Thursday, March 17th 2022

AMD FidelityFX Super Resolution 2.0 (FSR 2.0) Unveiled

AMD today unveiled FidelityFX Super Resolution 2.0 (FSR 2.0), the second major update to the company's ambitious gaming performance enhancement that seeks to improve performance with minimal loss of image quality, by lowering game render resolution, and applying a high-quality upscaling algorithm. The star-attraction with FSR 2.0 is its new upscaling algorithm that leverages temporal data. Data from past generated frames is used to predict and add details to the current frame. This improves image quality across all presets of FSR, and at all resolutions. In some cases, the "Performance" preset of FSR 2.0 may end up looking on par with the "Quality" preset of FSR 1.0. AMD's temporal upscaling algorithm doesn't need any machine-learning hardware on the GPU.

AMD is working to integrate FSR 2.0 with several contemporary game engines, and titles that are both already released or in development. It seems like an upcoming patch of "Deathloop" will add FSR 2.0 support, since the company exclusively used the game in its image-quality comparison previews. AMD will deliver a more elaborate presentation, with an under-the-hood look at how FSR 2.0 works, in its GDC 2022 presentation slated for March 23, 2022. The company is looking to release FSR 2.0 to gamers within Q2-2022 (between April to June).
Add your own comment

21 Comments on AMD FidelityFX Super Resolution 2.0 (FSR 2.0) Unveiled

#1
defaultluser
I'm excited, but because this will take months to get integrated, its going to be a boring slog!

At least we had launch-day games for the 1.0!
Posted on Reply
#2
Divide Overflow
defaultluserI'm excited, but because this will take months to get integrated, its going to be a boring slog!

At least we had launch-day games for the 1.0!
It's not out yet and it sounds like they are working to have games ready for it's launch. "The company is looking to release FSR 2.0 to gamers within Q2-2022 (between April to June)."
Posted on Reply
#3
Steevo
Vector adaptive accelerated video tech comes to the aid of gaming, I wonder if they realized FFT hardware with a sample pixel trace during Anisotropic filtering pass would hand them vector math at the same angle dependency as the AF pass which has been perfect for all vendors for years now. I would even almost guess that the difference in preset is merely the pixel sample box, 4X4, 8X8, 16X16
Posted on Reply
#4
Valantar
If those sample images are representative, this looks like a good improvement over 1.0. Hopefully it's not much more complicated to implement than 1.0 either (and has decent anti-ghosting mitigations built in).
Posted on Reply
#5
Chane
The stills look great, even performance mode. I'll need to see it in motion, though. I'm just happy it's a hardware-agnostic and hopefully that means broader implementation in games.
Posted on Reply
#6
Xuper
I created comparison , you can see more detail on wall and dome

