• 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) review roundup

Seeing Linus and Gamers Nexus in the OP.

Exit GIF
 

llien

Member
I'm not saying FreeSync is terrible.. or bad.. it's great. G-Sync just outdoes it a bit.
You are not getting it, are you?
Overdrive/etc to address ghosting is simply not part of the package.
Entire variable refresh rate thing is supported by all major upscaler chips FOR FREE.

Overdrive to fight ghosting is simply not part of the package.
 

llien

Member
wow so it's.... absolutely nothing.
Don't watch anything besides totally not paid shill DF, dude.
There are reviewers upscaling things with Adobe Premier and what not, but stick with billinear pics that look worse than in reviews by anyone else, besides DF.
 
Last edited:

rofif

Can’t Git Gud
Don't watch anything besides totally not paid shill DF, dude.
There are reviewers upscaling things with Adobe Premier and what not, but stick with billinear pics that look worse than in reviews by anyone else, besides DF.
You are the worst nvidia hate boner there is.
Clearly FSR is NOTHING and not up to par with DLSS.
You should be banned from all conversations like this. You are delusional and do anything to shill for AMD and shade on Nvidia.
You would say that white is black even if it was put in your face

it is clear that FSR is garbage while DLSS eliminates ghosting on foliage present with TAA, stabilizes motion clarity and pixel crawling. FSR does nothing of that. Just acts like worse smaa really
 
Last edited:

llien

Member
Clearly FSR is NOTHING and not up to par with DLSS.
Based on what? A single review out of, how many, 7 at this point?

A review which claimed that FSR causes ghosting, while TAA derivatives (DLSS 2 among them) does not?
A review that didn't show FPS gains from FSR? (showing GPU load is hilarious way to hide it)
Yeah, that's... legit.

I don't care who you are.
Are you fat or thing.
Tall or short.
Black or white.
Gay or straight.

In fact, it's not about you and me at all.
If you have arguments - go and state them.
If you feel too insecure to discuss upscaling (Jesus Christ), then don't discuss it.

If you want to discuss myself: I'm fine with that too, dude, just create a specific thread for that, I'll show up, I promise.
 

rofif

Can’t Git Gud
Based on what? A single review out of, how many, 7 at this point?

A review which claimed that FSR causes ghosting, while TAA derivatives (DLSS 2 among them) do not?
A review that didn't show FPS gains from FSR? (showing GPU load is hilarious way to hide it)
Yeah, that's... legit.


I don't care who you are.
Are you fat or thin.
Tall or short.
Black or white.
Gay or straight.
Hell, I don't even care if you are from Earth or outer space.

In fact, it's not about you and me at all.
If you have arguments - go and state them.
If you feel too insecure to discuss upscaling (Jesus Christ), then don't discuss it.

If you want to discuss myself: I'm fine with that too, dude, just create a specific thread for that, I'll show up, I promise.
Oh fuck off with the wholesome act.
Fps gains ?! If You run internal 1080p, I would expect it to run like a fucking 1080p if it looks like 1080p.
Better based on what? Based on my eyes and my comparisons and EVERYONE ELSES.
DLSS is the best anti aliasing technique there is. I repeat and show again and again and You are blind to anything.
The same arguments each month and You fail to see.

The fact that DLSS image is sharper and cleaner in motion than native and gives better fps results is really an achievement.
FSR looks just the same as resolution it upscales from with minimal edge enhancement. Even DLSS 1 looked years ahead.
In fact, DLSS in games like Death Stranding looks so good, it comes close to ground truth and I would expect it to be very heavy AA technique but it increases the fps.

For years, image sharpness and edge anti aliasing is not an issue.
The issue is with pixel crawling, in-texture aliasing, shader shimmering and image stability in motion or far away detail. DLSS deals with all of those and FSR does nothing.
I would love FSR to do anything special... this would mean we get better IQ on ps5 and xsex in the future but nope. Devs must use their own stuff.

edit:
ONE THING I will grant FSR is, if it's free and adds any sort of ege AA, that's good. Any more anti aliasing is a good thing. It's just nothing special
 
Last edited:

llien

Member
DLSS image is sharper and cleaner in motion than native
How convenient, ain't it?

On this very forum, one of your brethren, has challenged me to figure out which of the pics was DLSS quality and which true 4k.
It wasn't hard to figure, and, guess what, the one that added blur and wiped out details was the DLSS.

But "in motion" right? As in "go figure, how to compare". Very convenient.
Want to talk about ghosting? Or about how well TAA derivatives handle motion in general?

