[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vr / w / wg] [i / ic] [r9k / s4s / vip / qa] [cm / hm / lgbt / y] [3 / aco / adv / an / asp / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / qst / sci / soc / sp / tg / toy / trv / tv / vp / wsg / wsr / x] [Settings] [Search] [Home]
Board
Settings Home
/g/ - Technology


Thread archived.
You cannot reply anymore.



File: dlss is shit.jpg (1.16 MB, 1920x1080)
1.16 MB
1.16 MB JPG
The competition can't compete. 5700/XT image sharpening kills the Nvidia.
>>
>>71828594
>gaymes!!!!!
18+ website.
>>
>>71828594
I can't see the fucking difference
>>
>>71828594
DLSS is technology
>>
>>71828613
Still waiting for the 4K stream right now so had to use 1080p. But look at the round thing on the top of the tank on the right. DLSS turns it to mush.
>>
>>71828594
what does 78% sharpened meme? its the best looking one. is it native 4k sharpened with amd's new shit? or is it lower res than 4k upscaled and sharpened?
>>
>>71828594
Sharpened looks better than native 4k.
>>
File: 1.png (442 KB, 1223x320)
442 KB
442 KB PNG
>>71828613
I'm sorry anon but I think there might be cum in your eyes.
>>
>>71828637
Yup! Pretty neat. Also hardly any performance cost to it compared to Novideo.
>>
>>71828633
78% of 4k (around 1700p) with sharpening.
It actually looks better than native 4k.
Radeon Image Sharpening is crazy, it completely BTFOs DLSS.
>>
>>71828648
lol all that lost detail with DLSS. Nvidia are fucked.
>>
>>71828602
Electronic video games started out as "by adults for adults" kiddo
>>
>>71828633
>what does 78% sharpened meme? its the best looking one. is it native 4k sharpened with amd's new shit? or is it lower res than 4k upscaled and sharpened?

It's lower res upscaled with AMD and sharpened with AMD, gives superior performance and quality, also doesn't need proprietary DLSS bullshit
>>
>>71828672
>started out as
exactly.
times have changed, retard.
grow up.
>>
>>71828673
And best of all is developer agnostic, it's a driver level feature, so unlike DLSS doesn't require the devs to do anything to implement.
>>
>>71828688
It still requires an update to work with DX11 (which it doesn't) and per game implementation via profiles but there are still more games (DX9 , DX12 and Vulkan) that benefit from it than Nvidia's DLSS.
>>
>>71828688
>>71828673
thats amazing. can you do it for any resolution as well?
>>
>>71828594
>>71828648
Uhh but that's not screen resolution difference, the textures are clearly @ different resolution. Does DLSS change your texture resolution?
>>
DLSS is next gen technology provided by ML Tensor Cores. It's basically the power of Deep Learning brought to bear on video games. Every game gets a bespoke treatment using the power of the cloud for inference training. AMDs pajeet meme technology doesnt even begin to compare.
>>
>>71828706
Yes.
>>
>>71828706
Yes, it's a toggle in the Radeon control panel in the display options, which applies it to all games you run at whatever resolution you run them at. Currently there's no game-specific toggle, but I hope AMD implements that soon.
And as >>71828704 said it still requires the developers to a bit, but it's much easier than DLSS, and has a much lower performance cost (about 1fps on average on a 5700XT between sharpening on and off).
>>
>>71828730
and yet here is the evidence.
>>
>>71828724
>Uhh but that's not screen resolution difference, the textures are clearly @ different resolution. Does DLSS change your texture resolution?
Yes DLSS uses AI interpretation of image , so you get altered crap, Deep Learning is a meme.
>>
>>71828594
>that DLSS
Is this a joke?
Lmao
>>
>>71828594
>>71828663
Its not lost detail, this is a fake comparison. The textures on the tank arent loaded properly on the dlss image.

Proof: the guys shirt looks nearly the same across all three pictures. The only thing thats "blurry" is the tank.

