• 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 in 2021

Lots of people vocal on something they clearly won't use as it's "not good enough".

Tbh I don't like the idea if resolution/IQ cheating at all, but here we are

This ^^ - reminds me of the FX5800Ultra and the `we dont render what`s not on the rails benchmark` from back in the day so it would be competitive with the 9700PRO
 
But comparisons show from reviews that ultra quality very little difference between native? So what the game is originally showing at native is kept intacked so by your logic the original native game lack good texture quality. :rolleyes:
Simmering pixels?

Mind showing some examples?

From GPU Open: https://gpuopen.com/fidelityfx-superresolution/

Just look at the vaseline effect on the grass




Top of the mech warrior guy has some weird artifacting. E.g. on his left shoulder there is a vent and FSR blurs all the details.
The plants have lost a lot of detail
Specular highlights on the rock below the bird are gone.


If you download the high res originals images then you see many more problems.
For example, the big palm tree bottom right is pixelated and soft.


Will try and get more exampels later, but things like fine details of hairs, wires andare also problematic for FSR because it simply doens;t have the samples to reconstruct.
 
200%??

Did you watch any of the release video for DLSS 1.0 on battlefield and Metro? it was easy to see just looking at video of the gameplay no need for screenshots or zoomed in.

yeah, the initial BF4 DLSS release was dire, although the patched version a few weeks later was already better. Nvidia made a big mistake in letting that go out.
 
You are still massively downplaying the effort required, but I can see I am not going to change your mind so I will leave it there.


The source code ill be released in a few weeks to end this argument, but essentially it is a post processing effecting, although it leverages the depth map. AMD have published enough details to make it clear what FSR does and how it does it. The complexity is no more than CAS for example.

This was not a high engineering resource feature. That doens;t make it a bad thing, I think FSR is very good for the likely resources AMD put in to it. I would be supposed if more than a few good engineers plus a QA team etc. spent more than 2-3 months form research to production.
 
From GPU Open: https://gpuopen.com/fidelityfx-superresolution/

Just look at the vaseline effect on the grass




Top of the mech warrior guy has some weird artifacting. E.g. on his left shoulder there is a vent and FSR blurs all the details.
The plants have lost a lot of detail
Specular highlights on the rock below the bird are gone.


If you download the high res originals images then you see many more problems.
For example, the big palm tree bottom right is pixelated and soft.


Will try and get more exampels later, but things like fine details of hairs, wires andare also problematic for FSR because it simply doens;t have the samples to reconstruct.


Again you asuming we all saying FSR is perfect and DLSS is also perfect?

The answer is the both have issues and if you do what you doing pixel peaking you will find issues with both!!!

The fact is if you just playing the game you would struggle to see these little tiny issues its not like DLSS 1.0 that was easy to see just by watching the gameplay videos on the reviews should we start digging DLSS 1.0 video content up?

Sorry but if I open the full res the issues you pointing out become very less obvious.

Your defence mode over this DLSS is assuming we all saying FSR is perfect but I bet you think DLSS is?

0-RiftB%20FSR%20OFF+titles.png


1-RiftB%20FSR%20Ultra%20Quality+titles.png
 
yeah, the initial BF4 DLSS release was dire, although the patched version a few weeks later was already better. Nvidia made a big mistake in letting that go out.

Wasn't just battlefield though was it? I knew you would pick that game out :D The big Metro release DLSS was also like you just said dire
 
There are two factors for me here:

I just find it a bit odd to see it lauded so much when despite being quite good for spatial reconstruction it falls short of what AMD could have pulled off with a bit more effort and incorporating other techniques including temporal sampling even without a deep learning model. Having spent a lot of time playing with Quake 2 RTX's upscaling it is very noticeable to me.

Then there are those people whose position is inconsistent between when nVidia does it and when AMD does it.


This os my opinion as well.

FSR in itself is quite good, nice results for what it does, well balanced sharpening and really good geometric edge detail. But it is very disappointing that AMD didn't provide gamers with a state of the art temporal upscaler that could have provided much better results.

The sad thing is, all this wild praise for a simple upscaler might hold back the progress of technology and impact gaming. It would be much better for all gamers if AMD started pushing a real DLSS competitor and generated a new battle forcing both IHVs to put significant resources into better and better image quality at lower input resolutions.

In effect, AMD's marketing machine has essentially convinced the AMD fans that image quality doesn't matter, that deep learning is nonsense. Very sad.
 