Even DLSS 1 looked years ahead.
That is some crazy stuff, but anyhow this thread, but true AI DLSS (the 1.0) is not relevant to this discussion, only TAA derivative with some ML, also known as 2.0 is.


Oh fuck off
Ok, let me say that using words you seem to understand better: fuck off with ad hominems, focus on arguments.

Better based on what?
Pretty much any review, bar DF.
 

rofif

Can’t Git Gud
How convenient, ain't it?

On this very forum, one of your brethren, has challenged me to figure out which of the pics was DLSS quality and which true 4k.
It wasn't hard to figure, and, guess what, the one that added blur and wiped out details was the DLSS 2.

But "in motion" right? As in "go figure, how to compare". Very convenient.
Want to talk about ghosting? Or about how well TAA derivatives handle motion in general?


That is some crazy stuff, but anyhow this thread, but true AI DLSS (the 1.0) is not relevant to this discussion, only TAA derivative with some ML, also known as 2.0 is.



Ok, let me say that using words you seem to understand better: fuck off with ad hominems, focus on arguments.


Pretty much any review, bar DF.
I give up.
My arch nemesis won. congrats
 

rofif

Can’t Git Gud
Congrats to your arch nemesis, whoever that is.

A pity you weren't able to answer rather concrete questions and preferred to give up.
I have no way of recording or proving anything that was proven a bunch of times already and I am just tired of arguing.
I just know that TAA ghosts like crap on foliage and crawls on distant, smaller lines.
While DLSS looks pristine in motion and even foliage looks great.
So if FSR is just a bit of upscaling on top of TAA, it's going to have all the same problems.... so why bother you know....
Sure, there are some broken motion vectors in Death Stranding example but other than that, I wish every game had this good image clarity.
When I play horizon afterwards, i just see the grass "pixelating" during camera panning and geometry lines far away turning into foating pixels.
And I don't think DLSS 2 is blurry. It might be a visual side effect of cleaning the image but it's not blur
 

FireFly

Member
PS
DF is VSynced to 60fps. And then demonstrated GPU usage difference to show the "performance impact". The fact that this piece of embarrassment is perceived as legit review, speaks for itself.
Based on the "GPU usage" method of measurement, FSR enjoys around a 3-6% performance advantage over TAAU at the same internal resolution. While, KitGuru's performance benchmarks in Godfall show that FSR Ultra is 3.7% ahead in Godfall at the 77% render scale from 4K, and 1.9% faster scaling from 1080p.
 

llien

Member
I just know that TAA ghosts like crap on foliage and crawls on distant, smaller lines.
While DLSS looks pristine in motion and even foliage looks great.
Let's separate 2 things here, shall we: we are talking about reviews. Reviews might be doing and focusing on things that singled out gamer might not pay attention to (or not even see that game).

You cannot counter common findings by "but I don't see it in a handful of games I've tried".

So, NV's DLSS 2 did address some of the TAA's weaknesses, but not fully, one cannot state that ghosting is simply gone (especially when quick motion is involved).
Can we agree on this, or you insist 2.0 is never having ghosting?

So if FSR is just a bit of upscaling on top of TAA
It's purely spacial, nothing temporal about it.
It cannot add ghosting in principle, as it doesn't reference other frames, it takes DF level shilling to see ghosting caused by FSR.
 

llien

Member
"GPU usage" method of measurement
Why? Can you recall a single other case when "GPU usage" method of measurement (with locked FPS...) was used to judge the perf gain?
What is that if not an attempt to mislead people?



godfall-4k-gains.png
 

rofif

Can’t Git Gud
Let's separate 2 things here, shall we: we are talking about reviews. Reviews might be doing and focusing on things that singled out gamer might not pay attention to (or not even see that game).

You cannot counter common findings by "but I don't see it in a handful of games I've tried".

So, NV's DLSS 2 did address some of the TAA's weaknesses, but not fully, one cannot state that ghosting is simply gone (especially when quick motion is involved).
Can we agree on this, or you insist 2.0 is never having ghosting?


It's purely spacial, nothing temporal about it.
It cannot add ghosting in principle, as it doesn't reference other frames, it takes DF level shilling to see ghosting caused by FSR.
It does not add ghosting as it's 1 frame. I know.
But it also does nto work to eliminate any ghosting, trailing, crawling etc.
It just feels like a pointless effort to be hones. Everyone was hoping AMD was working on something else
 

llien

Member
Pretty much a nail in the coffin of any argument these other "tech youtubers"
Pretty much every tech youtuber touched on it being spacial, so did TPU.

