Info Hardware Unboxed: Massive DLSS to FSR comparison

[Heartbreaker]

26 games X 2 upscalers X 4 Modes (4K Qual to 1440p Perf) = Absurd amount of work.

Conclusion DLSS is better. Here is the Matrix image for all the games. More + signs indicate a bigger win (there are no FSR wins):



I'm more looking forward to the comparison to native he is working on...

 

[Mopetar]

I wouldn't be surprised if this were the case, and I don't really care since I despise both technologies, but you would want to use some kind of double-blinded study to make any kind of objective claims.

 

[KompuKare]

Versus native in a double blind would be interesting.

If native losses, then the AA on native must be the issue. Or upscalers are sharper than they should be.

 

[Heartbreaker]

I wouldn't be surprised if this were the case, and I don't really care since I despise both technologies, but you would want to use some kind of double-blinded study to make any kind of objective claims.

It's not a scientific study here. He shows you exactly the problems he is seeing. The guys at HWUB are quite fair and if anything they usually lean towards fighting against NVidia, so I don't think there is any kind of Pro NVidia bias going on with their results.

 

[SteveGrabowski]

Not too surprising. I'm usually pretty impressed with how good FSR and RSR look upscaling to 4k from a render resolution of 1440p or higher, while using either from a 1080p render (so 4k performance mode) doesn't look much better than native 1080p in my experience. Wish I could have gotten DLSS and a sufficient amount of VRAM but figured the extra VRAM was the better gamble when I went 6700 XT ($380) over 3060 Ti ($410). Though I would have probably chosen a hypothetical 3070 12GB had Nvidia ever made one even if was say $550.

 

[Mopetar]

It's not a scientific study here. He shows you exactly the problems he is seeing. The guys at HWUB are quite fair and if anything they usually lean towards fighting against NVidia, so I don't think there is any kind of Pro NVidia bias going on with their results.

I'm not accusing them of any kind of bias, and HUB has a good track record of identifying and correcting issues with any results they publish so they'd be one of the last sites that I'd even suspect of doing anything shady. Particularly anything that would favor NVidia given recent history with them trying to blacklist HUB.

They do about as good of a job of reporting and demonstrating why their results are the way they are, but it's still one person's subjective experience. They can't show you anything they didn't see or notice because they didn't see or notice it. Unless they release all of the testing footage for someone else to look over, you wouldn't be able to tell.

Frankly it's not worth it for any reviewer to even go to the level that HUB already have because it's not going to change anyone's mind and even a full blown study using appropriate methodologies and population wouldn't make most people budge either.

 

[DAPUNISHER]

Count me among those that are glad DLSS, FSR, and XeSS are in active development and use. I don't want it left to games to provide upscaling, or not. It allows older cards to stay out of the trash that much longer. And DLSS won't influence my purchasing decisions in the future, because the quality difference isn't worth paying a premium for by a longshot. I use FSR on my APUs, because they need it. Other then that, I avoid upscaling.

 

[Heartbreaker]

Unless they release all of the testing footage for someone else to look over, you wouldn't be able to tell.

Frankly it's not worth it for any reviewer to even go to the level that HUB already have because it's not going to change anyone's mind and even a full blown study using appropriate methodologies and population wouldn't make most people budge either.

That wouldn't make anyone's budget. Unless it's NVidia and then obviously no one would/should trust the results.

I think Tim at HWUB has a keen eye for artifacts and I don't think I would spot anything he missed.

I see little additional value in a double blind big population study in this case anyway. All that will tell is that most people are oblivious, and I already know that.

I think Tim mentioned somewhere in the video that he has many hours more video that he will probably make available to their Patreon members.

Also anyone with the right cards, and specific games can double check the results without needing his footage.

 

[Mopetar]

I'd trust Tim's eyes more than my own. Also I do recall someone did attempt a small scale study of this sort a few years ago when DLSS and RT were new. I'd only consider it anecdotal as the population size was still small (around 5 people) but the results seemed to suggest that common users had a much harder time at telling the difference or spotting artifacts that would give away DLSS/FSR.

