• Post category:Computer
  • Post comments:0 Comments
  • Post author:
  • Post published:23/06/2021
  • Post last modified:23/06/2021

Tweet

AMD FidelityFX Super Resolution

AMD unleashes its FidelityFX Super Resolution technology as an answer to combat with DLSS from NVIDIA. Will the new feature make enough of a difference, and is image quality, not an issue? Let’s try and see so you know what you can expect. 

Why FideltyFX Super Resolution?

We’ve mentioned this in all our Radeon RX 6000 series reviews; currently, AMD’s most significant deficit is that they do not have a solution at hand that matches DLSS from NVIDIA. NVIDIA wins every, and each time a game adds DLSS. NVIDIA did that very clever; they added Tensor cores in hardware on NVIDIA’s RTX series over two years ago, which will always work out better for NVIDIA. So here we are; a few years later, AMD had decided not to implement a hardware solution as NVIDIA does. As a result of negative feedback, they needed to develop all kinds of ominous ideas to answer the lack of that feature. 

I mean, when you think about it concerning technologies like DLSS or DirectML, AMD always will be behind in either image quality or performance. Even with pending DirectML support and now technology like FidelityFX Super Resolution, you need to realize; it still needs to be run over the existing compute engine as some sort of shader code. So no matter how we look at it, there will be a compromise, whereas NVIDIA can offload its ML supersampling to the tensor cores. It’s the sole reason why they implemented these specific DL/AI cores in the first place. Up to last week, we’ve not been 100% sure how AMD is applying FideltyFX Super-Resolution. Even after a briefing with AMD, they did not precisely explain that, but logic dictates this is a simply shader running over the render engine, with two extra passes.

AMD’s implementation

AMD presented an ‘answer’ towards NVIDIA’s DLSS and unveiled its FidelityFX Super Resolution. Nowhere in AMD’s documentation though AMD mentions DLSS, as they know, they’re not even in the same or similar ballpark. Where DLSS is a deep learning technique that uses artificial intelligence to improve rendered frames, FidelityFX Super Resolution is merely an algorithmic spatial scaler.

 

 

FSR does not use any machine learning or inference. While it is an interesting instrument that creates additional performance, we think we can already conclude that it is in no way comparable to an AI-powered image scaling system. However, my first impression of the technique applied is not at all negative; FidelityFX Super Resolution in its  ‘Ultra Quality’ mode certainly looks okay, but there’s bound to be a reduction in the visual quality of the games. In AMD’s defense, you can argue that DLSS also showed minor image deterioration in the early stages of the technology. The technology allows you to choose from four presets performance, balanced, quality, and ultra quality mode. 

So please do understand this well, FSR isn’t a representation of DLSS; Nvidia’s tech works by training an artificial intelligence (A.I.) model with high-quality scans, which they perform offline (try to imagine 16K still frames where their algorithm is learning from). DLSS can then fill in the extra information in a frame using that A.I. model. On its end, FSR uses a common and fixed supersampling technique with one upscaling and one sharpening pass.

 

 

A few weeks ago, AMD tried to explain in a complicated and confusing manner that it uses linear and non-linear upscaling facilitates both the preservation of the color and larger-scale features (large objects and shapes that are more easily observed by the human eye) of the image from linear upscaling as well as the preservation of finer features (e.g., curved features and features that are not effortlessly observed in low resolution) of the image from non-linear upscaling. The original (lower-res) image will pass through a linear upscaling network and a non-linear upscaling network at the same time, extracting different bits of information through each path. FSR blends these images to produce a pixel grid, then expands on each pixel for a higher-res image. After frame clean-up, you have your refurbished image. In the end, I would label this entire paragraph as a spatial upscaling technology with an added sharpening pass, as simple as that.

   

FSR Quality modeScale factorInput resolution for 2560X1440
FSR output
Input resolution for 3860×2140
FSR output
Ultra Quality1.3x1970×11082954×1662
Quality1.5x1706×9602560×1440
Balanced1.7x1506×8472259×1270
Performance2.0x1280×7201920×1080

  

FSR is open-source and will be part of GPUOpen; it can also run on Nvidia GPUs, including 10-series models that don’t support DLSS. AMD is claiming roughly a great performance increase even in the quality mode for games like Godfall on a GTX 1060, for example, boosting the frame rate from 27fps to 38fps. And that is a thing; games need to be supported as the developer needs to add the technique in the render engine. By the way, FSR is supported in Directx 11, 12, and VULKAN.

Hardware support

This is open source, and the games that implement it will also support it on NVIDIA hardware. As it stands on AMD’s side, supported are AMD Radeon RX 6000 series cards, but also the RX 5000 series, RX 500, 400, and Vega. On NVIDIA’s side, cards running backward to pascal (GTX 1000 series) would be compatible.

Game support

I mentioned this two paragraphs ago, but one of the bigger issues with FideltyFX Super Resolution is that it’ll need to be supported within the actual game itself, so your scope of games will be limited. Starting today, it’ll be part of the OpenGPU FideltyFX suite, and history taught us that only a few games implement FidelityFX in their engine. However, there is a positive note to be made here, the tech can boost framerates significantly, not just on OC but also AMD-powered consoles. If that takes off, you’ll see more widespread support for this technology in at the very least console ports.

While NVIDIA is paving away hard and fast with DLSS getting game support. Generally speaking, it’s not any different for NVIDIA as they need to whitelist the games that support this technology. Developers will be responsible for implementing the FSR compatibility in their games. AMD announced that the first batch of FSR games is scheduled to arrive in late June, with an update coming on release day — June 22. Over 10 game developers plan on using FSR in their games and engines in 2021. So the catch is simply: Not all games will be compatible, and perhaps we should reverse this; most games, if not, the vast majority of them likely will never support it.

1

Related Articles

Leave a Reply