Oh, so did AMD:

fsr10.jpg

fsr13.jpg


 

kingkaiser

Member
Regarding the credibility of digital foundry I still wonder how well Rich must have been paid by Intel to advertise the shitty i5 11400f as some kind of budget gaming revelation while completely neglecting the mainboard issues with the Intel chips and also how ridiculous overpriced those mainboards are, which completely contradicts his whole "budget" narrative. Oh and by the way, if my message shouldn't be clear enough, I will spell it out: Rich is a crook.
 

llien

Member
It does not add ghosting as it's 1 frame. I know.
But it also does nto work to eliminate any ghosting, trailing, crawling etc.
It just feels like a pointless effort to be hones. Everyone was hoping AMD was working on something else

This is true. (and I don't want to dive on "D 2.0 can add ghosting", e.g. check HUB review from OP, fast motion is a harsh condition for temporal processing)
But I think you are missing the point of FSR.

1) It needs to give fair speed bump over minor quality drop - according to pretty much anyone, but DF, it does, all reviewers like d"ultra" setting, (which gives 25-40% perf uplift)
2) It needs to be easy to implement (see the 2 hours reference of an actual dev, top fop of the second page in this thread)
3) It runs on pretty much anything.

I don't think there is any doubt about points #2 and #3.
For #1 there essentially a consensus, bar DF.
If there is no fair speed for quality gain, nobody would use it.
We'll see it soon enough.


I don't know if you caught that base resolution that ultra is upscaling from is not 1440p. Since post-processing done by FSR is not power hungry, they can afford to achieve the same perf uplift, while using better source material:

fsr15.jpg
 

FireFly

Member
Why? Can you recall a single other case when "GPU usage" method of measurement (with locked FPS...) was used to judge the perf gain?
What is that if not an attempt to mislead people?
I don't think Alex was trying to judge the performance gain, but rather the performance cost of enabling FSR vs. TAAU. Since the topic was whether developers should be integrating FSR instead of TAAU. In fact I would say the video was more developer than gamer focused, which is a criticism you can make in general of Alex's content.
 

ZywyPL

Banned
You should have taken this part of the image

1.png


2.png


3.png



Hint... the best looking one is number 3 and it's the only one that uses a reconstruction technique. 😉

The middle one has the most detailed tree/wall textures tho, especially the wall on the right looks flat and washed out on the 3rd picture. That's why I'm not a fan of UE's TAA, it's suppose to bring back the clarity/details, but does the opposite by blurrying all the textures.


FSR still has a long way to go obviously, but the initial results are IMO promising - it does the job really well at 4K, which is exactly were you need such tech the most, Ultra Quality mode gives little to no loss in the image clarity and texture details like in the below two screenshots from Godfall, one with native res and one with FSR QU, while boosting up the framerate quite a bit. Sorry but no one will convince me that "just lowering the resolution" to 1600p will give me that kind of image clarity, not even remotely close.

Godfall_FSR-off.jpg


Godfall_ultra-qualiy.jpg
 

llien

Member
I don't think Alex was trying to judge the performance gain, but rather the performance cost of enabling FSR vs. TAAU
KitGuru guys directly compared it vs TAAU (no VSync 60fps misleading bazinga).
They found FSR to be a bit better IQ-wise and for the most part, also faster (same internal resolution).

 

llien

Member
That very video.
King Hunt.
Neither bilinear, nor FSR can cause temporal artifacts (both are spacial).
Yet DF video has that for FSR.

So they PURPOSEFULLY set up a scenario to make typical filtering and FSR look bad compared to TAAU and DLSS.
 

FireFly

Member
KitGuru guys directly compared it vs TAAU (no VSync 60fps misleading bazinga).
They found FSR to be a bit better IQ-wise and for the most part, also faster (same internal resolution).

That's where I took the performance figures from in my post. They say that TAAU looks fractionally sharper @ 77% of 4K in Godfall at the cost of some texture shimmering. At 1080p, they think TAAU provides a "better experience" @ 77%. FSR looks better in Terminator due to the low contrast content, as noted by other sites.

So the way I interpret that is that at 77% of 4K in "normal" contrast scenes, TAAU can look subjectively better or worse than FSR, but that the more you lower the resolution, the more the advantages of TAAU manifest. You can give that a "negative" or "positive" spin, depending on what your expectations were.
 

Shmunter

Member
I’ve checked about 3 of the reviews now. After DF I was expecting it to be shredded, but instead others are claiming it does what it sets out to do on balance.

