Fidelity Super Resolution in 2021

[ICDP]

I wasn't aware of DLSS introducing latency issues. Ghosting yes, but not input lag. Is there any proof to show this latency issue with DLSS?

 

[Nexus18]

Zero input lag issues here and this is on both my 4k 60hz oled and 144hz monitor so not sure why people keep bringing this up??? People getting confused with ghosting/motion and input lag? (same way people get confused between input lag and pixel response time). As posted, is there any proof to back up such claims?

I think maybe it was COD where there was an issue with it? But if so, given that every other games doesn't have the issue, it's more likely down to cod and a conflict or something.

 

[ICDP]

https://www.theverge.com/2021/6/24/22549594/microosft-xbox-amd-fsr-xbox-consoles-support

Microsoft releases FSR Devkit to XBox developers. This is very big in terms of games and developer support for FSR. With the MS cross platform gaming any Xbox games that have FSR will have it on PC.

 

[D.P.]

Also another factor that needs to be looked at in terms of performance gain is that whilst these aim to have the same outcome i.e. retaining close to native res. with a performance uplift.... they are still rather different in the sense that dlss is also essentially doing AA work (which is arguably superior than the majority of current AA methods/implementations) so if dlss was to be purely just upscaling and nothing else, I suspect the performance uplift from dlss would be even larger.

Performance also needs to to be compared at similar IQ, and this is a problem since FSR is not actually comparable since it is only really providing good edge quality (texture quality is just standard bilinear filter and sharpening).

But thete are lots of nuances. Since DLSS replaces TAA with a better version, FSR has to be compared with TAA enabled in game to handle temporal artifacts and pixel shimmering. Image quality is not a strict linear scale on a single KPI, FSR gives great edges and poor textures, FSR cannot handle fine details like a wire because it isn't accumulating samples. DLSS provides verv good textures, temporal stability, fine details, but there is ghosting if not properly tuned or issues with the motion vectors.






A baseline might be DLSS on balanced snd FSR on ultraquality.

 

[D.P.]

Lets wait for FSR to land in Unreal Engine, then i'll have an Orange to compare to another Orange.

Edit: DLSS is as easy as plug in and play, right? Nvidia???

FSR wont appear in Unreal Engine. UE4 has TAA-U and UE5 has TSR (both very similar). These are superior to FSR so Epic wont waste resources on FSR

 

[JediFragger]

.

 

[shankly1985]

Performance also needs to to be compared at similar IQ, and this is a problem since FSR is not actually comparable since it is only really providing good edge quality (texture quality is just standard bilinear filter and sharpening).

But thete are lots of nuances. Since DLSS replaces TAA with a better version, FSR has to be compared with TAA enabled in game to handle temporal artifacts and pixel shimmering. Image quality is not a strict linear scale on a single KPI, FSR gives great edges and poor textures, FSR cannot handle fine details like a wire because it isn't accumulating samples. DLSS provides verv good textures, temporal stability, fine details, but there is ghosting if not properly tuned or issues with the motion vectors.






A baseline might be DLSS on balanced snd FSR on ultraquality.

So you are saying FSR ultra Quality is comparable to balanced?

 

[shankly1985]

FSR wont appear in Unreal Engine. UE4 has TAA-U and UE5 has TSR (both very similar). These are superior to FSR so Epic wont waste resources on FSR

Superior based on what exactly? On paper yes but in practice by how much?

You was only saying other day FSR 1.0 will be worst or just like DLSS 1.0

But it turns out that it's actually so much better.

Again I will say this it doesn't matter how these techniques are done so long the end goal is to have you an image quality of the native resolution.

It doesn't matter how you get there.

Again I will use Freesync vs Gsync again both achieve the same goal both very different approaches.

 

[JediFragger]

Tbf Ultra Quality FSR would be my goto, hardly any IQ hit for a fairly large fps increase! I don't think I'd personally consider the other options, but 99% of the time I wouldn't need to.

 

[D.P.]

So you are saying FSR ultra Quality is comparable to balanced?

no, i am saying they don't do the same thing do quality is hard to compare. If you only looked st edges,then probably FSR ultra is going to be similar to balanced in DLSS. If you look at texture quality or fine details such as wires, there is likely no FSR setting that is sufficiently good, but perhaps if yéu lowered DLSS to performance it would be possible to compare. If you look at ghosting then by design FSR doesn't have temporal accumulation, but the ghosting will be inherited and magnified by the underlying TAA, if you care sbout pixel shimmering and temporal stability then FSR doesn't even attempt to fix this so of course DLSS is much better.

