AMD FSR antialiasing discussion

  • Thread starter Deleted member 90741
  • Start date
Inferior = Slower and worse quality.
FPS difference is neglible, but does exist. As for quality, most reviews seem to agree that both have their turns to shine over the other and are more or less on same level.

Also FSR is clearly superior in the fact it doesn't require specialized hardware, which based on performance has little to no benefits in this case, to run.

And it took nVidia so long because they have done the whole work alone and from the start. Where was AMD 2 1/2 years ago?
It's quite often one company does something and other(s) follow(s) later, temporal scalers aren't the first and won't be the last where this happens for GPUs. Sometimes it's AMD first, sometimes Intel, sometimes NVIDIA.
 
Aside from quick plugin integration, which is no doubt coming to FSR 2.0 too just like it came to DLSS, how exactly is it supposed to be inferior this time? The fact it doesn't say "NVIDIA" on it doesn't count as a reason.

IIRC it took NVIDIA about two years from DLSS launch and a year from 2.x launch to get the plugin ready.

Without the plugin option (aka use something other than recent UE or Unity) it's not job done in minutes.




Seems it employs a layer of soft vaseline over the entire image which smoothes over details compared to DLSS. And looking at Epic's similar solution in Ghostwire, elements like hair on characters and particles will have the tipical artifacts, which DLSS doesnt have.

We're gonna of course need a sample of more than one game to see how this goes.
 
TSR and FSR 2.0 are both good upscaling processes and in the same category. Can't wait for TSR/FSR 2.0 comparison reviews.
If review comparisons indicate TSR is superior (or even similar result) is there any reason for developers to use FSR 2.0? Software bloat?
 
FPS difference is neglible, but does exist. As for quality, most reviews seem to agree that both have their turns to shine over the other and are more or less on same level.
Also FSR is clearly superior in the fact it doesn't require specialized hardware, which based on performance has little to no benefits in this case, to run.

Do not care what reviewers say. Here is a quality comparision between DLSS und FSR 2.0 without FSR sharpener in motion: Imgsli
DLSS on my 3090 is 10% faster, delivers better quality and uses slighty less power.

It's quite often one company does something and other(s) follow(s) later, temporal scalers aren't the first and won't be the last where this happens for GPUs. Sometimes it's AMD first, sometimes Intel, sometimes NVIDIA.

AMD copied what nVidia has done. Still waiting for the answer how that makes GPU cheaper.
 
So much bean counting here.

Even if you are a fan of Nvidia products you HAVE to admit what an impressive feat AMD accomplished here with FSR 2.0. It comes super close to DLSS and doing all that while running on a very wide range of hardware without the need for special hardware. Nvidia heavily overstated the importance of machine learning in upscaling.

Sadly I was am not able to try it out myself as I do not own the game, but from what I've seen so far in videos and image comparisons, I am very impressed. Kudos to AMD!
 
Do not care what reviewers say. Here is a quality comparision between DLSS und FSR 2.0 without FSR sharpener in motion: Imgsli
Could you test some spots with geometry occluded by smoke plumes in the background or some cases like the vent here if there are any?
It would be nice to see any screens with tiny particles, with lots of moving foliage in the background, with thin hair or with scrolled textures, these are usually the hardest to handle for different reasons
 
It’s always better to have more options, and even while DLSS is currently a better option, FSR may push NVIDIA to make it better in order to stay ahead of the competition, so it’s a good thing anyway.


Sent from my iPhone using Tapatalk
 
I'm not sure why people are surprised by initial FSR2 showing. Epic's TSR is basically the same thing and it was already compared in Ghostwire with DLSS with rather similar results - meaning that a modern TAAU it's pretty close to DLSS on static images and is a bit worse in motion. Same was bound to be true for FSR2 so why the weird reaction?

DLSS will still provide better quality and will still be able to do so with lower base resolutions - Performance preset is 100% usable with DLSS for example.

