What DLSS/FSR Upscaling Mode do you use?

[JustBenching]

Interesting you say that, I ran a DLSS test in Forza Horizon 5 and I thought that it appeared blurry. At the time I was using a 55" 4k TV which has poor pixel density as it is, but still, I expected more.

Pick from best to worst

Spoiler

 

[wolf]

Pick from best to worst

Spoiler

View attachment 380495
View attachment 380496

Don't hold your breath for much engagement on that, at least not candid engagement. Yet, I feel there would be a strong correlation between either ignoring it or offering criticism on your comparison, and those whom would happily support cherry picked RT on v off stills showing as little difference as possible. If it were a Venn diagram, it'd be a circle.

Like RT though, upscaling has a problem of consistency, some RT implementations/scenes are great and worth it, and some are effectively so pointless it's a net negative. For upscaling it's similar in the sense that some upscaling is excellent, and some.... Well yeah I can see why they don't like it, it was just never going to give great results on their setup.

 

[JustBenching]

Don't hold your breath for much engagement on that, at least not candid engagement.

YEAP, they disappeared. Makes you really wonder why

 

[AusWolf]

Pick from best to worst

Spoiler

View attachment 380495
View attachment 380496

The top picture appears slightly smoother, with fewer artefacts around the steering wheel and on the dry leaves on the ground. The texture of the car's roof looks all wrong, though, I don't know what happened there.

 

[JustBenching]

The top picture appears slightly smoother, with fewer artefacts around the steering wheel and on the dry leaves on the ground. The texture of the car's roof looks all wrong, though, I don't know what happened there.

So which one do you prefer? Might I add here, there is a 40% fps difference between the 2 images!!

 

[wolf]

The texture of the car's roof looks all wrong, though, I don't know what happened there.

It's the jeep trailcat iirc, and the texture is tough for any algo, top image manages it with minimal moire effect, bottom is just smoosh. I also find a fairly intense difference in the vegetation.

 

[JustBenching]

It's the jeep trailcat iirc, and the texture is tough for any algo, top image manages it with minimal moire effect, bottom is just smoosh. I also find a fairly intense difference in the vegetation.

Bottom isn't just native, it's native + 4x supersampled with MSAA Absolutely wild

 

[AusWolf]

So which one do you prefer? Might I add here, there is a 40% fps difference between the 2 images!!

I don't know... I see ups and downs in both. I'd have to see it in action, I guess. A 40% difference in FPS is pretty substantial, though.

 

[JustBenching]

I don't know... I see ups and downs in both. I'd have to see it in action, I guess. A 40% difference in FPS is pretty substantial, though.

So now you know why people use DLSS

 

[lexluthermiester]

A 40% difference in FPS is pretty substantial, though.

True! And if that gain can be achieved without making the displayed image look worse or terrible, then hell yes it's worth using.

For me personally, still not impressed. They keep promising that it'll look good and improve performance and have yet to deliver on that promise. In all fairness they're getting close. Close enough that I could be talked into using it daily, maybe. They need to find a way to render a scaled image to look as good as a native image.

So now you know why people use DLSS

Sure, if image quality isn't important. Hell, I'm a Frame-Rate-Is-Life kind of guy, which is why I never use AA of any kind. Scaling is a different beast and the quality of image difference are too distracting to justify using it in any form.

 

[wolf]

Sure, if image quality isn't important

Why be so reductive? This is a conversation and there's clearly room to talk about image quality weaknesses, but also their strengths and differences. To insinuate someone using DLSS doesn't value image quality at all is not an argument in good faith.

I can see you are clearly and vocally not an user or enjoyer of antialiasing of any kind, which I can respect as your personal choice but also appreciate that antialiasing is commonplace and not only desired but expected at least in some form.

 

[JustBenching]

Sure, if image quality isn't important. Hell, I'm a Frame-Rate-Is-Life kind of guy, which is why I never use AA of any kind. Scaling is a different beast and the quality of image difference are too distracting to justify using it in any form.

It's because image quality is important that im using DLSS. Each to his own I guess.

 

[lexluthermiester]

It's because image quality is important that im using DLSS. Each to his own I guess.

In it's current form DLSS can NOT look as good as an unprocessed native image.

 

[JustBenching]

In it's current form DLSS can NOT look as good as an unprocessed native image.

To me it does, if you are willing to take a blind test on it then ill admit i might just be blind

 

[wolf]

In it's current form DLSS can NOT look as good as an unprocessed native image.

Unprocessed native images have extensive issues in motion, and even in stills. "Good" in this context is a mixture of quantifiable and subjective.

Perhaps you might want to accept that your personal desires and tastes for image quality and processing when gaming aren't universally accepted truths for literally everyone on the planet.