The point is, you can't compare performance of the 2 if you ignore image quality, but image quality is not equivalent between the 2 because FSR is not a competitor to DLSS, it achieves very different results.



But overall, since DLSS used both temporal accumulation and DL, you will get away with rendering at a much lower base resolution to achieve sufficient quality, and as such the the performance will be much better.

 

[StarShock]

I think the best is to test FSR performance mode verses DLSS off as that's what most peoples graphics cards support, and then ignore the image quality as an ugly game with good performance is playable but a pretty game with bad performance isn't. Or at least I think that this is just as compelling an argument as the above

 

[D.P.]

Superior based on what exactly? On paper yes but in practice by how much?

You was only saying other day FSR 1.0 will be worst or just like DLSS 1.0

But it turns out that it's actually so much better.

NO it is not. FSR is veyr good at preserve geometric edge detail, but it does nothing at all for texture quality or fine details because away form any edges it is no more than a bin-linear filter with some sharpening on top, which is exactly what everyone was worried about. I even pointed out before FSR was released that FSR might use some kind of morphological AA to help with edges by leveraging the depth map, which is why FSR has to be incorporated in to the game engine before the final image. I said at the time if FSR comes with a well tuned sharpener then it would likely be comparable to DLSS and from a naive perspective, the sharpening will make people believe it is doing a better job than DLSS 1.

Again I will say this it doesn't matter how these techniques are done so long the end goal is to have you an image quality of the native resolution.

It doesn't matter how you get there.

Again I will use Freesync vs Gsync again both achieve the same goal both very different approaches.

I agree it doesn't matter about the technique but the results. But here is the problem, FSR is not even trying to do the same thing so the results are not really comparable.

If you look at texture quality alone, then FSR is terrible, is that only how we should judge it? No, FSR sets out to preserve geometric edge quality and it does a very good at doing that. Therefore it is a nice additional tool that will see some uses. But don't kid yourself into thinking it is a competitor to DLSS, AMD certainly doesn't think it is. AMD is trying to market FSR to limit the damage of DLSS and they are masters at pushing this anti-Nvidia agenda.
However, it is consumers that will loose out if AMD stop at FSR and don;t compete with Nvidia with their own DLSS alternative. And no, it doesn't necessarily need DL but the facts are to actually increase image quality during up-scaling you need a technology that provides more details or can leverage more information, or both.


FSR is mostly disapointing because it has added very little. I was really hoping for an open source TSR implementation.

 

[shankly1985]

@D.P.

If the end goal is to achieve a fake native resolution that is hard for the user to notice the difference be it glitches or add blur then surely the one that is closer to native is the better version?

FSR 1.0 doesn't suffer the issue that DLSS 1.0 did so it's better again you getting into the fine details of how these work but what matters is the end goal.

 

[fs123]

FSR wont appear in Unreal Engine. UE4 has TAA-U and UE5 has TSR (both very similar). These are superior to FSR so Epic wont waste resources on FSR

Are you still hanging on to the DF fake information regarding FSR/TAAU? In case you haven't noticed DF have been debunked and it has been proven FSR is better.

https://www.reddit.com/r/Amd/comments/o6skjq/digital_foundry_made_a_critical_mistake_with/

Spoiler: Native (DoF off)

Spoiler: FSR Ultra Quality (DoF off)

Spoiler: TAAU (DoF off)

The TAAU shot is blurry compared to the other two.

 

[nvidiamd]

A baseline might be DLSS on balanced snd FSR on ultraquality.

I am sure you are fair, you proved it so many times...

 

[ICDP]

So you are saying FSR ultra Quality is comparable to balanced?

Well I made the mistake of showing ignored content to see who you were responding too and I should have known better. That guy has been ******* on FSR (and everything AMD) ever since it was announced and he has made so many opinions masquerading as facts I have lost count how many are wrong. You would think people like him would learn to **** the **** up as they just keep showing themselves to be nothing but hot air and BS. So he is suggesting that FSR UQ is broadly equivalent to DLSS Balanced based on a set of metrics he has chosen as important.

These people (like the biased DF review) only look at the actual tech involved but form a very blinkered position where DLSS is seen as the ultimate aim. They ignore the other vital and very good aspects that even most reviewers acknowledged.

Open source
More easily implemented
Wider cross platform support
More potential userbase
No ghosting

They want to keep the narrative to only compare the actual output with a zoomed in screenshot so they can say "see DLSS does it better". And more often than not they would be right but if you are zooming in 2x or 3x times and pixel peeping to see where FSR fails and DLSS wins then we are doing it wrong. All other important metrics are declared irrelevant "oh but we must only look at the actual technology in a way I decide and ignore that it's limited to a very niche set of gamers". Yes FSR has its limitations and overall I prefer DLSS but FSR at 4K in Ultra comes very close and even FSR quality is not too far behind.

