• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

AMD FidelityFX Super Resolution (FSR) launches June 22nd

vpance

Member
d5311111.png

We already know FSR sucks in performance mode. A TAA upsample to 4K from 1600P comparison would have been vastly more useful.
 

CrustyBritches

Gold Member
The first step for FSR would be to look better than native 1440p upscaled to 4k, while FSR is operating from 1440p base resolution. I don't think I saw that comparison in the DF video. They said 4K FSR(Quality) has 1440p base res and posted comparison shots for 4K and 1440p resolution in the article though...
1440p Native:
godfall-native1440p.jpg

4K FSR(Quality/1440p base):
godfall-FSR-Quality4k-1440pbase.jpg

1440p Native, 720p slice for forum viewing:
Godfall-1440p-Native-Upscale4-K-720p-Slice.jpg

4K FSR(Quality, 1440p base, 720 slice for forum viewing):
Godfall-4-KQuality-1440p-Base-720p-Slice.jpg

FSR holds up pretty well from these still shots. Shimmer in motion might be a different deal, and it's hard to ascertain performance between the 2(all the FSR shots were 60fps lock). I agree with DF that it does a decent job at preserving edges, but loses inner surface detail. Either way I'm looking forward to trying it out.
 
Last edited:

vpance

Member
For whatever reason DF's footage seems to show more of a difference between native 4K and FSR Ultra. Didn't see nearly as much in some other vids.
 

ethomaz

Banned
For whatever reason DF's footage seems to show more of a difference between native 4K and FSR Ultra. Didn't see nearly as much in some other vids.
From what I see the difference in FSR Ultra and native 4k is noticeable but it is a fine trade-off between quality and performance.
 

llien

Member
Wait... so this doesn't suck?
I've seen GN and HU reviews so far.
Pretty good at ultra quality.

It is also notable that although perf gain is on par with DLSS, the source resolution that FSR is upscaling from is actually higher.

GN noted lack of games that have both DLSS and FSR.

Taking TAA from mostly static frames is misleading (it's not a typical gaming take and it looks better than it is) yet HU fell for it.

Surprise: FSR runs on much much older hardware than announced (including 900 series GeForce and what not).


Overall: it seems that ultra quality mode would be liked by most gamers that are looking for improved framerates.
 

Unknown?

Member
checkerboarding is possibly superior to FSR in this state tbh

just render stuff at 1920x2160 and checkerboard them away... re:village does this fantastically and everyone are content with IQ. no point of pushing blurry upscaling...
Insomniacs temporal injection can't be beat on console at least. Hopefully more games use a similar style.
 

DaGwaphics

Member
I think Linus Tech Tips showed that in their video.

Thanks for pointing that one out. It does a better job demonstrating the feature for what it is meant to be than most of the other vids. If the tech can give a better result than native 720p (which it seems to), I wouldn't rule it out on XSS or for low-end GPUs.
 

IntentionalPun

Ask me about my wife's perfect butthole
I thought they sold their stuff, what NV is doing only makes sense if they license their solutions.I don't really understand what they're doing then.
Making something popular that only works on their hardware... sells their hardware.
 
Last edited:

elliot5

Member
Thanks for pointing that one out. It does a better job demonstrating the feature for what it is meant to be than most of the other vids. If the tech can give a better result than native 720p (which it seems to), I wouldn't rule it out on XSS or for low-end GPUs.
The thing is it doesn't gain much performance at all for arguable image quality improvement, because at those low low internal resolutions it's more CPU bound than anything.
 

Rikkori

Member
For whatever reason DF's footage seems to show more of a difference between native 4K and FSR Ultra. Didn't see nearly as much in some other vids.
All these technologies have pros and cons. If you want to make it look bad you can focus on cons and edge cases where it's weaker, and then you're not "lying" straight up but it is dishonest by omission. As soon as you understand that DF is literally paid by Nvidia as a channel, then things fall into place (this is not a conspiracy btw, they have disclosed this and you can see the videos).

If you want to know which publication is honest then go find out which one points out that DLSS cannot reconstruct raytraced reflections in WD:L & CP2077 instead of just jerking off to the "AI magic". Spoiler alert: DF isn't one of them.

