AMD FSR antialiasing discussion

Yea, if RTX/intel gpu owners are to use FSR, sure. I was under the impression both had their own solutions. Maybe they can be used in tandem somehow.

 

TAA is pretty much spiritual successor to its idea of temporal jittering of sample locations.

With need for motion vectors and it being an AA method, it isn't really a FSR anymore.

Should be great as new easy to implement TAAU option though.

 

So much for spartial upscaling, huh?
Could have introduced last year instead of holding the whole market back with FSR.

 

Better than native in the sense that it's the same image quality at a higher framerate or what?

 

In the same sense as a temporally supersampled image with antialiasing is better than the same image without AA.

 

I want some of whatever you're smoking that makes you think having an easy to implement, optional scaling method available for those who want to use it is somehow "holding the whole market back"

 

I'd imagine they mean if AMD has another technique that offers better than native or as good as native but with performance increases of running at a lower resolution that works on all the same hardware that fsr currently works on , then they would have been holding things back with fSA ?

 

We can't know whether releasing FSR 1.0 affected FSR 2.0 schedule even a single bit, and even if it did, one can't say it would "hold back the market" in any way

 

It's not about working on the same hardware, it's about the engines being able to implement FSR vs FSR2. With temporal reconstruction you need motion vectors which increases the complexity of implementation a lot, and discounts certain games and engines from being able to implement at all without the developer having to go back and re-design part of the engine.

So FSR had a very low barrier of entry, proven by the number of implementations in a short time.

 

I certainly don’t believe FSR held back the industry. I do consider it pretty close to worthless though.

 

FSR on its own didn't held back anything but AMD themselves did because of a bad block on using anything but FSR in AMD sponsored games. In this sense AMD's policy on implementing FSR certainly held back some games.

 

Just because it is free doesnt make it good.
An example from the indie game "Enlisted" which has DLSS, FSR and temporal upscaling:
FSR UQ vs. TU@80%: Imgsli
FSR P vs TU@25%: Imgsli

AMD is doing one 180° turn with FSR 2.0. That should say you something about the usefulness of spartial upscaling. Temporal upscaling would have benefit the pc market.

 

One can always cherry pick comparisons to suit whatever view they want to present.
FSR is very useful for many, integratable to pretty much anything with ease, provides good quality all things considered and is free of many issues of temporal methods.
Temporal methods have artifacting issues and are a lot harder to integrate into game engine and they need the engine to support specific features. When they work fine the quality is better, but spatial methods have their place on the market too and will continue to do so after FSR 2.0 comes out.

Did NVIDIA do a 180° turn when they released NIS while they already had DLSS? Didn't think so. So why is AMD making a 180° turn here?

And again, none of what you said is "holding the market back"

(personally I still steer clear of any scaling techs)

 

Spatial upscale is a dead end probably. Curious to see how FSR 2 compares to UE’s temporal upscale.

 

Or FSR2 is a accelerated solution akin to DLSS on rtx gpu's, launching with rdna3+ products. AMD couldn't have been sitting still and let the competition fly freely, same for ray tracing.

 

https://videocardz.com/newz/amd-fsr...ally-launches-q2-2022-rsr-launches-march-17th

 

It's only a dead end if all games can provide motion vectors to be used for a high quality temporal upscaling method. Not all games can provide motion vectors, so it's not a dead end.

Regards,
SB

 

Games without motion vectors are a dead end. FSR is great for old games, but any cutting edge engines for forward thinking titles are going to have to think about temporal upscaling. Unless there's some unforeseen breakthrough in ai upscaling that can "imagine" the missing details.

 

What you say may be true (quite likely) for AAA games. However, games without any support for temporal upscaling will exist for many years to come because not all game developers make AAA games with AAA budgets with the know how to do good temporal upscaling. Bad temporal upscaling is far worse than not having any temporal upscaling.

Regards,
SB

 

I totally disagree on small studios. In fact, these companies use ready-made game engines like UE and Unity which already integrate TAA and DLSS in their core. TAA/DLSS is mostly a check box and doesn't require any extra development with these engines. On the other side, big studios doing AAA games generally use in house engines that need IHV effort and help to integrate their proprietary solution.