To perfectly demonstrate my point, he posted right below my post showing FSR is 100% coming to XBox. Yet not long ago he was saying this.

https://www.overclockers.co.uk/forums/posts/34842015

IN general this talk of FSR on consoles is bogus. Consoles already support various upscaling methods and techniques to reduce shader cost with hardware accelerated checkerboarding for example.

Given that better implementations of FSR already appear to exist such as TSR, then there isn't a big incentive to use FSR.

The problem AMD has being second to market and trying pretend that FSR is comparable to DLSS, is that reviewers will make that comparison and FSR will fall well short. Once that damage is done it will be hard to win mindshare beyond the AMD diehards

Not one aspect of that post has actually come true and in fact is pure BS. I bolded the parts in red because they are not presented as opinions, but as FACTS. I would ignore it if he had put a simple "In my opinion", or "I think" but he didn't. Yet many time he has tried to claim "I never said FSR wouldn't come to consoles".

• FSR is confirmed as coming to consoles (XBox)
  
• FSR has stood up quite well to DLSS according to the vast majority of reviewers. And my own testing on my RTX 3080

 

[nvidiamd]

If FSR is not added to UE it is because Epic has a deal with Nvidia, not because they have a "better" solution. That's BS, it is up to each game dev to decide which are the better solutions, i can make a deal with AMD and refuse to add DLSS because i already have a better solution.
Don't tell me a game dev will be so dumb that he will use FSR but refuse to use the "better solution" UE5 is already offering, if it is better. Or DLSS for that matter. It's all about money, the engines should integrate all the features from all major hardware vendors, especially if they are open sourced.

 

[nvidiamd]

I wasn't aware of DLSS introducing latency issues. Ghosting yes, but not input lag. Is there any proof to show this latency issue with DLSS?

It is possible to add latency in particular cases, if a game has very high framerates, depending how much time DLSS is taking for each frame. But even where it may decrease latency it has no chance against FSR, that thing will render the frames much faster than DLSS. And in competitive gaming where you play at lowest settings and the only thing you want to see on the screen is your enemy, FSR has the advantage.

 

[nvidiamd]

Look here the answer from Nvidia:
https://www.nvidia.com/en-us/geforce/news/nvidia-dlss-your-questions-answered/

"
DLSS is designed to boost frame rates at high GPU workloads (i.e. when your framerate is low and your GPU is working to its full capacity without bottlenecks or other limitations). If your game is already running at high frame rates, your GPU’s frame rendering time may be shorter than the DLSS execution time. In this case, DLSS is not available because it would not improve your framerate. However, if your game is heavily utilizing the GPU (e.g. FPS is below ~60), DLSS provides an optimal performance boost. You can crank up your settings to maximize your gains. (Note: 60 FPS is an approximation -- the exact number varies by game and what graphics settings are enabled)

To put it a bit more technically, DLSS requires a fixed amount of GPU time per frame to run the deep neural network. Thus, games that run at lower frame rates (proportionally less fixed workload) or higher resolutions (greater pixel shading savings), benefit more from DLSS. For games running at high frame rates or low resolutions, DLSS may not boost performance. When your GPU’s frame rendering time is shorter than what it takes to execute the DLSS model, we don’t enable DLSS. We only enable DLSS for cases where you will receive a performance gain. DLSS availability is game-specific, and depends on your GPU and selected display resolution."

FSR doesn't have this problem since it does everything in 1ms.

 

[BallistixOnZ490]

I tried it out finally in Anno 1800 and the results are astonishing.

At default everything maxed out, 2560x1440 8xAA, I get around 40-50 FPS at the start of a game tanking down to 30ish by the mid / end game after islands are filled up with buildings (same as what all reviews / benchmarks show).

Initially sticking FSR onto highest quality did next to nothing. Dropping it down to balanced ended up giving me 100+ FPS.

I found a balance at high quality and 4x AA offering 60-70 FPS, Anno 1800 being one of if not the most demanding RTS and mostly being CPU bound in the late game similarly to Civ 6.

So if demanding games will begin adding support for FSR, I don't need a new GPU for performance, only thing I'm missing out on is Ray Tracing which I suppose I can skip for longer until theres a massive backlog of ray traced games, and I wait for them to go on offer prices too.

Also I remembered to pick up the third Season Pass and some christmas DLC thing while it was a little over £1, as well as Anno 2205 complete for a tenner, Anno series being the only thing I ever need to use Uplay for.