Wednesday, March 16th 2022
AMD Set to Announce FSR 2.0 Featuring Temporal Upscaling on March 17th
AMD is preparing to announce their FidelityFX Super Resolution (FSR) successor tomorrow, on March 17th before a showcase of the technology at GDC 2022 as we previously reported on according to leaked slides obtained by VideoCardz. AMD FSR 2.0 will use temporal data and optimized anti-aliasing to improve image quality in all presents and resolutions compared to its predecessor making it a worthy component against NVIDIA DLSS 2.0. The slides also confirm that FSR 2.0 doesn't require dedicated machine learning hardware acceleration and will be compatible with a "wide range of products and platforms, both AMD and competitors".
The technology has been implemented in Deathloop where FSR 2.0 "Performance" mode with ray tracing increased frame rates from 53 FPS to 101 FPS compared to 4K native resolution with ray tracing. The slides do not reveal if AMD will make the source code for FSR 2.0 open-source as they have done for FSR and Intel is planning to do with XeSS. AMD is also expected to release Radeon Super Resolution which is an FSR driver implementation available for all games on March 17th.
Source:
VideoCardz
The technology has been implemented in Deathloop where FSR 2.0 "Performance" mode with ray tracing increased frame rates from 53 FPS to 101 FPS compared to 4K native resolution with ray tracing. The slides do not reveal if AMD will make the source code for FSR 2.0 open-source as they have done for FSR and Intel is planning to do with XeSS. AMD is also expected to release Radeon Super Resolution which is an FSR driver implementation available for all games on March 17th.
33 Comments on AMD Set to Announce FSR 2.0 Featuring Temporal Upscaling on March 17th
videocardz.com/newz/amd-teases-tomorrows-adrenalin-graphics-driver-with-radeon-super-resolution-launch
FSR 2.0 is a complete redesign of FSR now using temporal data and optimized anti-aliasing in a way which is much more similar to DLSS 2.0.
These piecemeal solutions from vendors are always going to be subject to weirdness and wonky ways.
The main competition will be Intel since also their solution will be open source and have also a DP4a version (+matrix solution)
I think Ada Lovelace in 6 months will probably introduce AI acceleration in many more fields other than resolution, having the capability to apply AI in fields like tesselation, texture generation, raytracing-path tracing (or general image synthesis like Intel's approach : isl-org.github.io/PhotorealismEnhancement/)
Performance being the "low end" where it will degrade image quality the most of these options, but therefore gives you the highest FPS boost
Ultra quality being the "high end" where it will degrade image quality the least, to the point of almost not even noticeable, but only gives a minor FPS boost
so basically they put it on the mode that gives the max FPS boost bc that higher number is what makes everyone's epeens hard. it's obviously a marketing tactic bc saying you only get 5FPS boost in ultra quality doesn't sound as good as 50FPS boost in performance.
FSR Ultra Quality - big, useful FPS bump over native, noticeable quality degradation when pixel peeping over native on fine edges where bad TAA is made worse. I don't notice this while actually playing a game (CP2077, HZD)
FRS Quality - almost no FPS improvement over Ultra Q, quality degradation and softness more noticeable but less so when actually playing the game. Not worth it, just use Ultra Q.
FRS Balanced - blurry, small FPS improvement, not worth it
FRS Performance - this setting is available
Cards tested:
RX 6600XT
RX 5600XT
GTX 1080
GTX 1050Ti
As for AA, Pixel density is important. I play at 4K on a 48 inch LG C1 OLED and i can see the pixel without AA. Indeed if you play at 4K on a 24 inch, you won't notice. Good AA increase image quality a lot. Bad AA degrade it significantly.
Also yes, resolving sub pixel content is better image quality than pixel popping it and out of existence because they don't align into the grid.
The statement that Native is the best image quality you can get will be made obsolete soon enough with more advance renderer. And when you start to get into VRS, sample count, mipmaps, ray count, filtering, denoising, etc, you start to realize that "native" becoming less relevant.
