• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Fidelity Super Resolution 2.0

Caporegime
Joined
12 Jul 2007
Posts
37,603
Location
United Kingdom
Dodgy motion was at 4k quality as well. You can over sharpen in the Nvidia CP. Might just need better optimization from dev.

8p7LV8z.png
As with the ghosting in motion noticed by HUB when aiming down the gun sights whilst using DLSS, neither solution is perfect tbh. Native is usually always king, unless the TAA solution is as bad as we see in Deathloop.
5tcJWSW.jpg
 
Caporegime
Joined
4 Jun 2009
Posts
26,867
TAA is just bad in all games ;)

Looking at the characteristics of where FSR 2 falls short and going by digital foundrys videos where they compared TSR and DLSS in ghostwire, FSR 2 is looking like it is very much based on TSR. Any info/idea on that Matt?
 
Caporegime
Joined
12 Jul 2007
Posts
37,603
Location
United Kingdom
TAA is just bad in all games ;)

Looking at the characteristics of where FSR 2 falls short and going by digital foundrys videos where they compared TSR and DLSS in ghostwire, FSR 2 is looking like it is very much based on TSR. Any info/idea on that Matt?
Some worse than others I think. As with the recent examples provided though in the last few posts, they each introduce their own issues. It really is a pick your poison isn't it.

No idea on if FSR is based on TSR. Given that FSR 1.0 was not based on lanzcos, I doubt it. Don't know for sure though sorry so just a guess on my part.
 
Soldato
Joined
6 Feb 2019
Posts
10,785
Dolmen launches on May 20 and supports DLSS, XeSS and FSR at launch making it the first game to support all three technologies

 
Associate
Joined
20 Nov 2020
Posts
1,027
Dolmen launches on May 20 and supports DLSS, XeSS and FSR at launch making it the first game to support all three technologies

Can't wait to see people saying that XeSS needs one year to implement since that much has passed from XeSS announcement until first release. :)
The reality is all three solutions need the same time. And it is not minutes or hours even when you have an engine plugin ( it will take at least a couple of days to add it into the game, test it and tweak it ). Think before writing something this stupid, even if you have a plugin you still need to at least add the option to your game menu, you can't do that in minutes. :)
On the other hand i see the Nvidia bodyguard said FSR works better on Ampere. But if we check other websites, for example this obscure one, we see this:
6800XT goes from 33.8 FPS in native to 55.5 FSR2Q to 74 FPS FSR2P
3080 goes from 46.3 FPS in native to 68.4 FSR2Q to 90.4 FSR2P

So we see the perf increase in Big Navi is more than 60% for FSR2Q and more than 100% for FSR2P
On Ampere the perf increase is less than 50% on FSR2Q and less than 100% on FSR2P.
So how can it work better on Ampere? :)
 
Caporegime
Joined
4 Jun 2009
Posts
26,867
Some worse than others I think. As with the recent examples provided though in the last few posts, they each introduce their own issues. It really is a pick your poison isn't it.

No idea on if FSR is based on TSR. Given that FSR 1.0 was not based on lanzcos, I doubt it. Don't know for sure though sorry so just a guess on my part.

I haven't looked since so maybe it has been "debunked" but was there not some lanzcos setting/code found in the source code for FSR 1? Obviously it wasn't just this though.

Can't wait to see people saying that XeSS needs one year to implement since that much has passed from XeSS announcement until first release. :)
The reality is all three solutions need the same time. And it is not minutes or hours even when you have an engine plugin ( it will take at least a couple of days to add it into the game, test it and tweak it ). Think before writing something this stupid, even if you have a plugin you still need to at least add the option to your game menu, you can't do that in minutes. :)
On the other hand i see the Nvidia bodyguard said FSR works better on Ampere. But if we check other websites, for example this obscure one, we see this:
6800XT goes from 33.8 FPS in native to 55.5 FSR2Q to 74 FPS FSR2P
3080 goes from 46.3 FPS in native to 68.4 FSR2Q to 90.4 FSR2P

So we see the perf increase in Big Navi is more than 60% for FSR2Q and more than 100% for FSR2P
On Ampere the perf increase is less than 50% on FSR2Q and less than 100% on FSR2P.
So how can it work better on Ampere? :)

Well yeah if you're adding in "everything" like that then yes of course there is more to it than just simply installing and enabling the plugin.....

From the developer for mario ray tracing mode:


OLXOFR9.png


And further info:

Person who posted the tweet here.

I'd say both FSR and DLSS are about on the same level of difficulty, but it depends on what information the engine has. FSR only requires the lower resolution image, while DLSS needs that, a subpixel jittered output and the screen space motion vector information.

I imagine the last one will be the toughest one depending on the game/engine, but anything that implements TAA should fullfill these requirements already.