 

[theouto]

Have people forgotten the meaning of preferences and "to each their own"?

No one side can convert the other side due to this being a very opinionated topic, it's just childish now.

Quick sidenote: The real test should be an image undergoing rapid movement (as usually videogames are undergoing rapid movement), and the captured images should then be shown as uncompressed PNGs, but even then, the only way of knowing if you like it or not is actually using it in real time, static images can only do so much.

 

[JustBenching]

No one side can convert the other side due to this being a very opinionated topic, it's just childish now.

That's not true, I can be easily convinced. Someone takes a blind test, detects the native 10/10 times, that's it, done, im now a native supporter and ill admit my eyesight is crap cause I can't tell them apart.

Quick sidenote: The real test should be an image undergoing rapid movement (as usually videogames are undergoing rapid movement), and the captured images should then be shown as uncompressed PNGs, but even then, the only way of knowing if you like it or not is actually using it in real time, static images can only do so much.

I agree, but I have to add, the real test should be done at ISO performance as well. There is no point comparing one running at 200 fps and the other one at 50. And when you do that (either via DLAA or DLDSR + DLSS Q) native is so far behind in image quality that it's a head scratcher why people are running native (assuming they have access to DLSS of course).

 

[wolf]

Have people forgotten the meaning of preferences and "to each their own"?

It's one of the most frustrating things about this topic, people that seem to speak in absolutes, like there's no room for discussion or preference whatsoever.

I don't deny that some people see what they see and like what they like, yet to talk as if the way they see the world is the ground truth that applies to all, especially when they're not even taking in and adequately considering all the possible sources of information, It boggles the mind...

Add to that a sprinkle of bias, stubbornness and forum mechanics, and you've got a recipe for a spicy debate that like clockwork becomes a race to absolutely nowhere.

If there were an eye roll or face-palm react on this forum that button would get a workout alright.

 

[AusWolf]

So now you know why people use DLSS

FSR does that, too, to be fair. And I still only need that 40% if performance is below my acceptable standard otherwise. I'm not in the "more performance is always better" group. Give me 60 FPS, and I'm good.

True! And if that gain can be achieved without making the displayed image look worse or terrible, then hell yes it's worth using.

For me personally, still not impressed. They keep promising that it'll look good and improve performance and have yet to deliver on that promise. In all fairness they're getting close. Close enough that I could be talked into using it daily, maybe. They need to find a way to render a scaled image to look as good as a native image.


Sure, if image quality isn't important. Hell, I'm a Frame-Rate-Is-Life kind of guy, which is why I never use AA of any kind. Scaling is a different beast and the quality of image difference are too distracting to justify using it in any form.

I still don't see the point in using it as long as your performance is acceptable. Which brings me to the fact that game devs tend to use upscaling as an excuse to get away with poor optimisation, which is wrong.

It's one of the most frustrating things about this topic, people that seem to speak in absolutes, like there's no room for discussion or preference whatsoever.

I don't deny that some people see what they see and like what they like, yet to talk as if the way they see the world is the ground truth that applies to all, especially when they're not even taking in and adequately considering all the possible sources of information, It boggles the mind...

Add to that a sprinkle of bias and stubbornness and you've got a recipe for a spicy debate that like clockwork becomes a race to absolutely nowhere.

Even when I speak about what I see and what I prefer, some others still see it as if I was talking in absolutes, which I am not. I don't know why some people are hell bent on convincing me of their truths.

It kind of feels like one can't share an honest opinion these days without others jumping at one's throat, especially if it's an opinion against some world-breaking glorious Nvidia feature.

Or should I write "this is my personal opinion - it has no reflection on yours" at the end of every single comment I make? Have we as a community become so intolerant?

 

[Degreco]

Look at powerlines or trees in some games to see where DLSS still needs some training. To be honest, in fast paced action games like Spiderman or Doom Eternal you won't see the artifacts, if there are any to begin with. But you will love the fps boost. In slower games you can see the (occasional) artifact, e.g. when you are slowly riding through a high poly landscape in RDR2 or flying low altitude in MSFS. So it's not surprising to see believers and naysayers in one place, they don't play the same games. Also, if you are evolutionized with crt monitors or lower resolutions, you sometimes don't perceive artifacts as graphical glitches. Other players, once they detect artifacts in their game of love, can never 'unsee' them. So opinions often vary.

Overall, this tech will only get better like any AI model: With more training on the base upscaling models and more corrections for specific games. At least in the case of DLSS the models are done by Nvidia inhouse, so they have no problems to get their hands on some GPUs

 

[wolf]

Or should I write "this is my personal opinion - it has no reflection on yours" at the end of every single comment I make?