TAA type upscaler will allow engine to use more detailed pixel and reuse them. Using the same GPU resource, you would have to render less detailed pixel at native if you have to render them every frame.
the more the image is downscaled and therefore "upscaled" (for lack of knowing proper tech term) through FSR's magic, the higher the FPS bump. It works the same way that a resolution scale slider works. I'll pull arbitrary numbers outta my rear here for the analogy/argument's sake. For reference native res is 100%
Ultra Quality would be ~90%
Quality would be ~80%
Balanced would be ~65%
Performance would be ~50%
the lower the %, the higher the performance boost bc it is rendering at a lower resolution, then going through the FSR magic to improve the image fidelity. that's the super basic nut shell description of this feature
Also doing this on such a low res like 1080p is basically pointless. 1440p is feasible though. But really this whole system is designed for HIGHER resolutions like 4K to be used on cards or games that couldn't necessarily get better frame rates otherwise.
Think about this, 50% of 1080p would be close to same res of a PS2 or Gamecube....not surprising it seems you thought that option looked like garbage
This pic alone should explain it. If not, read this article too
www.tomshardware.com/reference/amd-fsr-fidelityfx-super-resolution-explained
There is more to picture quality than the aspects of the screen or the way it is rendered. At around 110-120 PPI, you should NOT be able to discern individual pixels, unless you actively go scanning for them with your head tilted towards the screen, leaning heavily forward from regular seating position. That, or you're sitting way too close to begin with.
We're already heavily in the realm of diminishing returns here. The advantage of increased pixel filtering down to subpixel level is nit picking territory, honestly, if you 'just game' 9 out of 10 times you won't have any kind of meaningfully different experience from someone gaming at 100 PPI 1080p regular 16:9 panels.
You see what you want to see, most of the time and our brain is an expert in making it so. The relevance of all these technologies is extremely low, apart from the performance uplift you can get out of it on a card that might just drop below playable FPS. Thát is the purpose, the image quality is bonus, because it allows games to expand more on assets and other detail levels without killing the gameplay. The funny thing is, many games really don't deliver those assets at all, they just upscale the same crap you had in 2010 - never mind the impact of dreadfully repetitive art direction. But regardless, these newer AA techniques are definitely closing in on that holy combo of minimal perf hit for maximum IQ. At the same time though, you can question how much we really need them with current high pixel density monitors.
In all honesty, the more you game, the less graphics matter IMHO. A pretty picture only impresses the first time. Good gameplay concepts never cease to amaze. We see a similar nitpicking mentality in the realm of high refresh where zealots are now defending the purpose of 240hz and higher, even if most games never get that stable.
Its marketing, guys.
I am aware of how FSR and scaling are supposed to work and I spent quite some time, especially with the 1050Ti, assessing performance and visual tradeoffs. I'll assume you've done the same and if you haven't, you should try and see for yourself.
I agree that using lower quality modes past Ultra Quality should result in steadily improving FPS but these improvements are small compared to the initial big FPS boost from native to UQ.
I wrote that Quality gives almost no FPS improvement over Ultra Quality. You seem to assume I meant it gave a lower FPS improvement over Native, which is not the case. As the image quality in Quality mode is noticeably lower than Ultra Q mode, I feel that UQ is the sweet spot for FSR 1.0, at least for the GPUs, games, and display resolutions I tested. If you're playing CP2077 on a 1050 Ti, Ultra Quality makes it reasonable to play at 1080p as the FPS improvement is better than the image fidelity loss. IMO of course, others may not like that tradeoff. However in the same computer, I felt that the TAA implementation in HZD was really distractingly bad in combination with FSR to where I mildly preferred the generally soft but not distracting presentation of 80% raw scaling over FSR UQ or Q at 1080p. Honestly, I didn't like either and even though HZD is a great game, I'd just play something else until I got a better card.
funny you say that, I have done the same approach on certain cards and games. when I got AC Odyssey it didn't play too hot on a RX 580 unless like 1080p low settings, not very good looking on a 4K TV, also didn't help I was on like a 10yr old X58 system at that time too so CPU bottleneck was a real problem, so I waited until I got a RX 5700 XT which went into a whole new Z390 system and could play something like 1440p high settings instead. sometimes, it's worth the wait to enjoy playing the game with better hardware. but FSR is here to help gamers in that exact situation, so I think it's a great piece of tech that hopefully can only get better from here.