NVIDIA also provides some excellent debugging tools to make the integration much easier due to this additional requirement.

I'm by no means an expert on the technology and I haven't even finished reading most of the details in the integration guide yet, but I'd say a day of work is certainly enough to get decent results at least.

Certainly needs more days or even weeks to sort out the UI/UX details and any potential glitches that might pop up due to incorrect motion vectors or ghosting glitches on sudden scene transitions.

And from the developer of fabled woods:

Thank you! With the new plugin available through Unreal Marketplace, it's as simple as downloading and enabling =D I ran into 0 issues with implementation.

Nvidia stated in their announcement/press release for fabled woods:

Nvidia proudly notes that the Fabled Woods developers were able to implement the UE DLSS plugin in less than a day.

There are also videos on youtube showing the process.

So I would like to think proper/bigger game development companies with a full team of developers (who will have had far more experience) could squeeze in "adding DLSS (and probably FSR 2)" feature(s) without assigning much effort to these PBIs as part of their sprint planning or whatever work methodology game developers use/follow.

Your link doesn't work for the fsr 2 perf. thing. Also, it isn't just DF (presuming that is who you are referring based on the "nvidia bodyguard" comment?) saying that performance is better on ampere, HUB also are showing the same.....
 
Last edited:
Associate
Joined
20 Nov 2020
Posts
1,027
So I would like to think proper/bigger game development companies with a full team of developers can squeeze in "adding DLSS (and probably FSR 2)" feature(s) without assigning much effort to these PBIs as part of their sprint planning or whatever work methodology game developers use/follow.

Your link doesn't work for the fsr 2 perf. thing. Also, it isn't just DF (presuming that is who you are referring based on the "nvidia bodyguard" comment?) saying that performance is better on ampere, HUB also are showing the same.....
I think it is a lot more work than some of us think and tbh i don't see FSR2 being added in a lot of games unless AMD pays a lot of money, the way Nvidia did/does. There are two ways to do it, you either pay for their time to add your feature in their games or you build hype and put pressure on the developers like :" can't believe FC6 doesn't have the state of the art temporal upscaler" :)
Nvidia has the advantage of the number of DLSS ready cards already sold so it is much easier to put pressure on the devs. And they have a much bigger wallet.
This is the link ( it was translated first, that's why it wasn't working, and most likely those are the numbers. Anyone can do the maths and see the big navi has bigger gains from FSR2:
aaaaaa-jpg.png
 
Caporegime
Joined
4 Jun 2009
Posts
26,867
You literally have 2 people who have implemented it into their games in black and white saying how quick and easy it is.... I can't recall exactly but I think Dario wasn't even using the plugin way either.

Videos showing the exact process for the unreal engine plugin:



For FSR 2, amds own diagram (although we'll see just how this holds up in the next couple of months...):

sZP0ito.png


Unless you have read/seen something to suggest otherwise? If so, link please.

If DLSS wasn't being added to non-nvidia sponsored games, I would believe that but thankfully, dlss is being added to various games, which nvidia have no oversight to. Same as FSR. Developers are lazy, they will want to get away with the least work/effort required, if that means adding dlss/fsr to avoid having to spend time optimising their game, they will. There is no reason for developers to not use tools that are available to them for free with the click of a button. That's like saying to me with my job, I can't use any marketplace extensions/plugins for azure devops or any plugins for jenkins because it makes my job too easy then.
 
Man of Honour
Joined
13 Oct 2006
Posts
83,090
So I would like to think proper/bigger game development companies with a full team of developers (who will have had far more experience) could squeeze in "adding DLSS (and probably FSR 2)" feature(s) without assigning much effort to these PBIs as part of their sprint planning or whatever work methodology game developers use/follow.

The people who worked on the FSR implementation in Quake 2 RTX and the talk surrounding DLSS (which they can't implement due to the open source complications noted in the Tweets in your post) said it is basically a days work to implement especially if you have a working TAA implementation already - but can take longer to test fully for both.

(Those who have coding experience can see the extent of coding required for FSR support in Quake 2 RTX here: https://github.com/NVIDIA/Q2RTX/tree/master/src/refresh/vkpt )

As an aside the implementation in Quake 2 is basically FSR 2.0 though not quite - as they already combined it with a temporal upscaling implementation - one advantage in Quake 2 is you can easily fine tune the resolution scale and sharpening so as to get the FPS you desire with minimal quality loss. It does have a slightly undesirable effect that if you stand still the quality of the image increases a bit which is noticeable.
 
Last edited:
Associate
Joined
20 Nov 2020
Posts
1,027
You literally have 2 people who have implemented it into their games in black and white saying how quick and easy it is.... I can't recall exactly but I think Dario wasn't even using the plugin way either.