DLSS causes some blur but that loss of detail isn't possible without lower resolution textures. Or in this case it seems like the gayme is loading the wrong lod textures just for the tank
>>
Fucking Jewtube slow updating the 4K stream.
>>
>>71828756
Cope
>>
>>71828747
been a joke for a while
upscaled 2k has always been a better choice. dlss is ass
>>
>>71828746
I get that but what we see here is clearly high-res vs low-res texture and not overall screen resolution difference. Like >>71828756
mentions it's almost like LOD fucked up and didn't load close-up textures. Is DLSS the reason? Fuck if I know, but that would mean Nvidia fucked up something in the drivers.
>>
>>71828756
thats not proof and thats not how it works. dlss constructs the image based on AI, similar to how you see random weird shit in deep fakes. it can and does mess up some pieces of the image because thats how AI works. nvidia made a big gamble and unless their AI can improve a lot, it looks like it wont pay off here.
>>
Stream here.
https://www.youtube.com/watch?v=7MLr1nijHIo
Wait for the 4K version because Youtube is being it's lousy self as per usual.
>>
>>71828648
DLSS on the right. Detail is GONE!
>>
File: IMG_9206.jpg (1008 KB, 3840x2160)
1008 KB
1008 KB JPG
>>71828746
DLSS doesnt alter game textures its a post process, maybe dont talk about things you don't understand


Here is a real dlss comparison, you can see that dlss is even blurrier than taa (kek) but the textures are visibly the same. DLSS isnt that great but OPs image is flawed
>>
here is another factor, it's all shit upscaling to 4k it's great for consoles and maybe some PC users
but it does interesting things to native 1440p too
from comparison I've seen it's basically like running AF x64, if it was possible, for textures and distant objects
>>
>>71828847
Why bother making the game detailed when novidia tech will make it blurry as fuck anyway. This is retarded.
>>
>>71828602
t. anime lover
>>
>>71828847
Look at the back of the headrests in the rear seats. Blurry shit with DLSS.
>>
Holy shit Njeetia PajeetLSS is completely destroyed.
>>
>>71828847
>look at this image from nvidia marketing materials, goy
>>
>>71828847
>>71828946
More importantly it's barely less overhead than native 4K. DLSS is suppose to be budget 4K, but even 1440p TAA looks better with less resource demand.
>>
>>71828756
>I thought this might be a texture issue for DLSS but I captured the footage twice and it looked the same both times
HUB

Dilate.
>>
>>71829068
As far as your memeing goes, it really doesn't seem like result of DLSS, at least not directly. I'd rather expect novideo drivers to fuck with textures when he turns on DLSS.
>>
>>71829090
Nvidia drivers are a joke in 2019.
>>
Damn this shit is juicy, might get the 5700 even though i have a 1080p monitor at least to when i ugprade
>>
>>71828594
>1920x1080 JPEG compressed screenshot of compressed YouTube video playing at 1080p
>supposed to compare image quality of 3840x2160 images
What's the point? You literally cannot even see 2560x1440 level of detail in the screenshot you posted, not to mention 3840x2160 because it's too low resolution. Beyond that you even JPEG'd it when the whole point would be to evaluate fine detail. This screenshot is entirely worthless, though DLSS looks so bad that I'm more inclined to believe that it isn't even using the same texture resolution. Even DLSS renders at higher resolution than your screenshot, so looking this bad at 1920x1080 doesn't really make sense.
>>
>>71829118
>>71828648
>>
>>71829118
As I said. Had to work with what I had and PNG was too big for 4chan. I could have cropped it but still. It's till fucking obvious which is better even with these constraints.
>>
>>71828940
Couldn't be more wrong.
Cope more, faggot.
>>
File: 1492890072159.png (235 KB, 380x380)
235 KB
235 KB PNG
>>71828602
>Vidya is for kids only
Your life must be boring, might as well kys now
>>
>>71829106
Theyve been a joke for years but blind fanboys dont care because they'll literally never use amd ever anyways. I sold my crossfire 7950s during the litecoin craze of 2014 and picked up a 780ti. Incredibly, I had way more issues with the 780ti than i ever did with TWO amd gpus in crossfire.

