AMD's FSR 3 upscaling and frame interpolation *spawn

In the process of bashing the wrong target you forgot that framegen ("FSR3") can be used without scaling ("FSR2"), which frees you from those issues. But that's not important is it?
It can't, FSR3 is tied to FSR2, and FSR2 is horrible image quality wise, even in it's FSR2 native AA mode, which looks worse than DLSS2 Quality.

Only DLSS3 can work without scaling, DLSS3 can work on native resolution, and on FSR1/FSR2/XeSS/DLSS/DLAA images. DLSS3 generates frames from practically anything you throw at it.
 
Curious if AMD did it this way to stop people from using DLSS upscaling with FSR FrameGen for 2000/3000 series nvidia cards.
 
It can't, FSR3 is tied to FSR2, and FSR2 is horrible image quality wise, even in it's FSR2 native AA mode, which looks worse than DLSS2 Quality.
DLSS Quality is amazing, though. Calling something 'horrible' just for not matching that is a gross exaggeration. As if there's no middle ground between 'amazing' and 'terrible'. At higher resolutions and FSR settings, the image quality is still quite decent overall and very usable.

But yes, FSR frame gen being tied to FSR2 is definitely a frustrating drawback.
 
DLSS Quality is amazing, though. Calling something 'horrible' just for not matching that is a gross exaggeration. As if there's no middle ground between 'amazing' and 'terrible'. At higher resolutions and FSR settings, the image quality is still quite decent overall and very usable.

But yes, FSR frame gen being tied to FSR2 is definitely a frustrating drawback.
FSR is actually horrible. It is worse than the basic resolution scale sliders we used to have. Those would soften the image. FSR fills nearly every moving pixel of the screen with eye searing artifacts.
 
You might want to read up on FSR 3.
FSR Native AA != FSR Scaling and it's free of several issues caused by scaling (while it shares the other steps). Framegen works with native aa, it isn't tied to scaling.

edit: for clarifications sake, I don't stand these techs at all personally.

FSR is actually horrible. It is worse than the basic resolution scale sliders we used to have. Those would soften the image. FSR fills nearly every moving pixel of the screen with eye searing artifacts.
It's BS posts like this that really drive the negative tensions around here.
 
FSR Native AA != FSR Scaling and it's free of several issues caused by scaling (while it shares the other steps). Framegen works with native aa, it isn't tied to scaling.

edit: for clarifications sake, I don't stand these techs at all personally.


It's BS posts like this that really drive the negative tensions around here.
FSR native still looks worse than standard TAA. You're paying a 15% performance penalty to degrade your IQ.
 
FSR Native AA != FSR Scaling and it's free of several issues caused by scaling (while it shares the other steps). Framegen works with native aa, it isn't tied to scaling.

Unfortunately not really, at least from the examples I've seen so far. It still has the shimmering issues FSR2 can present, just less. Even DLAA can sometimes have artifacts that aren't present in TAA from my testing, but the majority of the other aspects are vastly improved so it's still usually a big win over native TAA. Haven't seen that with the FSR native examples I've viewed.
 
But yes, FSR frame gen being tied to FSR2 is definitely a frustrating drawback.
One more serious draw back though is the problem of FSR3 needing VSync to work properly. In that case, FSR3 is being used to max out the refresh rate of the monitor, which can and will increase latency.

First you get the added latency from VSync, but you also get another latency on top, as FSR3 will decrease your "real fps" in favor of generated fps. Let's say you are getting native 80fps on a 120Hz screen, enabling FSR3 with VSync on that screen will slash your real fps to 60fps, so that FSR3 can insert generated frames equally between real frames (FSR3 can't do fractions, it needs one generated frame for every real frame). So in effect you get your latency increased from going down from 80fps to 60fps.

So in total you get multitudes of latency piling up on top of each other: VSync, the frame generation itself, then the reduction of real fps, and Anti-Lag+ not working with FSR3.
 
Both games with FSR3 so far run the hud at internal frame-rate - I do not think many will opt to run the hud at native due to CPU concerns and how it does not make much sense If you have diagetic hud/3D HUD.
I don't have any clue, but surely they could be. The only thing is that AFMF has HUD artifacts, DLSS3 FG also does, but FSR3 FG don't have on both games, which points out that they are probably rendering at display frame rate, and CPU will not be a concern in that case.

The HUD can be rendered asynchronously on a different renderer thread on the CPU, games never fully utilize your CPU because that would cause your operating system to freeze and your inputs to not be computed.

