AMD Officially Unveils FidelityFX Super Resolution: Open Source Gaming Scaling
In addition to announcing its new GPUs for Radeon RX 6000M laptops, AMD is also using Computex tonight to officially unveil its FidelityFX Super Resolution technology. Gaming scaling technology has garnered a lot of interest since AMD first announced last year that they were working on the technology, today AMD finally lifts the veil on the technology, at least for a brief moment. Overall today is more of a teaser of what’s to come on June 22nd.sd, while AMD plans to reveal more information on the technology, but for now, this is the largest release of information on the technology since AMD’s initial announcement.
To quickly recap, AMD announced FidelityFX Super Resolution last year as part of the launch of the Radeon RX 6000 series. The technology in development was intended as AMD’s response to the increasingly popular Deep Learning Super technology. Sampling (DLSS) from NVIDIA, an advanced image scaling technique that NVIDIA introduced to allow their GPUs to render games at a lower resolution (and therefore higher performance) without the severe blow to the image quality. Research on DLSS and similar smart scaling techniques has become increasingly intense, as scaling offers a tantalizing way to improve game performance via what is a post effect. – relatively inexpensive processing.
While NVIDIA seems to have finally hit its stride with DLSS 2.0, the downside for anyone outside of NVIDIA is that it is NVIDIA-only technology. Which, for AMD, means this is another value-added software feature from NVIDIA that NVIDIA can use to outsmart AMD. And while it is not strictly necessary for AMD to match NVIDIA on a one-to-one software feature base, it’s clear that DLSS is the start of a bigger change in the gaming rendering landscape, it is. is therefore an area where AMD will try to catch up. , both to negate an NVIDIA advantage and to give PC game developers another tool in their arsenal for better performance.
And that’s how FidelityFX Super Resolution was born. While today is more of a teaser than a detailed technical breakdown, AMD is confirming a few major facts regarding their approach to smart game scaling.
First of all, FSR, as AMD likes to call it, will be another of AMD’s GPUOpen technologies, meaning it will be released as open source and free to developers. And not only will developers be able to use it on AMD GPUs, they will also be able to use it on NVIDIA GPUs.
AMD doesn’t go into the specific technical underpinnings of the runtime model here – I’m assuming it’s implemented as a shader – but they do confirm that it doesn’t require any kind of tensor or other deep learning material. As a result, the technology can be used not only on recent AMD Radeon RX 6000 series cards, but also on the RX 5000, RX 500 and Vega series. Meanwhile, while not officially supported to the same extent on NVIDIA cards, according to AMD FSR will work (on day one) on NVIDIA cards dating back as far as the Pascal-based GTX 10 series (which precedes DLSS Support). In fact, the only modern graphics hardware that is not supported at this stage are the current generation game consoles; AMD might get there one day, but for now they’re focusing on the PC side.
At this point, AMD is not disclosing which games will support the technology, but the current message is that developers will need to take an active role in implementing the technology. That is, it doesn’t look like it can simply be applied in a fully post-processing fashion to existing games like AMD’s Adaptive Contrast Sharpening technology.
Following its June 22sd launch, AMD will release FSR on GPUOpen. Overall, the company says more than 10 “game studios and engines” in 2021 will implement FSR, with more details coming on the 22nd.sd. Expect to see Godfall among these games, as AMD is using him as a sample game for today’s announcement.
Afterwards, AMD also reveals that FSR will have four quality modes. Similar to DLSS, I would expect these modes to be all based on the upscaling factor used, and the smaller the upscaling factor used (the closer you are to native resolution), the lower the upscaling factor is. quality is high. Formally, these modes are Ultra Quality, Quality, Balanced and Performance modes.
For today’s Computex presentation, AMD is releasing the performance numbers for Godfall, running on a 6800XT. This card gets 49 fps when running in 4K with the “epic” picture quality preset and ray tracing. This goes up to 78 fps with FST in ultra quality mode, 99 fps in quality mode, 124 fps in balanced mode and 150 fps in performance mode. The exact advantage will depend on the card and game used, of course, but overall AMD touts the technology’s performance mode as improving game performance more than 2 times over native 4K rendering.
However, the million dollar question – and the question AMD won’t really answer until the 22ndsd – is what the resulting image quality of FSR will look like. Like other scaling techniques, FSR will live or die depending on how clean the image it produces. The scaling techniques here aim for “good enough” results, so they do not need to match native quality, but they should be sufficient to produce a reasonably sharp image without serious spatial or temporal artifacts. .
And, to switch to editorial mode, this is where AMD worries me a bit. During our pre-briefing with AMD, the company confirmed that FSR will be a purely spatial scaling technology; it will work frame by frame, without taking into account the motion data (motion vectors) of the game itself.
For GPU junkies, many of you will recognize that this is a strategy similar to how NVIDIA designed DLSS 1.0, which involved spatial scaling using pre-built neural network models. -trained and game-specific. DLSS 1.0 was ultimately a failure – it could not consistently produce acceptable results, and time artifacts were all too common. It wasn’t until NVIDIA introduced DLSS 2.0, a dramatically expanded version of the technology that incorporated motion vector data (essentially creating Temporal AA on steroids), that they finally got DLSS as we know it in order. Steps.
Considering NVIDIA’s experience with spatial-only scaling, I’m concerned that AMD will repeat NVIDIA’s early mistakes. Spatial is much easier to do on the backend – and requires a lot less developer work – but the lack of motion vector data presents some challenges. In particular, motion vectors are the traditional solution to countering temporal artifacts in TAA / DLSS, ensuring there are no frame-to-frame quirks or other rendering errors from moving objects. . Which is not to say that spatial-only scaling cannot work, only that, if it is competitive in image quality with DLSS, that would be a big first for AMD.
Unfortunately, AMD is doing itself a disservice in this regard with today’s presentation. In their slideshow there is a single split image with FSR apparently enabled, which they use for a performance comparison of the GTX 1060. I went ahead and extracted the raw image from the slide set. released to the press and uploaded it here, to try to preserve image quality as much as possible. (Don’t forget to click on it to see the photo in full resolution)
Taking a hit at NVIDIA by comparing the GTX 1060 running at native 1440p versus FSR in quality mode, the demo slide shows that performance is significantly improved, taking the GTX 1060 from 27 fps to 38 fps. Unfortunately, the success of the picture quality is quite noticeable here. The building and bridge are more blurry here than in the native resolution example, and the tree in the background – which is made up of lots of fine detail – easily gives up on the fact that it’s running at a lower resolution.
That said, since I haven’t seen the technology in person in motion yet, I’m not able to argue how good it all looks in action. But at least for static screenshots, it raises eyebrows. Although I’m giving credit to AMD here for posting a (seemingly honest) screenshot like this, rather than picking something that would overstep FSR.
In any case, there is a lot more technical information about FSR that we haven’t seen yet, and that AMD will present on June 22.sd. And, given the open source nature of FSR, AMD has little incentive or ability to withhold technical information for too long. So the questions raised today with their brief teaser will be a good talking point for the actual launch of FSR.
Finally, ahead of FSR’s launch, AMD is soliciting feedback on which games users would like to see supported by the technology. AMD is likely going to use it to guide its developer relations priorities, so if you’re interested in providing feedback, be sure to stop by. AMD FSR survey site.
Overall, it goes without saying that a lot of people are eager to see what AMD can do with game-scaling technology, both to level the playing field with NVIDIA and for the bring in a wider range of equipment and games. This is an ambitious project by AMD, and it will no doubt be interesting to see how it all falls into place when AMD launches the technology on June 22.sd. So be sure to stay tuned for it!