Ill buy an intel gpu before I ever give nvidia my money again
>>
>>71829168
>might as well kys now
What sophisticated, adult logic you possess.
>>
>>71829132
The fact that it's 1920x1080 is the biggest flaw. Again - your image literally cannot display all the detail 2560x1440 or 3840x2160 can. It's fucking retarded to compare a 2560x1440 upscale to 3840x2160 at 1920x1080. You cannot see the extra detail or any upscaling artifacts because your image is literally too small.

DLSS looks obviously shit, but having seen DLSS myself the image in the video is so obviously shit that the video is either faked (lowered texture resolution) or the game is suffering from some other issue, like the textures were still streaming and were not yet displayed at full resolution. I very much doubt DLSS alone is doing that, especially since the resolution DLSS is rendering at is also higher than this screenshot, as such there should be no significant difference in the detail reproduced in this image. DLSS adds some "oil painting effect" which is ugly and very obvious, but that's not what we're seeing here.
>>
File: literally.png (1.16 MB, 1845x384)
1.16 MB
1.16 MB PNG
>>71829250
>You cannot see the extra detail or any upscaling artifacts because your image is literally too small.
>the video is either faked
>I very much doubt DLSS alone is doing that
>there should be no significant difference in the detail reproduced in this image
This is your brain on NVIDIA.
>>
File: stfu.png (55 KB, 847x274)
55 KB
55 KB PNG
>>71829250
>>
nvidiots are the current cancer of pc hardware. at least intel fags were willing to admit when amd pulled ahead, usually.
>>
>>71829290
I'm talking about OP's 1920x1080 screenshot anon, not whatever you posted now.
>>
>>71829237
I knew you'd appreciate reasoning over as a simple "kys faggot"
>>
>>71829337
A JPG captured from the upscaled image on my 4K monitor. :))))
>>
>>71829219
Same, I switched to AMD from Nvidia in 2014 and it's been smooth sailing since, only times I had issues was the first two months of Vega drivers, but they got fixed quick enough.
But everywhere you go you read this garbage that AMD drivers suck, when the Radeon Settings menu is light years ahead of the pajeetware Nvidia control panel.
>>
>>71829219
>>71829106