Once again, reality shifts by focusing too heavily on minutia instead of the bigger picture. Meh.
 

llien

Member
So the way I interpret that is that at 77% of 4K in "normal" contrast scenes, TAAU can look subjectively better or worse than FSR, but that the more you lower the resolution, the more the advantages of TAAU manifest.
I think that's a fair assessment.

You can give that a "negative" or "positive" spin, depending on what your expectations were.
"I would prefer to use FSR for gaming" (stated verbatim) is a quite clearly positive spin.
 

CuNi

Member
Ilien thread being salty again?

Ilien thread bein salty again.
I said it in the other thread, FSR is new, it does a good job for being new.
It does not hold up against DLSS 2.X but DLSS is out for some time already and has seen improvement, so will FSR.
Only hope for AMD will be that DLSS 3.0 won't release too soon or once it comes out not totally make FSR obsolete.
 

Marlenus

Member
The issue I have with the DF video is Alex makes the blanket claim that TAAU is better than FSR without proving it for all input resolutions.

Another use case for FSR is to combine it with VSR/DSR. With this you can take a native image, use FSR performance to 2x upscale it and then let VSR/DSR supersample it back to native and it seems you get a nice IQ increase for almost no performance hit.
 

Bo_Hazem

Banned
Demons Souls 1440p TAA looks much blurrier than 4K FSR UQ. Loved the 60FPS but much preferred the crispness of native.

I’ve yet to play Ratchet but I’m sure it’s it’s one of the best of examples of DSR TAA.

Haven't played Demon's Souls myself, but that's even good news if true. FSR will be the standard then. It was already impressive how Returnal looks being from 1080p base.
 

vpance

Member
Haven't played Demon's Souls myself, but that's even good news if true. FSR will be the standard then. It was already impressive how Returnal looks being from 1080p base.

Yeah I think it’s a good option to have for consoles, especially the UQ preset. Tested it a bit with Anno and it’s pretty hard to spot the degradation at TV viewing distances.
 
Last edited:

M1chl

Currently Gif and Meme Champion
It should be supported with most games that lack solid upscaling like Insomniac games and Demon's Souls.
Well sadly in that case it's going to do fuck all, because for this to work, you already have to have quite high internal resolution, because FSR make the resolution lower and try to guess and reconstruct the edges. So no, we need something else for this.

Anyday now:
xbox-image-7607.jpg


Believe™

Hopefully they didn't meant FSR.

FSR made me realized how cruel I was to CB. Because in VILLAGE, in 60FPS that shit is absolutely pristine.
 
From what I've gathered the Ultra & quality modes are great for 4K, Ultra works well at 1440p - anything below that isn't quite good enough. Having an RTX GPU means I get to use both if needed so I'm happy enough it turned out better than DLSS 1.0 at least.

FSR made me realized how cruel I was to CB. Because in VILLAGE, in 60FPS that shit is absolutely pristine.

One thing I've noticed since I got a PS5 is that games like GOW & Days Gone, which used CB and have had a 60fps patch, look much cleaner than on Pro at 30fps. I guess having double the frames to work with really helps a technique like CB
 
No, 2.0 is 90% TAA sprinkled with some mild AI. (that is why it is exhibiting all the strengths and weaknesses of TAA)
What you describe was 1.0.

No that was 2.0.
I don't know what you were trying to imply with these pictures, mate, that 16k super resolution images and per game training is exclusively 1.0. (which everyone, but people watching DF exclusively have figured, sucks)
Or maybe you are dumb those pics are from NVidias own website.
And the picture shows it compares itself to a 16k image.


The NVIDIA DLSS 2.0 Architecture
A special type of AI network, called a convolutional autoencoder, takes the low resolution current frame, and the high resolution previous frame, to determine on a pixel-by-pixel basis how to generate a higher quality current frame.

During the training process, the output image is compared to an offline rendered, ultra-high quality 16K reference image, and the difference is communicated back into the network so that it can continue to learn and improve its results. This process is repeated tens of thousands of times on the supercomputer until the network reliably outputs high quality, high resolution images.

Once the network is trained, NGX delivers the AI model to your GeForce RTX PC or laptop via Game Ready Drivers and OTA updates. With Turing’s Tensor Cores delivering up to 110 teraflops of dedicated AI horsepower, the DLSS network can be run in real-time simultaneously with an intensive 3D game. This simply wasn’t possible before Turing and Tensor Cores.

This is the second time your tried to claim this bs. It's totally false.
 
Top Bottom