The main rendering pipeline is very concerned about CPU because it needs to gather every data and state from the game in order to render it, while also having to compute physics on an internal tick rate. On the other hand, the HUD don't need half of this.

There will be corner cases for sure, but as far as my experience with graphics programming goes, your UI will not be concerned about CPU as long as you have enough threads available to off-load it, and do not have to deal with writes to data shared with the UI and the Scene during the rendering pipeline.
Unfortunately not really, at least from the examples I've seen so far. It still has the shimmering issues FSR2 can present, just less. Even DLAA can sometimes have artifacts that aren't present in TAA from my testing, but the majority of the other aspects are vastly improved so it's still usually a big win over native TAA. Haven't seen that with the FSR native examples I've viewed.
Yes, it'll be worse for some cases and better in the others. I hope AMD improves FSR2 image stability because it's indeed possible and is the major issue that FSR2 has at the moment.

I personally hate TAA, I just didn't got rid of it on Cyberpunk 2077 because the SSAO breaks, so I'm just using dynamic res with 90 max and min for internal resolution. The frame rate is lower but it solved a lot of the shimmering that the TAA implementation has on this game.
 
I don't consider it hyperbole at all. It's among the worst and most artifact ridden upscaling techniques in recent history.
At least provide some kind of source for your claims, like a video analysis showing TAA vs. FSR. Because the DF ones found it was resolving more detail.
 
I don't have any clue, but surely they could be. The only thing is that AFMF has HUD artifacts, DLSS3 FG also does, but FSR3 FG don't have on both games, which points out that they are probably rendering at display frame rate, and CPU will not be a concern in that case.
They are rendering at the internal frame-rate in both games AFAIK, not the display frame-rate.
 
At least provide some kind of source for your claims, like a video analysis showing TAA vs. FSR. Because the DF ones found it was resolving more detail.
My source is my eyes. I have watched every DF video and used it myself. It's an eye sore. Extreme ghosting, sizzling and disocclusion artifacts galore, not to mention the high image instability. This thread has already linked and discussed some of the recent examples.


I mean come on, this is just awful.
 
Last edited:
Interesting comparison between all 3 upscalers, especially regarding the native AA mode in AC Mirage.

FSR2 Upscaling: bad.
The FSR 2.2 image quality in this game is suffering from shimmering and flickering on three leaves, vegetation, distant shadows and thin lines, and these shimmering issues are visible even when standing still, across all resolutions and quality modes. The ghosting issues, mainly on small particle effects, and small amount of disocclusion artifacts around the main character and NPCs are also present with FSR 2.2 enabled. Also, the overall image has a very soft look with FSR 2.2 enabled in "Quality" mode across all resolutions, even if the sharpening slider is set to the value of 100

XeSS Upscaling through DP4a: better than FSR2.
XeSS essentially has the same performance gains as FSR 2.2 while producing better image quality at the same time using the DP4a instruction set, which is quite an impressive achievement

DLSS Upscaling: the best.
the DLSS Super Resolution implementation offers the best image quality across all resolutions and quality modes when upscaling is enabled. With DLSS in "Quality" mode you can expect slightly improved rendering of the details on tree leaves and vegetation in general, a sharp overall image with perfect stability in motion and on small particle effects, and the absence of any form of ghosting or shimmering artifacts
the DLSS image has the sharpest look, while FSR 2.2 is completely the opposite with a very soft overall image, and XeSS is in the middle between DLSS and FSR 2.2 in terms of sharpness


FSR Native: moderate improvements, still the worst though
With FSR 2.2 running in "Native AA" you can expect a sharper image, but it doesn't help with the shimmering issues, which are noticeable even when standing still

XeSS Native: good
With XeSS running in "Native AA" mode you can expect an improved quality of built-in anti-aliasing, which means less visible pixelation and jaggies in vegetation in particular, but with a slight cost to performance compared to the TAA solution

DLSS Native (DLAA): looks the absolute best
With DLAA enabled, the overall image quality improvement goes even higher, rendering additional details and offering the best graphical experience overall compared to the TAA solution, FSR, DLSS or XeSS

 
Worth noting according to Oliver here, FSR 2 is turning in a better image presentation than even native with TAA in Warhammer:

 
What's up with the afmf vs fsr3 frame gen in that video. the forspoken part has afmf blurring the side of a building but the roof is sharp yet fsr3 fg has the building sharp but the roof blurred.
 
Back
Top