Where has the D.P. and or Rorff goalposts ended up today? Let's show ignored content to have a look (I feel like punishing myself) at the stupid levels of debate we have been brought down to today.

Ah, I see we are onto the argument that FSR introduces some blur and that AMD could have done better therefore FSR is crap. Oh but we still have the very much debunked "it's just a simple upscaler" nonsense. If only those FSR reviews had mentioned that FSR is on v 1.0 and AMD plan to improve it. Or that it does cause some blur but that to see it you need to pixel peep and zoon in to tell. Or that it isn't just a simple upscaler. Oh, wait, you mean they did mention these things and so did pretty much every other rational unbiased person who has tested it and posted here.

AMD released a technology that is not limited to the select few who have Tensor Cores, it does almost as well and better than most expected, is widely praised as a game changer for most GPU owners and is cross platform and easy to implement. Yet still they will say it's crap, NOT because it is, but because it isn't from their beloved Nvidia.

Nothing to see here, back to ignoring the Nvidia trolls.
 
Last edited:
This os my opinion as well.

FSR in itself is quite good, nice results for what it does, well balanced sharpening and really good geometric edge detail. But it is very disappointing that AMD didn't provide gamers with a state of the art temporal upscaler that could have provided much better results.

The sad thing is, all this wild praise for a simple upscaler might hold back the progress of technology and impact gaming. It would be much better for all gamers if AMD started pushing a real DLSS competitor and generated a new battle forcing both IHVs to put significant resources into better and better image quality at lower input resolutions.

In effect, AMD's marketing machine has essentially convinced the AMD fans that image quality doesn't matter, that deep learning is nonsense. Very sad.

Its bloody version 1.0 for god sake :cry: Honestly DLSS 1.0 was much worst.
Who knows what happens next? It surely improves yes? its not going to get worst, just like DLSS 1.0 it improved because of the poor feedback it received on release.
 
Again you asuming we all saying FSR is perfect and DLSS is also perfect?


The answer is the both have issues and if you do what you doing pixel peaking you will find issues with both!!!

The fact is if you just playing the game you would struggle to see these little tiny issues its not like DLSS 1.0 that was easy to see just by watching the gameplay videos on the reviews should we start digging DLSS 1.0 video content up?

Sorry but if I open the full res the issues you pointing out become very less obvious.

Your defence mode over this DLSS is assuming we all saying FSR is perfect but I bet you think DLSS is?


Quite the opposite, my stance all along has been that FSR and DLSS both have pros and cons and do some things better than others. The problem comes to the limitations of FSR due to the underlying technology and the fact no reviewer has yet compared FSR to DLSS due to AMD reviewer guidelines, therefore the relative strengths and weakness have not been properly conveyed.
 
Its bloody version 1.0 for god sake :cry: Honestly DLSS 1.0 was much worst.
Who knows what happens next? It surely improves yes? its not going to get worst, just like DLSS 1.0 it improved because of the poor feedback it received on release.


YEs, DLSS 1.0 was worse overall, but it showed the potential if you understood the technology. The problems this is 2021, FSR has to be compared to DLSS2.2.

Who know where FSR will go, except we know that they have started down a path with no temporal accumulation and no deep learning. That limits their options, and the weaknesses of FSR aren't going to be easily overcome by tweaking their filter.


Anyway, I need to go do the shopping.
 
Quite the opposite, my stance all along has been that FSR and DLSS both have pros and cons and do some things better than others. The problem comes to the limitations of FSR due to the underlying technology and the fact no reviewer has yet compared FSR to DLSS due to AMD reviewer guidelines, therefore the relative strengths and weakness have not been properly conveyed.

And why not just let that happen? instead of getting into the fine details of why DLSS should be better because it uses gold dust technology to create machines from skynet to create the next frame?

You made the assumptions before SFR was reviewed and it turns out you was wrong that FSR was infact better than DLSS 1.0 you called this time and time again AMD would never create something as good but the supress that DLSS 1.0
 
YEs, DLSS 1.0 was worse overall, but it showed the potential if you understood the technology. The problems this is 2021, FSR has to be compared to DLSS2.2.

Who know where FSR will go, except we know that they have started down a path with no temporal accumulation and no deep learning. That limits their options, and the weaknesses of FSR aren't going to be easily overcome by tweaking their filter.


Anyway, I need to go do the shopping.