Otherwise FSR2 is what FSR1 should have been - a great state of the art TAAU solution available to everyone now, actually capable of competing against other options.
I don't remember reading on this forum even once that dlss2 is very similar to some taau soluton available on consoles, rather opposite, thats is in their own league. Now all of sudden everybody expected fsr2 will be close to dlss2 ? ;)
 
Could you test some spots with geometry occluded by smoke plumes in the background or some cases like the vent here if there are any?
It would be nice to see any screens with tiny particles, with lots of moving foliage in the background, with thin hair or with scrolled textures, these are usually the hardest to handle for different reasons

PCGH has covered you: https://www.youtube.com/user/PCGHDigitalausgabe/videos

Maybe support them with a click to their website: https://www.pcgameshardware.de/Fide...e-277617/Specials/FSR20-Test-vs-dlss-1394934/
 
Do not care what reviewers say. Here is a quality comparision between DLSS und FSR 2.0 without FSR sharpener in motion: Imgsli
DLSS on my 3090 is 10% faster, delivers better quality and uses slighty less power.
If you like sharper maybe you should try with the sharpening, it's included in FSR 2.0 and user controllable for a reason.

AMD copied what nVidia has done. Still waiting for the answer how that makes GPU cheaper.
I haven't made such claims
 
Nvidia heavily overstated the importance of machine learning in upscaling.

For the end user it might not matter as much, but the AI/ML route is going to be easier than hand-coded solutions for developers. With everything going the ML/AI model these days, i'd say its the more advanced and future proof solution. I suspect AMD updating FSR to the deploy ML features in the future too.

I don't remember reading on this forum even once that dlss2 is very similar to some taau soluton available on consoles, rather opposite, thats is in their own league. Now all of sudden everybody expected fsr2 will be close to dlss2 ? ;)

FSR2.0 is just what TAAU/UE5 and consoles have been having for a while now. Its coming to pc and thats great.
 
Comparison 1:
FSR 2.0 has problems reconstructing the filigree elements less than one pixel wide and is also the only method that works out the snow unnaturally.
In a direct comparison, DLSS is in a completely different league, with many details and a high level of image stability pleasing to the eye.
However, the often encountered side effect of this "magic" should not go unmentioned: Selective streak/tail artefacts, for example with snow.

Comparison 2:
FSR 2.0 Quality calms the picture, works out details better and thus achieves a pleasing moving impression.
But here, too, Nvidia's DLSS is in the lead - note, among other things, the reconstruction of the radio mast on the right in the picture.

Motion:

FSR 2.0 undoubtedly looks good, considering that pixels are saved here internally,
but the process produces a little more jitter in motion than DLSS. In the following video you can see this on the sheet metal lying on the floor.

Conclusion:

According to our random samples, FSR 2.0 does not come close to DLSS 2.x overall, the Nvidia method scores in a direct comparison with both a higher image steadiness and detailed reproduction
However, DLSS still suffers from selective smear artifacts with fine, high-contrast elements, which FSR 2.0 does not have in this form

I await Digital Foundry's extensive analysis on this as well.

Nvidia heavily overstated the importance of machine learning in upscaling.
The importance of AI is still present, DLSS performance mode is still significantly better than FSR 2, This is an important mode for getting your fps up in heavy games.
 
Company fanboys are once again completely ruining this forum. Hard to discuss anything without dick waving.

Well that techpowerup article wasn't more than a rant to NV but ok. For me personally i dont give a fuck if its AMD or NV, my platform accepts both. I dont have to choose sides since both will slide into my gaming systems. FSR2.0 entering and giving us pc gamers console-level reconstruction for all GPU's is a very, very good thing to happen.

Edit: i have a 2080Ti in my main gaming system, 3070m laptop and another gaming system with a 6800XT.
 
Last edited:
Company fanboys are once again completely ruining this forum. Hard to discuss anything without dick waving.

It’s unavoidable when all of the upscaling tech being discussed is company sponsored. UE’s TAAU is the only entry untainted by IHV marketing. But yes it would be nice to have more discussion of actual technology and less sniping.
 


They're testing older gpu's with it
Primarily the biggest market for FSR. Older weaker hardware.
We still use a variety of non deep learning methods in data science even though our industry is dominated by deep learning as buzz words. Appropriate technology for the right solution is still better than a nuke for every problem.
 
Found a rotating object in my level: Imgsli
It looks fine to me)

In case of weather effects, handcrafted algorithm fails because it was crafted for a common case.
Snowfall or raindrops confuse handcrafted weighting logic due to rapid color change (so that a common weighting logic is no longer optimal).
NN on the other hand can be easily trained to account for such cases)

Though, the best solution would be to render particles in a separate pass, but that's not always possible.
 
Last edited:
Back
Top