I use both on my two computers (rx 580 and gtx970) and both have their share of problems. But if amd ironed out their incompatibilities with some older games (Most of the time you can get them to run properly but it's easier on the nvidia side of things) and improved their performance on opengl in winblows I would just go amd and never look back. I don't like nvidia as a corporation and even in their infinite budget they still find ways to fuck up their drivers and to milk the consumer with barely passable cards (rtx 2060 with 6gb of vram? No thanks the gtx 970 it's the last vram starved gpu I'll ever use)
>>
>>71829348
Good, this does not change anything about the inadequacy of OP's image. I wanted to see the difference between native 4K and AMD's new upscaler, which the image is essentially useless for because any extra detail or upscaling artifacts are obliterated by the downscale. I don't care about DLSS because it was obviously shit from the beginning and we knew that since it released.
>>
>>71829384
>but having seen DLSS myself the image in the video is so obviously shit that the video is either faked (lowered texture resolution) or the game is suffering from some other issue, like the textures were still streaming and were not yet displayed at full resolution. I very much doubt DLSS alone is doing that
>I don't care about DLSS because it was obviously shit from the beginning and we knew that since it released.
So this is the power of the Tensor cores...
>>
>>71829349
The only thing that makes me want an Nvidia card is better OpenGl and nvidiaProfileInspector, my RX 580 has been great so far.
>>
>>71829412
That was in response to this part of the post I quoted >>71829132
>It's till fucking obvious which is better even with these constraints.
>>
>>71829437
Having had both, yes Profile Inspector is a great feature that I wish AMD would implement. As for OpenGL, what even uses that anymore these days?
>>
>>71829500
>As for OpenGL, what even uses that anymore these days?
Not that anon, but emulators. Most notably Wii U and Switch emulation has always been crippled on AMD cards. Even though they are both supposed to get a Vulkan renderer who the fuck knows when.
>>
>>71828594
Why did they turn down the texture quality for Nvidia?
>>
>>71829646
Because Nvidia cheats in the driver.
>>
Wow all the butthurt nvidiots in this thread
>>
>>71828594
Nvidia has always produced worse image quality, so DLSS is just Nvidia doing what they do best.
>>
>>71829500
All 3d modeling software uses OpenGL.
>>
>>71829658
What do you mean?
>>
>>71829709
Nvidia is probably compressing their shit to get higher fps.
>>
>>71828594
Can I use this for VR to take advantage of the Index's 144hz or the Pimax's full resolution in VRChat?
>>
>>71828594
I generally call BS on all DLSS and AMD Sharpening threads but damn the one in the middle looks better, especially the reflections on the back on the tank.
>>
>>71829709
Novidya takes every shortcut to improve FPS, which is why people say AMD has better image quality
>>
File: a.png (3.11 MB, 1347x1539)
3.11 MB
3.11 MB PNG
>>71829716
Nah, something is fucked in their setup. DLSS looks bad but not that bad. Here are some crops I made myself, one is 4K native and the other is "4K" DLSS. DLSS is obviously blurry compared to native but it doesn't shit on textures as bad as what is shown in that video. If you look at the plants though, you'll see that DLSS is completely disastrous on those though.
>>
>>71828594
>youtube
can't see the difference since that shit is blurry regardless.
>>
File: b.png (3.59 MB, 1347x1539)
3.59 MB
3.59 MB PNG
>>71829716
>>71829806
And here's the other one. Open them in different tabs at 100% zoom to accurately judge the difference. It's very obvious which one is DLSS and which is native but it doesn't look like what this video shows.
>>
>>71828594
How much fps boost do you you get with Radeon Image Sharpening?
Retard.
>>
>>71829806
>>71829826
Top is DLSS. Why did they never fix it for BFV? In Metro Exodus they fixed it after complaints
>>
>>71828594
Worst bit about this is you have to wait for updates from Nvidia to get any better results while AMD's just does it on the gpu then and there for practically no performance loss lmao.
>>
>>71829806
>>71829826
DLSS makes the foliage look like its fucking melting lmfaooooooo
>>
>>71828594
Maybe now finally Nvidia will release the DLSS sdk so anyone can develop their own DLSS filter to apply on any game.
>>
>>71828825
So you can get +30% fps for almost free? That's amazing.
>>
>>71829854
Those images are old, to be fair. I took them 3-4 months ago. I don't know if they fixed it or not now but it was god-awful then. You can clearly see that what DLSS does to the textures does not look like what this video shows, though. It doesn't look like in this video though, so BF V updates either made DLSS even worse since I last tried it or there is something fucked up with their game, settings or testing process.

What is shown in the video does not look like DLSS, it looks like low-res textures. In fact if you compare the foliage in my DLSS shot to the foliage in the video, the "DLSS effect" is missing too. So yeah, something is either fucked or DLSS has been changed drastically in BF V.

>>71829892
This sort of artifact is quite typical of certain neural network upscaling algorithms. You can get things like waifu2x to do this too, but you can also get excellent results if they are used properly. NVIDIA has seriously fucked something up to get it this bad, it literally looks worse than any traditional upscaler.
>>
>>71829437
>>71829500
There was RadeonPro, but it ripd. I don't know if it will work on newer cards and drivers. Worked fine on 290 and Catalyst, might try it when I get home on my Vega
>>
>>71829826
Those pictures convinced me that the tank one is legit. Dont you see that DLSS (your first picture) does exactly the same thing on e.g. the stones as with the tank - it literally smears out all the small sharp details on the textures into a smudged blurry flat looking surface.
>>
>>71829826
It's actually quite noticeable. Everything looks like it's been thought the waifu2x upscaler.
>>
>>71829826
DLSS has ghosting while moving in game. Something you don't see in pictures.
>>
>>71829944
>This sort of artifact is quite typical of certain neural network upscaling algorithms.
Which is why it's a fucking meme in the first place, on one hand you have amds image sharpening that does it on the gpu, you know it can do it and you have it in your hands. On the other hand you have dlss that you have to wait for updates for and hope that nvidia actually does it and doesn't just one day give up on it. Almost like paying for an online game only to find out the company decided to close the servers down. On top of this it's shit.
>>
>>71829500
Neverwinter Nights 2, still gets stutters. As mediocre as it is I was playing Prey 2006 and it was stutrering a lot, I thought it was a OS compatibility problem but it uses OpenGl 2.0. It does have a Linux port so thag would probably run better but still.
>>
>>71829978
I'm not saying that DLSS is necessarily disabled, I'm saying that in addition to that something looks wrong with the textures. If you look at those white dots on the glove texture in my DLSS shot you will notice that they are more blurry, but they ARE still there. They aren't reduced into absolute nothingness. Same for the finer detail on the gun like the little nicks and wear on the metal, they are blurrier but still present. It's obviously worse, but it's not like they're completely gone like what is going on in this video.