Videos showing the exact process for the unreal engine plugin:


Unless you have read/seen something to suggest otherwise? If so, link please.

If DLSS wasn't being added to non-nvidia sponsored games, I would believe that but thankfully, dlss is being added to various games, which nvidia have no oversight to. Same as FSR. Developers are lazy, they will want to get away with the least work/effort required, if that means adding dlss/fsr to avoid having to spend time optimising their game, they will. There is no reason for developers to not use tools that are available to them for free with the click of a button. That's like saying to me with my job, I can't use any marketplace extensions/plugins for azure devops or any plugins for jenkins because it makes my job too easy then.
You don't read your own posts. Because that guy says " it will take few days to ...". You don't just activate a plugin or add the motion vectors or whatever. All these things need to be tested and tweaked, the features need to be added to the game UI and so on. That takes time. It is not a minute process and it is not a five hours process. It will take at least several days if you want to make sure the feature works ok.
 
Caporegime
Joined
4 Jun 2009
Posts
26,867
You don't read your own posts. Because that guy says " it will take few days to ...". You don't just activate a plugin or add the motion vectors or whatever. All these things need to be tested and tweaked, the features need to be added to the game UI and so on. That takes time. It is not a minute process and it is not a five hours process. It will take at least several days if you want to make sure the feature works ok.

Again, re-read the posts, watch the 2 videos above and look at amds own diagram for timescale....

Pretty much every game has TAA implemented now so that's not even really a factor to take into consideration.

I'm by no means an expert on the technology and I haven't even finished reading most of the details in the integration guide yet, but I'd say a day of work is certainly enough to get decent results at least.

Here's another tweet explaining his POV:

5UUqJz0.png


And that's from someone using it for the "first time". For unreal engine, it appears to be the easiest and quickest to add dlss. For a more experienced developer or/and someone who has done this before, I can guarantee you it will be an easier and quicker process for them or in the case of Dario now that he has already done this, future implementations will be even quicker/easier for him.

Like I said, of course if you're talking about "end to end" i.e. right from implementation to "thoroughly" testing it then yes it will take longer, I'm referring purely to just adding DLSS here and more so to engines that have it added as a plugin. It's not as much of a hardship to add like what some people think, especially when compared to FSR 2 from the looks of it. Do you really think adding the UI selector for dlss is going to take a seasoned developer "days" to do... If that kind of task took any of the developers I work with days to do, they would be getting a serious talking to or sacked.
 
Last edited:
Soldato
Joined
6 Feb 2019
Posts
10,785
Dodgy motion was at 4k quality as well. You can over sharpen in the Nvidia CP. Might just need better optimization from dev.

8p7LV8z.png


So neither solution is as good as native for motion clarity, but dlss is much closer to native though not quite there
 
Caporegime
Joined
12 Jul 2007
Posts
37,603
Location
United Kingdom
So neither solution is as good as native for motion clarity, but dlss is much closer to native though not quite there
At the very end of the video he showed another weird ghosting issue with DLSS too that looks very similar to that FSR example posted earlier. Makes me think it might be an issue unique to this game rather than tied to either tech.
 
Soldato
Joined
25 Sep 2009
Posts
8,394
Location
Billericay, UK
Hopefully the Lossless Scaling Tool gets updated so FRS 2 can be run on every game.

Edit

Drat, that's a bummer. According to the Lossless Scaling Tool developer FRS 2 can't be forcibly injected into the render pipeline as things like depth buffer, motion vector buffer, and jittered render samples now can't be obtained outside the game engine.
 
Last edited:
Caporegime
Joined
18 Oct 2002
Posts
26,464
Location
y0 Mommas a**
I think it is a lot more work than some of us think and tbh i don't see FSR2 being added in a lot of games unless AMD pays a lot of money, the way Nvidia did/does. There are two ways to do it, you either pay for their time to add your feature in their games or you build hype and put pressure on the developers like :" can't believe FC6 doesn't have the state of the art temporal upscaler" :)
Nvidia has the advantage of the number of DLSS ready cards already sold so it is much easier to put pressure on the devs. And they have a much bigger wallet.
This is the link ( it was translated first, that's why it wasn't working, and most likely those are the numbers. Anyone can do the maths and see the big navi has bigger gains from FSR2:
aaaaaa-jpg.png
Nvidia 3000-series is def a compute powerhouse, might be what's helping it here?
 
Associate
Joined
3 May 2021
Posts
462
Location
Italy
I still dont understand why AMD isnt tapping into DirectML
ML is not the solution to every problem, especially when you're looking for consistency. When you use fixed algorithms you know how they can fail and correct their flaws to a certain extent, with ML you end up with endless corner cases that you cannot easily fix.
 
Top Bottom