UeykcQW.jpg
 
The first step for FSR would be to look better than native 1440p upscaled to 4k, while FSR is operating from 1440p base resolution. I don't think I saw that in the DF video. They said 4K FSR(Quality) has 1440p base res and posted comparison shots for 4K and 1440p resolution in the article though...
1440p Native:
godfall-native1440p.jpg

4K FSR(Quality/1440p base):
godfall-FSR-Quality4k-1440pbase.jpg

1440p Native, 720p slice for forum viewing:
Godfall-1440p-Native-Upscale4-K-720p-Slice.jpg

4K FSR(Quality, 1440p base, 720 slice for forum viewing):
Godfall-4-KQuality-1440p-Base-720p-Slice.jpg

FSR holds up pretty well from these still shots. Shimmer in motion might be a different deal, and it's hard to ascertain performance between the 2(all the FSR shots were 60fps lock). I agree with DF that it does a decent job at preserving edges, but loses inner surface detail. Either way I'm looking forward to trying it out.
Yeah from your shots it really does seem to refine a lot of details, but muddy up others, the stone on the base of the pillar for instance is a lot more defined with FSR, at the same point the orange writing on the small stone became too ‘chunky’ to define as well.

I‘m kinda wondering how well it would work as a fake supersample (render at native to the screen, uspcale up with FSR, then downscale it back down to native), kinda figure the shimmer effect would probably be a problem though, and adding TAA to the native image to mitigate it might just make it blurrier in all the wrong spots.
 
Last edited:

Reizo Ryuu

Member
FSR holds up pretty well from these still shots.
How did you get both shots in the same size? Did you upscale the 1440p shot to UHD before making the 720 slice for both?
Because the UHD FSR shot looks markedly sharper to me in many areas, I would've expected it to be the other way around.
 

CrustyBritches

Gold Member
How did you get both shots in the same size? Did you upscale the 1440p shot to UHD before making the 720 slice for both?
Because the UHD FSR shot looks markedly sharper to me in many areas, I would've expected it to be the other way around.
Sorry if I didn't clarify that this is from the DF article where they have a full comparison suite of 1440p and 4K shots.

They let the game upscale and captured the shots in native 4K. I downloaded the 1440p native and 4K FSR(Quality) shots and selected a 1280x720 portion of the image from each 4K shot and included it for use on the forum that typical scales the images down which has a supersampling effect and makes comparisons hard. You can match them against the originals to see they're unaltered. It's a forum thing.

From those still shots I'd say 4K FSR(Quality, 1440p base) looks improved over 1440p native. 4K FSR(Ultra) looks ok, too. It's meant to be used in a situation where 4K/60fps isn't possible and you're in the 30-40fps range. In that case it needs to look better than 1440p/1800p native or TAA upscaled. In Terminator it seems like suitable solution. We'll have to see how each game handles it.
 
Why the fuck every goddamn YouTuber looking into FSR has those half perfect crisp footage on one side and half giant pixellated shit on the other on EVRY fucking thumbnail? Quality and especially Ultra Quality setting is not at all blurry or pixellated unless extremely zoomed in. Like they like alluding it is complete shit, and making me wonder if they are getting paid by nV to do skew these videos however they can...
 
All these technologies have pros and cons. If you want to make it look bad you can focus on cons and edge cases where it's weaker, and then you're not "lying" straight up but it is dishonest by omission. As soon as you understand that DF is literally paid by Nvidia as a channel, then things fall into place (this is not a conspiracy btw, they have disclosed this and you can see the videos).

If you want to know which publication is honest then go find out which one points out that DLSS cannot reconstruct raytraced reflections in WD:L & CP2077 instead of just jerking off to the "AI magic". Spoiler alert: DF isn't one of them.

UeykcQW.jpg
CP2077's presets were fucked so the implementation of DLSS was terrible. It had texture resolution reduced for DLSS which messed up the DLSS reconstruction. You had to fiddle with it a bit to fix it.
 
It is extremely cheap to integrate and runs on any crap.
Most game devs will likely support it.
So are the temporal upscaling solutions they likely have already implemented in their engines though.

There's no reason to implement a spatial solution when you already have a superior temporal solution which runs on everything.

