As the inputs both need would be so similar, a good chunk of the work needed to implement any one could drastically reduce the effort to then add any more. I say this with 0 experience implementing anything into any game.

Only time will tell if one surfaces as the dominant tech and the others totally fall away, and that's before seeing what might change for any of them too, ie DLSS opened up and running on everything, XeSS having superior IQ to them both, more competing reconstruction/upscalers join the race etc. We don't know what we don't know and the future could hold many yet-not-thought-of developments.

 

[GamerGuy]

DLSS running on everything? I don't think nvidia would want anything to do with this, how old they sell their Tensor equipped GPUs then? Besides, there're some holdouts who've insist it's DLSS or nothing.

 

[stimpy88]

Most "A.I." is a lie anyway.

 

[ModEl4]

Is Polaris compatibility enabled?
RX 590 mentioned only at 1080p ultra performance (360p input res) are they gonna enable it in all the polaris models?

https://cdn.videocardz.com/1/2022/03/AMD-FSR-GPU-SUPPORT.png

 

[Punkenjoy]

Is Polaris compatibility enabled?
RX 590 mentioned only at 1080p ultra performance (360p input res) are they gonna enable it in all the polaris models?

https://cdn.videocardz.com/1/2022/03/AMD-FSR-GPU-SUPPORT.png

I suspect it's going to work fine with other polaris cards, but they might not have the horse power to run newer games at lower resolutions. More specifically looking at RX 460 and maybe RX 470/570 . I think the RX 580 will probably be fine.

One interesting thing is AMD is claiming not using AI is an advantages. It have advantages but it have draw back too. Having tailored algorithm is probably more flexible and you can see what happen under the hood where AI have to be trained and it will behave the way it learned and you don't really know how that affect all scenario.

On the other hands, you need very smart people to adapt to all the case and fine tune it.

We will see