Discussion TPU: DLSS tested in Final Fantasy

[BFG10K]

https://www.techpowerup.com/reviews/NVIDIA/DLSS_Super_Sampling_AA_Tested_Final_Fantasy/

As if we needed more evidence of the disaster that is Turing.

This glorified upscaler DLSS is a blurry jagged mess in comparison to TAA @ 4K. Also at the same 1440p resolution, TAA is far faster given it's basically free to enable.

Oh, and DLSS only works at exactly 4K resolution in this game. So much for "deep learning" and "AI".

Epic fail.

 

[amenx]

Actually pic 2 doesn't really favour DLSS, since it is still clearly blurrier than the alternative (FXAA), it isn't nearly as pronounced as when comparing to the first picture though. One thing you still suffer from though is the fairly hefty over-sharpening artifacts you get with DLSS, which you don't get with the post-processing AA,

pic 5 is a 1440P image and one that runs 30% faster than DLSS, so you would bloody well hope that DLSS looks better here, otherwise it would obviously be a complete failure.

Also I actually already pointed out picture 5 in my original post (and noted that it looked worse than DLSS), so I don't know why you're accusing me of cherrypicking?

Apologies. Only read the first sentence "doesnt look good" and thought all pics were DLSS.

 

[BFG10K]

BTW: DLSS looks much superior to TAA when you play FF15. But i guess playing games dont matter anymore, too.

It sounds like you have first-hand experience? In that case you must have a 4K monitor.

So tell me, did you purchase the monitor with the intention of running all your other games 2560 upscaled with AA?

Not to mention that 4K high refresh monitors are $two-thousand a pop, which makes DLSS even more ridiculous. Who the hell spends $2K to intentionally upscale?

It's astonishing the contorted explanations we get for this garbage in a desperate attempt to emotionally justify it.

If AMD had shipped an upscaling AA method on a $1200 card that only works at exactly one resolution, it would've rightly been panned as excrement. But when nVidia does it, it's "performance innovation".

Article you quote pretty much says the opposite:

I can look at pictures just fine, I don' t need a reviewer to form an opinion for me.

 

[arandomguy]

If the criteria is to run every game at 4k high refresh without any compromise then people are going to be waiting for quite some time.

What the actual question should be is given the need for compromise does DLSS provide a better user experience versus other options such as running non native, lower settings, lower FPS, and etc?

 

[beginner99]

While I agree that Turing release is crap and the price hikes plain offensive, I always have trouble spotting actual differences in these still images. Yeah they look a bit different but I can't really say one is worse than the other. And if it isn't clear lookin minutes at still images, then it will barley matter in game play and hence the method with highest fps is the better one.

 

[sontin]

It sounds like you have first-hand experience? In that case you must have a 4K monitor.
So tell me, did you purchase the monitor with the intention of running all your other games 2560 upscaled with AA?

DLSS works just fine with DSR. So i dont need a 4K display. Oh if i would have bought a 4K display i would take every option to improve performance when the picture is sharp enough to give me the illusion to have a native picture. But lucky me i dont have to.

I can look at pictures just fine, I don' t need a reviewer to form an opinion for me.

You want pictures? Here is a comparision between TAA and DLSS when you are not in a city:
TAA: https://s15.directupload.net/images/181216/ube92xlc.png
DLSS: https://s15.directupload.net/images/181216/yiufurh4.png

So do you still believe that TAA is the way to go in FF15? TAA works fine when there isnt a lot of geometry in the picture. But it falls apart when you ran through the wilderness of FF15.

 

[Spjut]

I wonder if DLSS holds up any better in motion. For example, when Digitalfoundry has analyzed the checkerboard rendering on PS4, they say it's alot forgiving in motion but breaks apart when you pause and analyze the scenes.

 

[zebrax2]

@sontin
While the increase in performance is great i don't think the loss in visual fidelity when using DLSS, using the screenshot you posted for comparison, are worth it. The trees and bushes looks absolutely terrible on the DLSS screenshot. The details on the leaves are almost completely gone for both the trees and bushes making them look blocky and 2d-ish

 

[maddie]

@sontin
While the increase in performance is great i don't think the loss in visual fidelity when using DLSS, using the screenshot you posted for comparison, are worth it. The trees and bushes looks absolutely terrible on the DLSS screenshot. The details on the leaves are almost completely gone for both the trees and bushes making them look blocky and 2d-ish

Agree, and will add the white railing running across the scene. A mess with DLSS, but trying to argue this is pointless, as to each, their own opinion.