However, that would make the whole notion of such a study far more comical as it's possible that average users may not be able to tell the difference between the two technologies. It reminds me of another such study about smartphone cameras where an Android-focused site and its users picked the iPhone as having the best camera, which led to a fair amount of butthurt among some users of the site.

 

[Heartbreaker]

I'd trust Tim's eyes more than my own. Also I do recall someone did attempt a small scale study of this sort a few years ago when DLSS and RT were new. I'd only consider it anecdotal as the population size was still small (around 5 people) but the results seemed to suggest that common users had a much harder time at telling the difference or spotting artifacts that would give away DLSS/FSR.

However, that would make the whole notion of such a study far more comical as it's possible that average users may not be able to tell the difference between the two technologies. It reminds me of another such study about smartphone cameras where an Android-focused site and its users picked the iPhone as having the best camera, which led to a fair amount of butthurt among some users of the site.

I'd bet most average users couldn't tell the difference between both technology and native either.

 

[Mopetar]

It's easy to notice the difference with native if you know what you're looking for, but while gaming you may not be looking at those parts of the screen/scene.

Both technologies have gotten a lot better and will no doubt continue to improve, but I dislike them for the simple fact that the introduce so much subjectivity into comparisons and the utterly misleading marketing that is being put out due to these technologies.

Frankly it's bad for the consumer. The bigger issue is when the technology gets good enough that even reviewers are having a difficult time spotting the difference during regular gameplay that the companies will just flip the switch so that it's an always on technology so they can sell less capable hardware while still claiming performance improvements.

 

[maddie]

I'd bet most average users couldn't tell the difference between both technology and native either.

Didn't you get the memo? It's better than native.

 

[GodisanAtheist]

I'd trust Tim's eyes more than my own. Also I do recall someone did attempt a small scale study of this sort a few years ago when DLSS and RT were new. I'd only consider it anecdotal as the population size was still small (around 5 people) but the results seemed to suggest that common users had a much harder time at telling the difference or spotting artifacts that would give away DLSS/FSR.

However, that would make the whole notion of such a study far more comical as it's possible that average users may not be able to tell the difference between the two technologies. It reminds me of another such study about smartphone cameras where an Android-focused site and its users picked the iPhone as having the best camera, which led to a fair amount of butthurt among some users of the site.

- I'd be fascinated by a double blind with a decent sized group.

I wonder how many people would actually prefer the upscaling solutions to native due to over sharpening or "motion blur aka ghosting" or softened textures etc.

I keep thinking about how a lot of default settings on TVs are oversaturated to hell because the average Joe likes the stimulating colorful pop over the signal accurate stuff that might be important to a professional.

 

[DAPUNISHER]

- I'd be fascinated by a double blind with a decent sized group.

I wonder how many people would actually prefer the upscaling solutions to native due to over sharpening or "motion blur aka ghosting" or softened textures etc.

I keep thinking about how a lot of default settings on TVs are oversaturated to hell because the average Joe likes the stimulating colorful pop over the signal accurate stuff that might be important to a professional.

I should think Nvidia did focus groups for it. Linus testing his employees, and some of them picking frame generation told me all I need to know. Which is that it's like art. People like what they like. Also that someone that isn't a pixel peeping gamer, evaluates what they are seeing differently.

I have to confess, if AMD releases their version, and it works on APUs, I will be stoked about it.

 

[coercitiv]

I'm assuming @DAPUNISHER is talking about their Async Reprojection test:

 

[WelshBloke]

- I'd be fascinated by a double blind with a decent sized group.

You could just compare an upscalled screenshot with a reference one by subtracting one picture from the other. IIRC you can do that with Photoshop. I'm pretty sure they used to do that to compare video card rendering on different quality settings back in the day.

 

[Saylick]

- I'd be fascinated by a double blind with a decent sized group.

I wonder how many people would actually prefer the upscaling solutions to native due to over sharpening or "motion blur aka ghosting" or softened textures etc.

