AMD FSR antialiasing discussion

  • Thread starter Deleted member 90741
  • Start date
pharma said:
You anticipate the inconsistency from game to game to disappear across different engines if FSR 2.0 is used?
Click to expand...
I think it should because the game should require a specific input pipeline to work. I'll know more once I run through the spectrum analyzer and determine that truth. But as of my current viewpoint of TAAU - very inconsistent behaviour engine to engine, even title to title, but as for title to title, it's likely they're doing some additional work on it. So I guess it's not fair to say that.
 
troyan said:
What is more universal than UE4 own TAAU?
Click to expand...
an algorithm that can apply to UE4 and outside of UE4?
edit: I applaud anything that is made and given out to everyone that is
  • hardware agnostic
  • engine agnostic

both XeSS and FSR2.0 are ideal in that respect. The XeSS has steeper requirements for inputs, but I suspect FSR2.0 will gain much faster traction. Though I'm going to set a low bar here, these algorithms needs to compare themselves against a standard bicubic upscale something that is universal. Comparing against custom TAAU solutions will happen, which is fine, but if a generic algorithm can hit solid across all engines and hardware, that will bring the overall graphical quality up moving bicubic out the door.
 
Last edited:
I guess FSR 2.0 may become a standard used in proprietary engines which don't have their own TAAU solution yet. But beyond that? Most other engines (UE, Unity) have TAAU already and it's unlikely that FSR2 will be any better.

Then again we need to see how XeSS will fare on non-Intel h/w. If it will do fine most of the time then why would anyone use FSR2 over XeSS?
 
DegustatoR said:
I guess FSR 2.0 may become a standard used in proprietary engines which don't have their own TAAU solution yet. But beyond that? Most other engines (UE, Unity) have TAAU already and it's unlikely that FSR2 will be any better.

Then again we need to see how XeSS will fare on non-Intel h/w. If it will do fine most of the time then why would anyone use FSR2 over XeSS?
Click to expand...
May depend on whether you want to spend the labour to develop a solution further. I assume here AMD will continue to take this on and improve FSR over time. If companies like Intel, Nvidia, and AMD are willing to do this for you, the question becomes whether a company should continue to develop their own custom solution. There are pros and cons of course - time/labour/money being an important factor here if you could devote your resources elsewhere.

Epic and Unity are the exception of course, this is their core business. They are more engine development than anything else now and they have different use cases that extend beyond games. But for everyone else (like Ubisoft) who leverages a large list of different engines all with different TAAU results, will it continue to make sense.
 
DegustatoR said:
If it will do fine most of the time then why would anyone use FSR2 over XeSS?
Click to expand...
Because FSR2 can be faster and despite of the XeSS being open source, NN in the XeSS is still pretty much a black box unless all DL environment (training samples, etc) and training/inference code is open sourced as well.
Though, looking at TSR, I don't expect FSR2 to be integration friendly. Epic's TSR uses a whole bunch of auxiliary buffers and masks to make it work without the lossy neighborhood color clipping.
Unless AMD's devs are magicians, I don't expect FSR2 to be significantly better here (there are some shortcuts, such as triangle IDs buffers, but they are all expensive).
If FSR2 requires lots of auxiliary buffers in the same manner as TSR, it will be hard to implement into an engine and debug, which is probably why it was announced in just a single title and in Q2.
That's a speculation though, so as always lets just wait and see how it will pan out.
 
I still game at 1080 so there's not a lot of advantage in it for me. My rig is pretty much designed to be comfortable at 1080 and it pretty much is.

I don't think using FSR to upscale it in 4K using super-scaling to make it fit my 1080 screen is going to improve image quality much, and I doubt it'll help performance.

It's just not a thing I use, nothing against it and I hope they improve it so when I do need it I can use it. :)
 
digitalwanderer said:
I still game at 1080 so there's not a lot of advantage in it for me. My rig is pretty much designed to be comfortable at 1080 and it pretty much is.

I don't think using FSR to upscale it in 4K using super-scaling to make it fit my 1080 screen is going to improve image quality much, and I doubt it'll help performance.

It's just not a thing I use, nothing against it and I hope they improve it so when I do need it I can use it. :)
Click to expand...

This can be big for the steam deck , esp if they can enable it on any game like FSR. 540p to 720/800p if it looks good enough could be the solution to a lot of games performance and battery life issues on the steam deck. I can also see this being big on consoles proper. Series s doing 720/900 to 1080p could look good depending on the quality of FSR 2. Series X/ PS5 doing 1440p to 4k could also look good. Laptops could also take advantage of this.

Like I said , I'd always take 4k over this and dlss if I had the choice , I'd even upgrade hardware to get to 4k. But I rather have more options to play around with for when I can't hit 4k
 
I totally agree. I got nothing against the tech, it's just not useful for me right now.

I see huge benefits if they can make it look decent, I'm sort of curious to see the new improved 2.0 tbh just for the reasons you stated. Just because I don't use it doesn't mean it's not interesting to me or won't benefit someone else. :)
 
Malo said:
FSR2 requires access to frame color, depth, motion vectors and previous frames. I don't really see it being any easier to implement than DLSS. It's definitely not going to get supported like FSR.
Click to expand...

Wonder how it will compare to DLSS with hw acceleration. FSR2 seems a very viable alternative, and will probably give NV all the more reason to improve their solution and that in turn will make AMD do so aswell.
 
You must log in or register to reply here.

Similar threads

Cyan
Microsoft's Auto SR: Universal AI upscaling discussion.
Replies
7
Views
2K
Janne Kylliö
J
B
DLSS versus FSR, again, neither being PSSR *spawn
2 3
Replies
41
Views
4K
homerdog
homerdog
Unknown Soldier
3DMark AMD FSR Feature Test
Replies
1
Views
1K
Putas
P
BRiT
Intel XeSS anti-aliasing discussion
9 10 11
Replies
200
Views
36K
Shifty Geezer
Shifty Geezer
M
Proposed upscaling technique - render vectors to pixels, convert to vectors, upscale and re-raster *spawn
2 3
Replies
43
Views
4K
JoeJ
J
Back
Top