I think the only devs that will support it... are the ones AMD "sponsors"
 

Rikkori

Member
CP2077's presets were fucked so the implementation of DLSS was terrible. It had texture resolution reduced for DLSS which messed up the DLSS reconstruction. You had to fiddle with it a bit to fix it.
I am well aware of that, but first of all that's an image of WD:L lol. Secondly the lod bias doesn't affect effects such as reflections, so those effects will always be rendered at DLSS's starting resolution rather than what it wants to reconstruct to, f.ex. at 4K if you use DLSS Perf then those reflections will be 1080p in every way and not be affected by DLSS's reconstruction. In fact, it's closer to console quality than the pinnacle of PC power:

4k.jpg


xbox.jpg
dlss.jpg
 
Last edited:
I am well aware of that, but first of all that's an image of WD:L lol. Secondly the lod bias doesn't affect effects such as reflections, so those effects will always be rendered at DLSS's starting resolution rather than what it wants to reconstruct to, f.ex. at 4K if you use DLSS Perf then those reflections will be 1080p in every way and not be affected by DLSS's reconstruction. In fact, it's closer to console quality than the pinnacle of PC power:

4k.jpg


xbox.jpg
dlss.jpg
What exactly am i looking at here? Anyway reflections are a post processing effect AFAIK. Not sure if it possible for DLSS to have a fix for that but IIRC DLSS is a post processing effect so it is possible future versions fix this.
 
So what I understood from Steve is that since FSR is using just the single frame, the current one, and not a few frames aggregate like DLSS, it is in fact better in terms of ghosting, especially alpha textures in motion. So in the ghosting case alone FSR is better than even DLSS 2.0 or all the newer iterations since it is using temporal aggregate from the conception no matter what.
 

llien

Member
So what I understood from Steve is that since FSR is using just the single frame, the current one, and not a few frames aggregate like DLSS, it is in fact better in terms of ghosting, especially alpha textures in motion.
It is better in the sense of not introducing ghosting.
On the other hand, ghosting is the other side of TAA's strengths, it is the compromise one pays for having access to more frames.

CP2077's presets were fucked so the implementation of DLSS was terrible
In an NV sponsored game.
In other words: https://en.wikipedia.org/wiki/No_true_Scotsman
 

Zuzu

Member
I just tried the Riftbreaker demo and ran the GPU benchmark tool once at 4k native and then once at 4k with FidelityFX set to ultra. At 4k native I got an average of just under 58fps and at 4k with FidelityFX I got an average between 76fps and 77fps. That's a nice bit of improvement. The image quality still seemed very good to me with FidelityFX on.
 
Last edited:

MaKTaiL

Member
I just tried the Riftbreaker demo and ran the GPU benchmark tool once at 4k native and then once at 4k with FidelityFX set to ultra. At 4k native I got an average of just under 58fps and at 4k with FidelityFX I got an average between 76fps and 77fps. That's a nice bit of improvement. The image quality still seemed very good to me with FidelityFX on.
I just tried The RIftbreaker on my RX 580 as well. I ran at 4K FSR Performance Mode and it looks incredibly sharp on my screen. It ran at 60fps locked all the time. FSR is amazing IMO.
 
From all those videos I understand that the geometry sharpness, or sharpness at the edges to be correct, do not get any hits. It is the details within the textures, the sharpness on the texture details get a little blurry depending on FSR setting and how aggressive it is. With my monitor at a close distance like just one and a half feet, I can not tell native 4K and Ultra FSR apart. Testing with my 50" TV, at a distance where I normally sit I can't still tell them apart, only when I get closer to within a foot I can tell it gets a little blurry. So for me, all the FPS it can give at Ultra is worth it, whatever it is doing to get higher FPS I can't tell apart unless at a big screen at a very close distance. That is a win in my book. Wish it was game agnostic and could be turned on for everything at the driver level. I hope this being a free performance option would mean that a lot of developers include this in their games going forward.
 

EverydayBeast

thinks Halo Infinite is a new graphical benchmark
For once I’m on the casual gaming side they all literally look the same between the fps modes maybe I’m under estimating ray, hd settings, fps 120+ but it’s mostly to a certain degree the same.
 

yamaci17

Member
Because the things AMD does are never a black box people already discovered tricks like this:

jstpKMa.jpg




dunno what this has to do with nvidia dlss being a black box

you can use the same hack/trick on dlss too. in fact, i already to get superior supersampling on dlss supported games (dsr'ed to 4k and back to 1080p or 1440p with dlss performance/quality)

don't get me wrong btw, i'm not trying to discredit or defend anything here :)
 
Last edited:

Kenpachii

Member
Because the things AMD does are never a black box people already discovered tricks like this:

jstpKMa.jpg






Just use resolution scale, not a fan of DSR because u will require to drop borderless window mode, which practically for anybody with multiple screens ( practicually everybody on PC ) is a big no go.