You quoted me and so appear to be asking my honest opinion here so I'll give it. Your opinions, from my perspective, could stand to be phrased more like an opinion. Please don't go overboard with what I've just said and take that too far, like what I just quoted from you, your example is well made but purposely over the top. Yet, a little could go a long way. Hey I could be wrong, but you kinda asked. If you want to discuss further I'd be happy to quote specifics and give feedback.

I want to be clear here though, what I'm actually talking about referencing in this thread in the last page or two isn't you, it's another user who I won't name in the interest of civility, but I'd wager you know.

 

[JustBenching]

FSR does that, too, to be fair. And I still only need that 40% if performance is below my acceptable standard otherwise. I'm not in the "more performance is always better" group. Give me 60 FPS, and I'm good.

I've always said that even FSR does - in most cases - look better than native at iso performance.

Saying I don't always need that 40% performance is weird. Of course you do, it allows you to buy a cheaper GPU than the one you'd have normally bought.

And it's an honest question, in a blind test what % would you get correct you think between native and DLSS?

 

[AusWolf]

You quoted me and so appear to be asking my honest opinion here so I'll give it. Your opinions, from my perspective, could stand to be phrased more like an opinion. Please don't go overboard with what I've just said and take that too far, like what I just quoted from you, your example is well made but purposely over the top. Yet, a little could go a long way. Hey I could be wrong, but you kinda asked. If you want to discuss further I'd be happy to quote specifics and give feedback.

I don't know about others, but personally, I think a forum is a place meant for sharing ideas and learning from others, and I intend to use it that way. My truth is mine alone, and it's not the ultimate one. Nevertheless, it's based on a personal experience just as valid as yours, or anyone else's. There is no right or wrong.

Ultimately, I may hate upscaling, and you may love it, but I still think we're both right in our own ways. It's quite similar as the 4080 vs 7900 XTX argument. Some people will gladly pay more for a better RT performance. Some people don't care. And they're equally right.

This is what I'd say to everyone registered here: there is no ultimate truth. Whatever kind of experience you prefer is yours alone.

I want to be clear here though, what I'm actually talking about referencing in this thread in the last page or two isn't you, it's another user who I won't name in the interest of civility, but I'd wager you know.

Fair enough.

I've always said that even FSR does - in most cases - look better than native at iso performance.

Let's just agree to disagree there.

Saying I don't always need that 40% performance is weird. Of course you do, it allows you to buy a cheaper GPU than the one you'd have normally bought.

Not if it comes at a cost of worse image quality. My personal limit for a GPU is roughly £500, and I try to choose the one that gives me the highest raster FPS within that budget. Going a tier lower shoots me in the foot in the long run - and if I have to use upscaling, then in the short run, too.

And it's an honest question, in a blind test what % would you get correct you think between native and DLSS?

A screenshot is different from a live game. You can spot details in a screenshot that you wouldn't in a live game (pixel peeping), or some differences can become more apparent in the long run.

For example, I just fired up Senua's Saga: Hellblade 2, and had to turn on FSR for it to be stable at 1440 UW. At first I thought it looked really not bad. But after about 20 minutes of gameplay, the blur became more and more apparent.

I do have to say, FSR 3 is a massive improvement over previous versions, and if FSR 4 is an even bigger improvement, then I'll be happy to consider using it daily in the newest games, especially if it makes me able to postpone my next GPU upgrade a bit more.

On another note, I'm currently playing The Talos Principle, where there is no FSR, DLSS, dynamic resolutions or any of that crap, and I have to say, it looks absolutely glorious!

 

[wheresmycar]

Pick from best to worst

Spoiler

View attachment 380495
View attachment 380496

Its a mix bag. If we're going by still images, if possible i'd tweak the settings to find the middle ground. Top image looks smoother but the plants/veg look a little mushy/blurred. Bottom image carries minor artefacts and the plants/veg look over-sharpened. Both to my distaste.

I got an idea... if you're willing to aid and assist. How about 2 short video samples using 1440p. Create a new thread with a poll. This could be valuable for those unfamiliar with DLSS and might even spark their interest. Starting a new thread will also help attract a broader audience to pitch in.

 

[lexluthermiester]

Unprocessed native images have extensive issues in motion, and even in stills.

Utter Moose Muffins.

Why be so reductive?

Because I expect excellence. I expect something that improves performance and not also degrade quality.

I still don't see the point in using it as long as your performance is acceptable.

Agreed. I'll take that a step further: I would rather reduce resolution than use scaling.

Which brings me to the fact that game devs tend to use upscaling as an excuse to get away with poor optimisation, which is wrong.

Also agree.