Posted on Reply
#7
wolf
Performance Enthusiast
Very exited to see how good this is outside of marketing slides. I'll believe it's better than 1.0 for now, given its temporal that shouldn't be a hard bar to jump over. More options ftw.
Posted on Reply
#8
GamerGuy
wolfVery exited to see how good this is outside of marketing slides. I'll believe it's better than 1.0 for now, given its temporal that shouldn't be a hard bar to jump over. More options ftw.
Need to see FSR 2.0 in action, being temporal, I'm afraid it'd have the same ghosting effect as DLSS. Only time can tell, but I and my RX 6900 XT are ever so ready for this!
Posted on Reply
#9
Valantar
GamerGuyNeed to see FSR 2.0 in action, being temporal, I'm afraid it'd have the same ghosting effect as DLSS. Only time can tell, but I and my RX 6900 XT are ever so ready for this!
I would say I agree, except as a fellow 6900 XT owner we're probably among the people least in need of this :rolleyes:
Posted on Reply
#10
GamerGuy
ValantarI would say I agree, except as a fellow 6900 XT owner we're probably among the people least in need of this :rolleyes:
Oh, but there are titles that pushes my RX 6900 XT even at exactly half 4K, 3840x1080 to be precise. Metro Exodus PC Enhanced is a fine example especially with RT enabled, but unfortunately, nVidia has locked the game dev (4A?) into DLSS only.
Posted on Reply
#11
Valantar
GamerGuyOh, but there are titles that pushes my RX 6900 XT even at exactly half 4K, 3840x1080 to be precise. Metro Exodus PC Enhanced is a fine example especially with RT enabled, but unfortunately, nVidia has locked the game dev (4A?) into DLSS only.
Yeah, I enjoyed Metro: Exodus with everything cranked at 1440p, but I can imagine it struggling at 4k - there were areas where it did so at 1440p even. Still, we're still in a rather privileged position when our complaints are "my GPU can't handle every game with settings cranked at max resolution" :D
Posted on Reply
#12
medi01
ValantarIf those sample images are representative, this looks like a good improvement over 1.0. Hopefully it's not much more complicated to implement than 1.0 either (and has decent anti-ghosting mitigations built in).
Uh, spacial to TAA derivative, it sure is more complex (need motion vectors)
Posted on Reply
#13
Valantar
medi01Uh, spacial to TAA derivative, it sure is more complex (need motion vectors)
Sure, but how difficult are those to actually get out of the game? Can't AMD just write sample code for various engines and help devs adjust it as needed?
Posted on Reply
#14
ToTTenTranz
ValantarI would say I agree, except as a fellow 6900 XT owner we're probably among the people least in need of this :rolleyes:
Unless you're trying to play Cyberpunk with RT global illumination...
Posted on Reply
#15
ModEl4
Anandtech decided to update their FSR 2.0 article judging validity in a report from Computer base.de and although at first they thought (like most of the people) that the method is going to be DP4a based, now according to Computer base.de there is confirmation from AMD that the method they using is purely non-ML and DP4a-free (I suppose Anandtech crosschecked Computer base.de story with their AMD sources before updating their article) www.anandtech.com/show/17317/amd-teases-fsr-20-temporal-upscaling-tech-for-games-coming-in-q2
Supposedly FSR 2.0 is using advanced algorithms that can detect relationships between different frames and resolutions and produce much better results than FSR 1.0.
That's interesting, I guess we will know more in a few days after the GDC session.
www.computerbase.de/2022-03/fidelityfx-super-resolution-2.0-amd-fsr-2.0-stellt-sich-im-zweiten-quartal-gegen-dlss/
Posted on Reply
#16
wolf
Performance Enthusiast
GamerGuybeing temporal, I'm afraid it'd have the same ghosting effect as DLSS.
Can't say I ever really notice ghosting. If you're super duper sensitive to it you might not like FSR 2.0 then (and I suppose TAA in general) but it's never been even close to eye catching or bad for me using DLSS. I welcome our temporal future.
Posted on Reply
#17
defaultluser
Divide OverflowIt's not out yet and it sounds like they are working to have games ready for it's launch. "The company is looking to release FSR 2.0 to gamers within Q2-2022 (between April to June)."
When they announced 1.0, they had seven games ready-to-go

On launch, FSR supported the following seven games:
  • 22 Racing Series
  • Anno 1800™
  • Evil Genius® 2: World Domination
  • Godfal
  • Kingshunt
  • Terminator™: Resistance
  • The Riftbreaker™
The fact that there are none this time means we are months away; I guess lintels new FSR competitor forced them to rush this out?
Posted on Reply
#18
Valantar
defaultluserWhen they announced 1.0, they had seven games ready-to-go

On launch, FSR supported the following seven games:
  • 22 Racing Series
  • Anno 1800™
  • Evil Genius® 2: World Domination
  • Godfal
  • Kingshunt
  • Terminator™: Resistance
  • The Riftbreaker™
The fact that there are none this time means we are months away; I guess lintels new FSR competitor forced them to rush this out?
None? I believe Deathloop is a game? I see your point, but this might as well just be them not wanting to spoil things, especially knowing how variable game launch windows can be.
Posted on Reply
#19
ToTTenTranz
defaultluserThe fact that there are none this time means we are months away; I guess lintels new FSR competitor forced them to rush this out?
AMD used a very different approach between announcing FSR1.0 and FSR2.0.

For FSR1.0 they announced a soft-launch date weeks in advance,, to then show it off in a dedicated presentation. For FSR2.0 they updated their webpage and posted a couple of videos on youtube, without any warning.

AMD's marketing teams are steadily distancing themselves from Raja Koduri's methods. They now disclose the info when they're closer to put the solution in the market, which IMO is a good thing.
What we saw last week was a precursor to their GDC presentation which is happening the day after tomorrow. They probably mentioned only Deathloop because of some sort of cross-promotion deal, but I'd bet they'll be talking about other games adopting FSR2.0 during that presentation.

Most games that already implement motion vectors might have FSR2.0 enabled with a low development cost/time, and that would include all titles with DLSS2.
Posted on Reply
#21
ToTTenTranz
ToTTenTranzMost games that already implement motion vectors might have FSR2.0 enabled with a low development cost/time, and that would include all titles with DLSS2.
Lo and behold, going for games with DLSS2 to nullify Nvidia's proprietary solution is exactly what AMD is targeting with FSR 2.0.






Fastest way to implement FSR2.0 is to adapt a game that already has DLSS2. Less than 3 days' work on games that already support DLSS2.
Posted on Reply
Add your own comment
Jul 4th, 2022 08:32 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts