• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

AMD Announces FidelityFX Super Resolution 3 (FSR 3) Fluid Motion Rivaling DLSS 3, Broad Hardware Support

Status
Not open for further replies.
FSR3 being able to run on all relatively recent GPUs makes it far more likely that DLSS3 frame generation was locked to Ada for non-technical reasons.
Nvidia has a paper on it actually comparing it against all their cards with tensor. Has FPS and latency data. They all have fps improvements with frame gen, but latency is worse on all previous generations to Ada, but it doesn't go into detail on how impactful the difference is and if their anti latency tech helps. The 3000 series also only has a very minor incr ase latency impact over Ada.
 
If implementing a technology in your game takes time, then why would you spend extra time to implement two very similar technologies when you could do just one that fits all, and call it a day? Maybe there's no conspiracy here, just capitalism, and limited game development times.

Yes, Nvidia agrees, developed and released the open source Streamline framework which serves this exact purpose. One implementation to service all three upscalers.

Intel adopted it and Nvidia provides DLSS precompiled for it, but AMD has refused to write and maintain an FSR plug-in for it.
 
Yes, Nvidia agrees, developed and released the open source Streamline framework which serves this exact purpose. One implementation to service all three upscalers.

Intel adopted it and Nvidia provides DLSS precompiled for it, but AMD has refused to write and maintain an FSR plug-in for it.
That doesn't really answer my question, though. If you were a developer, and had to choose which tech to use because of the limited time available, would you use the one that all gamers can enjoy, or the one that only X% can?

Nvidia has a paper on it actually comparing it against all their cards with tensor. Has FPS and latency data. They all have fps improvements with frame gen, but latency is worse on all previous generations to Ada, but it doesn't go into detail on how impactful the difference is and if their anti latency tech helps. The 3000 series also only has a very minor incr ase latency impact over Ada.
Well, they didn't refrain from enabling RT on certain GTX 1000 and 1600 cards, so I don't see why they should not enable FG on Turing onwards.
 
That doesn't really answer my question, though. If you were a developer, and had to choose which tech to use because of the limited time available, would you use the one that all gamers can enjoy, or the one that only X% can?

Streamline implementation takes just the time of a single implementation and would support all technologies, it'd be a no brainer if AMD had decided to participate.

But in the shoes of a game developer targeting specifically the Windows PC platform, I'd probably implement DLSS and test games primarily on Nvidia hardware simply due to their market share (88% of the discrete graphics market share in Q3/22) and very high marketability of the tech (trendy name, influencers love it, etc.)
 
If implementing a technology in your game takes time, then why would you spend extra time to implement two very similar technologies when you could do just one that fits all, and call it a day? Maybe there's no conspiracy here, just capitalism, and limited game development times.

again my only annoyance is if you are only going to include one version at least make it good and not the shoddy implementation we typically get when it comes to FSR2.
 
It uses Async Compute to work.



So far it sounds promising but there is some confusion about the driver base version only maybe working on 7000 series and the newer latency reducing antilag+ only working on 7000 series.
Thanks @oxrufiioxo. That's a good read and rather than pitching camps and advocating for multi-billion dollar companies, we should actually discuss this article. Regular FSR3 uses a combination of motion vector input from FSR 2 and optical flow analysis as the basis for interpolating frames. The actual computation relies on asynchronous compute to reduce the burden. The driver based solution, on the other hand, only uses optical flow and that makes it more limited.
 
Streamline implementation takes just the time of a single implementation and would support all technologies, it'd be a no brainer if AMD had decided to participate.

But in the shoes of a game developer targeting specifically the Windows PC platform, I'd probably implement DLSS and test games primarily on Nvidia hardware simply due to their market share (88% of the discrete graphics market share in Q3/22) and very high marketability of the tech (trendy name, influencers love it, etc.)
88% is still not 100%. FSR is supported by 100% of current gen GPUs. For me that's enough reason to choose it over DLSS without a blink of an eye. Proprietary technologies (from any company) should die a painful death, imo.

again my only annoyance is if you are only going to include one version at least make it good and not the shoddy implementation we typically get when it comes to FSR2.
That I can agree with.
 
88% is still not 100%. FSR is supported by 100% of current gen GPUs. For me that's enough reason to choose it over DLSS without a blink of an eye. Proprietary technologies (from any company) should die a painful death, imo.


That I can agree with.

I'm all for things being open like with adaptive sync basically replacing Gsync in most monitors but only when the open version is as good or better than what a proprietary version offers. IF FSR could be substantially better on just RNDA 2/3 hardware for example I would prefer that vs a lesser more open version.

Especially when it comes to upscaling because currently if you can implement one they are all easy to implement. IT actually should be embarrassing for the developers at this point who only support FSR because modders mod in a better quality DLSS.

I actually like intel's approach better of having a version that works better on their hardware but a still lesser more open version.
 
Low quality post by R0H1T
If AMD's driver level implementation of frame interpolation is decent this will make Nvidia look really bad
Unfortunately it's highly unlikely to do the job properly, otherwise, why would they demo it with irrelevant trash like Forspoken instead of Starfield?
 
The biggest Takeaway for me was Frame Generation in all Games.
Problem is that it's driver level and part of hypr-rx and therefor will be locked to rdna3 / 7000 series only... FSR3 will need to be directly integrated into the game for it to compatible with "all gpu's"

Driver level frame generation i guess AMD wins assuming NVIDIA isn't working on something similar already but then again NVIDIA only supports RTX 40 series with DLSS3 so i guess AMD wins this round without even trying.
Except that that's going to be a part of the new hypr-rx driver features and that tech is locked to rdna3 / 7000 series only!

Thanks @oxrufiioxo. That's a good read and rather than pitching camps and advocating for multi-billion dollar companies, we should actually discuss this article. Regular FSR3 uses a combination of motion vector input from FSR 2 and optical flow analysis as the basis for interpolating frames. The actual computation relies on asynchronous compute to reduce the burden. The driver based solution, on the other hand, only uses optical flow and that makes it more limited.
And the driver level version will only run on 7000 series / rdna3 gpu's... It's part of hypr-rx and that's locked to those cards only.
 
Problem is that it's driver level and part of hypr-rx and therefor will be locked to rdna3 / 7000 series only... FSR3 will need to be directly integrated into the game for it to compatible with "all gpu's"

Except that that's going to be a part of the new hypr-rx driver features and that tech is locked to rdna3 / 7000 series only!

And the driver level version will only run on 7000 series / rdna3 gpu's... It's part of hypr-rx and that's locked to those cards only.
So? According to AMD themselves, the AFMF solution will be a shitty variant of FSR3's frame generation. Why are you crying about it not being supported by everything when FSR3 will be and it's the superior system?
 
Another proof that you don't need fancy Tensor Cores or whatever to run DLSS 3, just as scumbag nGreedia callously told us.
Once a greedy scumbag, always a scumbag.
Good job AMD.
 
Can't wait to see this in action and make up my own mind. A shame AMD are locking the driver level one to RX 7000 only (or higher, presumably).

As for the soup of mental gymnastics, flawed logic and inability to accept evidence and draw reasonable conclusions in the last 6 pages, I suppose it's become par for the course on forums, depending on the shade of your glasses.
 
Can't wait to see this in action and make up my own mind. A shame AMD are locking the driver level one to RX 7000 only (or higher, presumably).

As for the soup of mental gymnastics, flawed logic and inability to accept evidence and draw reasonable conclusions in the last 6 pages, I suppose it's become par for the course on forums, depending on the shade of your glasses.
Looks like only Anti-Lag+ is limited to RX7000.

 
Problem is that it's driver level and part of hypr-rx and therefor will be locked to rdna3 / 7000 series only... FSR3 will need to be directly integrated into the game for it to compatible with "all gpu's"


Except that that's going to be a part of the new hypr-rx driver features and that tech is locked to rdna3 / 7000 series only!


And the driver level version will only run on 7000 series / rdna3 gpu's... It's part of hypr-rx and that's locked to those cards only.
Who wants driver level frame generation in older games that run at hundreds of FPS on any RDNA 2 GPU anyway? Newer games that actually benefit from it will support it on game level, so no issues there.

This is a typical example of making a problem out of a non-problem just for the sake of it.
 
I don't like frame gen. in general. No matter if the frames are created by hardware like DLSS 3 frame gen. or by software like FSR3.
The thing is that the cards like 6800/XT and better can use it without many negatives of the tech (latency) while it's needed on the cards that are worse than the aforementioned.
And that's the problem. The lower end cards do not have the power to achieve normal base fps, so that the frame gen tech can work from.
Anyway, the image reconstruction is far more beneficial than this.

Practically as I see it, it will be useful mainly on consoles while on pc space, the cards that can use it properly, do not even need it.
Yes, most likely it will be worse than DLSS 3 frame gen. but it doesn't matter.
A credit has to be given to AMD anyway that it can be used on any card (with async compute) no matter the results.
 
Well believe it or not this tech will make it to consoles & portable consoles & then you'll thank AMD, or Nvidia, for it so there's that :pimp:

Besides the hyperbole surrounding pros & cons of each implementation this is a godsend for smaller/mobile devices!

And yes it's probably also coming to an Exynos near you.
 
Sadly there are much better, more mature discussions being had about this elsewhere. The camping is killing the vibe here.

Later.
 
Unfortunately it's highly unlikely to do the job properly, otherwise, why would they demo it with irrelevant trash like Forspoken instead of Starfield?
What should they demo it with, enlighten us. Also why would it even matter what game it is ? You do understand how this works, right ? It just inserts a new frame in-between 2 consecutive ones, it works the same everywhere.
 
To everyone going on about Streamline above.

Streamline doesn't support anything outside the PC ecosystem, with AMD supporting three different consoles (including the Nvidia powered Switch), Streamline would split FSR implementation for little benefit to itself. If Nvidia truly wants Streamline to be the default implementation approach, it'd bring support for GNM, GNMX and NVN to Streamline (even if its upscaling solutions themselves don't get ported).

As for Intel supporting Streamline, of course it would, XeSS is limited to PC, and as a low marketshare competitor, can use the implementation to push its own technology.
 
What should they demo it with, enlighten us. Also why would it even matter what game it is ? You do understand how this works, right ? It just inserts a new frame in-between 2 consecutive ones, it works the same everywhere.
They should demo it with the biggest game release in recent years, and officially AMD sponsored title, but they don't even mention Starfield as FSR3 supporting title. Makes no fucking sense. No on gives a shit about Forspoken.
 
What should they demo it with, enlighten us. Also why would it even matter what game it is ? You do understand how this works, right ? It just inserts a new frame in-between 2 consecutive ones, it works the same everywhere.
Exactly this!

In order for FG to work (whether you call it DLSS 3 FG or FSR 3, it doesn't matter), you need 2 queued frames in the render pipeline to insert a 3rd one in between. If you want a decent, lag-free gaming experience, you should limit queued frames to 1. Conclusion: frame generation is pointless. Moving on. :)
 
Status
Not open for further replies.
Back
Top