 

[amenx]

Gamernexus examines DLSS vs TAA in FF.

 

[Dribble]

So that was interesting - it's DLSS that's flickering, and that's the biggest problem with it not the loss of detail. Personally the flickering would get on my nerves pretty quickly when I noticed it. Over to Nvidia to see if they can fix that (which should be possible, they just need to tweak the AI to get it to learn to inhibit flickering).

 

[BFG10K]

DLSS works just fine with DSR. So i dont need a 4K display.

Not sure if you're on 1080p or 1440p but regardless, either DLSS or DSR isn't getting integer scaling (maybe both), so you're getting scaling on top of scaling, which makes things even worse than regular DLSS.

Comedy gold.

Oh if i would have bought a 4K display i would take every option to improve performance when the picture is sharp enough to give me the illusion to have a native picture.

Resolution is the last thing you'd drop. There's a fundamental difference between native & non-native. All the "deeeeeeep learning" in the world can't resolve that.

So do you still believe that TAA is the way to go in FF15? TAA works fine when there isnt a lot of geometry in the picture. But it falls apart when you ran through the wilderness of FF15.

Did you see the video? Specifically the flickering geometry and texturing that DLSS has during movement, or the randomly shifting pixels while standing still?

The last problem actually happened to AMD's awful MLAA implementation back in the day.

And again, the TAA used in Final Fantasy isn't the best. Unreal 4's is better.

 

[sontin]

Resolution doesnt matter when games need Post-AA to smooth transitions between objects. When you use TAA in FF15 the picture is not "native" anymore. It gets modified by the TAA in such a way that you lose a lot of details. You are basically waste performance to generate a native 4K picture and then lose most of the details to TAA...

I can produce videos to show how bad TAA is in this game. But it doesnt tell the whole story. Thats the problem today with such sites/channels like Gamersnexus. They have an agenda. Computerbase has a much better video, 5mins long nearly uncut: https://www.computerbase.de/2018-12/final-fantasy-xv-nvidia-dlss-test/

Download and watch it on a 4K display/TV. You see crawling with TAA, too. And i bet most people will choose the sharper DLSS image over the blurry TAA one.

 

[antihelten]

You want pictures? Here is a comparision between TAA and DLSS when you are not in a city:
TAA: https://s15.directupload.net/images/181216/ube92xlc.png
DLSS: https://s15.directupload.net/images/181216/yiufurh4.png

So do you still believe that TAA is the way to go in FF15? TAA works fine when there isnt a lot of geometry in the picture. But it falls apart when you ran through the wilderness of FF15.

DLSS still clearly blurs significantly compared to TAA in those shots:

TAA
DLSS

In fact it very much seems to have that whole vaseline/water color look to it that FXAA and certain smartphone camera algorithms have been criticized for in the past:

TAA
DLSS

It also looks like there are some patterns that it handles poorly, like here, where it turns a checkered pattern into a striped pattern:

TAA
DLSS

Now there are clearly some issues present in the TAA shots that you don't see in the DLSS shots (example), but those seem to be caused by FF15 using screendoor transparency instead of proper alpha blending (due to it using deferred rendering I guess), and not an issue with TAA per say.

Now one could of course argue that all of this is nitpicking, but when DLSS apparently doesn't even perform any better than TAA, when normalizing the image quality (which happens at roughly 1800P according to techspot.com), then it becomes really hard to see what purpose the existence of DLSS really even serves, especially when it is proprietary and requires training before it can be implemented.

 

[sontin]

Do you play your games with a 10x zoom?!

4K is great for pictures but not for games.

 

[Rifter]

I agree with the OP, both DLSS and RT are just complete gimmicks, tacked on so they could try and justify their price hikes. Neither is actually useful in a real life gaming situation. RT has potential but until they can get 1080p@60FPS at $250 price point(so xx60 series) its never going to become mainstream, the install base of hardware has to be there for the devs to put the work into the software side and make it happen on a large scale to become well adopted.

I was waiting for this gen of Nvidia cards to upgrade to after vega launch was not good with insane prices. I actually ended up with a vega after all when the price dropped for black friday and it turned out this gen from Nvidia was a dud.

 

[antihelten]

Do you play your games with a 10x zoom?!

4K is great for pictures but not for games.

You do understand that the exact same thing could be said when comparing DLSS with 1440P TAA (which is significantly faster), right?

In other words DLSS looks better than 1440P TAA if you zoom in, but at native resolution and in motion, you likely won't notice the difference. You will however notice the 30% performance boost you get with 1440P TAA.