I tried performance out with riftbreaker, more performance but blurry as shit. Ultra quality gives a bit better performance at 1080p but still quality degrades noticable. However with RT ultra in the game i get like 10-30 fps with drops on a 1080ti, with performance i get like 60-100 fps which is pretty darn impressive.

Not enough disk space to try anno out atm. However with anno CPU is the main bottleneck anyway, so yea don't see much use for FSR here.

I could see using higher resolution + FSR being a thing, same performance as 1080p but with higher base resolution.

The main issue so far is, games that need it don't support it, it seems. Also not sure why they choose godfall to demonstrate it. Absolute nobody plays it the same goes for terminator. Put it in games that are actually played on PC. Like battlefield games, AC games, Cyberpunk, Witcher etc etc.
 
Last edited:

RoyBatty

Banned
Two comparisons of mine

1440p native without CAS vs FSR Ultra

https://imgsli.com/NTg2NzQ

1440p native without CAS vs 2160 FSR Balanced

https://imgsli.com/NTg2NzU


How it works
FidelityFX Super Resolution is a spatial upscaler: it works by taking the current anti-aliased frame and upscaling it to display resolution without relying on other data such as frame history or motion vectors.
At the heart of FSR is a cutting-edge algorithm that detects and recreates high-resolution edges from the source image. Those high-resolution edges are a critical element required for turning the current frame into a “super resolution” image. FSR provides consistent upscaling quality regardless of whether the frame is in movement, which can provide quality advantages compared to other types of upscalers.
FSR is composed of two main passes:
  • An upscaling pass called EASU (Edge-Adaptive Spatial Upsampling) that also performs edge reconstruction. In this pass the input frame is analyzed and the main part of the algorithm detects gradient reversals – essentially looking at how neighboring gradients differ – from a set of input pixels. The intensity of the gradient reversals defines the weights to apply to the reconstructed pixels at display resolution.
  • A sharpening pass called RCAS (Robust Contrast-Adaptive Sharpening) that extracts pixel detail in the upscaled image.

EDIT:

FSR comparison on DotA 2

 
Last edited:

Hypno285

Banned
What the heck is the point of using DSR if you're going to use either DLSS or FSR?

You might as well run the game from that resolution that's higher then native at whatever DLSS or FSR runs the game at.
 
Last edited:

assurdum

Banned
Insomniacs temporal injection can't be beat on console at least. Hopefully more games use a similar style.
Temporal injection is not that good neither at very low resolution basis. They used it from the PS3 era and frankly in such console didn't improved almost anything. But surely around the 1440p DRS is very useful. It's not that miraculous though imo. I suspect FSR works better in comparison.
 
Last edited:

Armorian

Banned
What the heck is the point of using DSR if you're going to use either DLSS or FSR?

You might as well run the game from that resolution that's higher then native at whatever DLSS or FSR runs the game at.

You will get much better framerate with very similar picture quality

4x DSR is the only one worth using (with smoothness to 0%) so games without build in scales have to run in full resolution (5120x2160 in my case), but DSR provides amazing results even when internal res is lower than 4x it still looks great (much better than monitor native). So for example in Dirt 5 i Use 4x DSR with in game 75% internal res, FSR and DLSS provides more options (and better results than standard scaling).
 

Hypno285

Banned
You will get much better framerate with very similar picture quality

4x DSR is the only one worth using (with smoothness to 0%) so games without build in scales have to run in full resolution (5120x2160 in my case), but DSR provides amazing results even when internal res is lower than 4x it still looks great (much better than monitor native). So for example in Dirt 5 i Use 4x DSR with in game 75% internal res, FSR and DLSS provides more options (and better results than standard scaling).

DSR and VSR do the exact opposite of what FSR and DLSS do.

You might as well just use DSR and VSR alone instead of applying more vaseline to your screen.
 
Last edited:
Top Bottom