I keep thinking about how a lot of default settings on TVs are oversaturated to hell because the average Joe likes the stimulating colorful pop over the signal accurate stuff that might be important to a professional.

Heh, this "phenomenon" occurs in many fields, like speakers for example. Beats by Dre are inferior headphones. Just ask any audio engineer and they'll agree, but Beats by Dre caters to a demographic that likes overemphasized bass.

 

[wanderica]

Heh, this "phenomenon" occurs in many fields, like speakers for example. Beats by Dre are inferior headphones. Just ask any audio engineer and they'll agree, but Beats by Dre caters to a demographic that likes overemphasized bass.

Very true. Those headphones even have weights in them to make them feel more sturdy, but they have lots of bass (if not a good frequency range), they look cool, and are marketed well. People gobble them up for twice the cost of a pair of Sennheisers that eat them for breakfast.

On topic:
I'm not sure I'd be able to tell the difference between FSR or DLSS myself, so I can't blame the music industry too much. I'm fairly confident I could pick out native though. I guess I just don't get it anymore. Back in my day, we loved upgrading so we could max settings at a higher framerate, run better AA, or (if you go back far enough) even play the latest game. The kid in me feels like all of this upscaling and fake frame stuff, while interesting tech, just feels a bit like cheating to me. The responsible adult, however, only wishes I could get it on my 1080 Ti so I didn't have to spend $1K for an upgrade. In short, I'll use it if I ever have to in a pinch, but at that point, I'd rather buy the game on console and be done with it.

 

[Dribble]

The kid in me feels like all of this upscaling and fake frame stuff, while interesting tech, just feels a bit like cheating to me.

This always makes me laugh - the renderering pipeline of every game you ever played is full of cheating - that's how they make it work. To describe existing cheating as fine and new ways of cheating as bad is silly. All that matters is the end result.

 

[BFG10K]

Can't deny it, FSR looks objectively worse in those tests, sometimes badly. But then DLSS has problems too, like ghosting which every version claims to have fixed. Better than native? Nah, neither of them are.

Right from the start I said I didn't like the direction this whole thing was going. Upscaling isn't a feature, it's a last resort for more performance.

 

[soresu]

Upscaling isn't a feature, it's a last resort for more performance.

Agree 100%.

To me it has more use as a VR booster, especially for standalone performance.

They have already employed every trick in the book to increase perf before dynamic foveated rendering became possible, so these would be another tool in the box for that.

 

[VirtualLarry]

This always makes me laugh - the renderering pipeline of every game you ever played is full of cheating - that's how they make it work. To describe existing cheating as fine and new ways of cheating as bad is silly. All that matters is the end result.

One of the "graphics gurus" on a project I was on in my 20s at a game studio, he had a function that took a 24-bit color pixel and sort of "intelligently decimated" it down to a 16-bit color value, that could then be used as an index into some 16-bit color LUTs, themselves 16-bit or 128KB in size. Something to do with lighting calcs, I think.

You would think that the "cheating" by throwing away that data might have caused visual issues, but it didn't.

I guess that was a "win" rather than dealing with each color channel and a seperate LUT for each.

 

[Kocicak]

I do not have much graphic card experience, but a truly native unsharpened original image from a gameplay looks very underwhelming to me, much worse than upscaled sharpened image.

 

[GodisanAtheist]

I do not have much graphic card experience, but a truly native unsharpened original image from a gameplay looks very underwhelming to me, much worse than upscaled sharpened image.

Playing Hogwarts Legacy and mother of god the native TAA is just hot garbage. Any kind of low light or dark area is ghosting galore.

I wish we had an option to turn off TAA in games that include it just to see what *true native* looks like.

Shame we lost the older brute force AA options that were computationally heavy but really made stuff look great.

 

[Heartbreaker]

Shame we lost the older brute force AA options that were computationally heavy but really made stuff look great.

Same here. I'm on the trailing edge for GPU so the only AA I have on old games is brute force, but there is no artifacts, and no loss of sharpness, just removal of edge jaggies.