 

[sontin]

You do understand that the exact same thing could be said when comparing DLSS with 1440P TAA (which is significantly faster), right?

The difference is magnitude bigger than between 4K/TAA and DLSS:

The 25% performance lost is more justable here than going from DLSS to 4K/TAA.

In other words DLSS looks better than 1440P TAA if you zoom in, but at native resolution and in motion, you likely won't notice the difference. You will however notice the 30% performance boost you get with 1440P TAA.

If you cant see a real difference than you can just scale 1440p up.

 

[antihelten]

The difference is magnitude bigger than between 4K/TAA and DLSS:

I don't know why, but something is clearly not right with that guys videos. 1440P TAA shouldn't be anywhere near that blurry. In fact if you look at his 4K TAA video comparison it is also blurry compared to DLSS, which is the exact opposite of what every other site out there is seeing. So excuse me if I don't put much stock in his comparisons.

 

[coercitiv]

4K is great for pictures but not for games.

This little gem must remain etched in the fabrics of the forum.

I wonder where all the 4k gaming critics were up until a few months ago. It's almost as if we needed help from Nvidia to see the truth.

 

[BFG10K]

Resolution doesnt matter when games need Post-AA to smooth transitions between objects. When you use TAA in FF15 the picture is not "native" anymore. It gets modified by the TAA in such a way that you lose a lot of details. You are basically waste performance to generate a native 4K picture and then lose most of the details to TAA...

What are you talking about? Post-processing only works on pixels because it happens after rasterization. That applies to DLSS because it's just an upscaling postfilter. That means resolution very much matters because it determines how much data you have to sample.

I mean are you really trying to say there's no different between 640x480 and 4K on a 4K display as long as you use TAA?

4K is great for pictures but not for games.

You heard it here first, folks. Native resolution doesn't matter and neither does 4K, and upscaling postfilters are the pinnacle of image quality. It's why upscaled DVDs are the industry gold standard.

Gsync modules don't work for desktop pictures, so why does nVidia insist on ramming them into each and every one of the 4K high refresh monitors if 4K isn't for gaming?

 

[wilds]

Does anyone here actually use TAA though? I think it looks terrible to use any form of AA at 4k.

 

[BFG10K]

Does anyone here actually use TAA though?

It looks stunning in Unreal 4 engine games and is consistent, even with the smallest indie developers.

Very little blurring, perfect image quality (completely clears up even the most stubborn specular aliasing), and zero performance cost.

Even on 1080p 27" native when I tested it, it still looked like an offline render, and that's a relatively low PPI.

I think it looks terrible to use any form of AA at 4k.

AA still makes a visible difference at 27" 4K, but I tend to agree from a simplicity of configuration standpoint, I personally wouldn't use it. At that level of pixel density a plain image looks good enough.

 

[wilds]

It looks stunning in Unreal 4 engine games and is consistent, even with the smallest indie developers.

Very little blurring, perfect image quality (completely clears up even the most stubborn specular aliasing), and zero performance cost.

Even on 1080p 27" native when I tested it, it still looked like an offline render, and that's a relatively low PPI.


AA still makes a visible difference at 27" 4K, but I tend to agree from a simplicity of configuration standpoint, I personally wouldn't use it. At that level of pixel density a plain image looks good enough.

I game on a 27" 4k monitor funnily enough. I'll have to keep playing around with it. Aliasing is clearly visible for me, but I prefer that over the blur I get from TAA.

We all must have also come across poor TAA implentations on at least 1 game. The worst offender I have come across is Rainbow 6 Siege, which is terrible looking. Glad to hear it looks better on Unreal.

I would love to see some 4k no AA screenshots, but the 1440p no AA screenshot will suffice.

R6 Siege TAA:
https://images.nvidia.com/geforce-c...active-comparison-001-4x-txaa-vs-4x-msaa.html

TAA in R6 provides a massive performance boost though. It allows resolution to scale as low as 160p which is insanely low. Because of this, even a GT 1030 can be a 144hz card.

 

[Adawy]

It looks stunning in Unreal 4 engine games and is consistent, even with the smallest indie developers.

Very little blurring, perfect image quality (completely clears up even the most stubborn specular aliasing), and zero performance cost.

Even on 1080p 27" native when I tested it, it still looked like an offline render, and that's a relatively low PPI.


AA still makes a visible difference at 27" 4K, but I tend to agree from a simplicity of configuration standpoint, I personally wouldn't use it. At that level of pixel density a plain image looks good enough.

I have to agree, and it's not just Unreal Engine, Unity as well, Inside looked absolutely incredible with TAA.