NOW COME ON.....

Why should FSR be compared to DLSS 2.2? you surely need to look at the first stepping stones of both companies right? Just like I have said its unfair to compare the performance of AMD first RT performance against 3000 series its actually against 2000 series for RT performance alone.

How much RDNA 3 will bring is the next big question.

RDNA 2 is nvidia last years competition its just amazing how much ground they got in normal performance and power consumption this time around.
 
Its bloody version 1.0 for god sake :cry: Honestly DLSS 1.0 was much worst.
Who knows what happens next? It surely improves yes? its not going to get worst, just like DLSS 1.0 it improved because of the poor feedback it received on release.
In that case should we not criticise it so AMD can improve it ASAP.?
When I tested it I admit I was impressed but quality loss in textures is visible to me on a 4K screen, not enough to disable it for the performance increase it provides but definitely visible. Geometric edges are nice and sharp tho.
Hopefully textures quality is the first thing AMD will improve in the next version.
 
In that case should we not criticise it so AMD can improve it ASAP.?
When I tested it I admit I was impressed but quality loss in textures is visible to me on a 4K screen, not enough to disable it for the performance increase it provides but definitely visible. Geometric edges are nice and sharp tho.
Hopefully textures quality is the first thing AMD will improve in the next version.

Yes definitely all software should have the pros and cons listed to be improved and I am sure Amd already knows its limitations.

Nvidia was forced to make changes with DLSS the biggest issue for nvidia at the time was marketing and brand name RTX was under a lot of backlash.

That premium price tag was under the hammer so what Nvidia did and rightly so was put all the effort into improving DLSS because they had to right you got hundreds people buying into this DLSS that was left switched off for a bit even when the GPUs released.

The difference WITH AMD approach is its not branded into the price of the GPU its a free software update for everyone.

What does this do? Browny points if its good, competition is good for us all

You now have GPUs from both side offer a taste of RT and image upscale.

Just like AMD's RT performance got some back lash you can bet the house RDNA 3 will focus a lot on the RT side of the GPU.
 
Most have never argued that FSR does not introduce some texture blur even at 4K, so it's a strawman argument introduced by the usual suspects. Most reputable reviews already have established that FSR is close to native at 4K that you have to look closely to tell it's on. We already know as the resolution gets lower so does the quality of FSR. But the same is true of DLSS to a lesser dgree but still there.

Is DLSS better, yes. Is it better by much, NO at least not at higher resolutions. Though At 1080p FSR is pretty poor to be honest and DLSS fairs much better at lower resolutions. AMD made a very good tech that works on a much wider user base of GPUs. Had they made it more reliant on recent GPU tech that would have limited the impact it has had.

AMD made the right choice.
 
Last edited:
This was not a high engineering resource feature. That doens;t make it a bad thing, I think FSR is very good for the likely resources AMD put in to it. I would be supposed if more than a few good engineers plus a QA team etc. spent more than 2-3 months form research to production.

Once AMD hit it with future hardware-level optimisations I think then we're gonna see it shine (obv 2.0+).
 
. It would be much better for all gamers if AMD started pushing a real DLSS competitor and generated a new battle forcing both IHVs to put significant resources into better and better image quality at lower input resolutions.
.
It would be better for gamers if nvidia and amd just built better GPUs that didn't rely on upscaling to work

In effect, AMD's marketing machine has essentially convinced the AMD fans that image quality doesn't matter, that deep learning is nonsense. Very sad.

Pretty certain that was nvidia marketing machine when they convinced pc gamers that upscaling is cool now.
 
"build better gpus"

Easier said than done.....

- we're not at that point where it is even possible, probably at least another 2 gens away, probably 3/4 if you want native res. ray traced reflections too
- if it were even possible now or for next gen can you imagine the price cards will be

Thankfully dlss/FSR allows us to enjoy ray traced intensive games 60+ fps @1440 or 4k now (and for nvidia users, they have been able to enjoy it for the last what 10 months now? [when games actually started to use ray tracing better] as opposed to having to wait potentially another 2/3+ years till we get to the stage we are at now.

Also, dlss/fsr will allow developers to add even more graphical effects (outside of ray tracing) to games now since we can claw back a significant amount of performance with the ever so slight hit to "native" clarity/sharpness.

Sadly though, for PC, developers will more likely see FSR/dlss as a chance not to optimise games performance though......
 
Back
Top Bottom