>>71829985
Yeah, though waifu2x can produce excellent results with decent quality input images and if you don't run it with maximum denoise like a retard.

>>71829999
I never noticed ghosting with DLSS, but then again I never used it for anything more than fucking around in the practice area to take a couple of screenshots.

>>71830111
Neural network upscalers can and definitely do produce better results than traditional ones. Waifu2x when used properly and on the right kind of content is a good example, but mpv and madVR also have neural network upscalers which can look very, very nice. Something is just fundamentally fucked with DLSS itself, the core idea of using neural net upscalers is sound.

I personally find this whole upscaling shit borderline useless anyway. Just buy a monitor which your GPU can power properly and play native. You'll get the best image quality then. If you want sharpening or such, there's always Reshade which is customizable as fuck and can produce some amazing results with a couple of effects tuned properly (like adaptive sharpen and/or Clarity).
>>
>>71829806
It looks like one of those "deep learning" google paintings.
You know the one, with fractal dog faces.
what a joke from nvidia.
>>
>>71830198
>If you look at those white dots on the glove texture in my DLSS shot you will notice that they are more blurry, but they ARE still there. They aren't reduced into absolute nothingness. Same for the finer detail on the gun like the little nicks and wear on the metal, they are blurrier but still present. It's obviously worse, but it's not like they're completely gone like what is going on in this video.

There are a bunch of other places in your picture where the small details are completely gone though and reduced into nothingness, like the surface textures on the stones to the left of the gun. The only difference between your picture and the tank one is that on the latter, the effect is on an object that is close to the camera, very big, and in your face.
>>
>>71830254
Nah, I still don't think that's right. The missing detail on the stones is super fine, in the 4K image it basically turns into little dots of strands of color that are 1-2px wide. DLSS renders at a lower resolution (around 1440p I think) so a lot of detail on that level is going to be lost simply because there isn't enough resolving power at the lower resolution. It doesn't remove larger things like the bolts on top of the tank or such, those are large enough to be their own shapes, not like dots of dust on those stones.

I don't know, where did they take this video? Is it a standard benchmark I can run? I can try to replicate their results and see if it really does look as bad as the video.
>>
>>71829806
>>71829826
Looks like it was run through an hq2x filter.
>>
>>71830417
I remember this was after an update to DLSS in the game, before this update the result looked like more like "normal" blur and not this sort of smeary oil painting shit. I think they tried to make it sharper and fucked it up.
>>
>>71829854
Nobody complained about DLSS looking like shit, because nobody uses it.
>>
>>71830402
Clearly the DLSS algorithm doesnt work the way you think it works. There are many details that are much smaller than the stone surfaces but which DLSS has not smeared away completely into a featureless blur, for example the leaves. It seems DLSS just has a tendency to go haywire with the smudge tool when confronted with the wrong texture, and in OPs example it completely shat its pants with the tank.
>>
>>71829138
>t. resetera user and gayms hater
kys commie
>>
>>71829806
>>71829826
So, this is the power of Deep Learning, wow.
>>
>>71828737
Okay Lisa I'll buy your fucking card.
>>
File: 1548157693453.gif (1006 KB, 500x403)
1006 KB
1006 KB GIF
>>71830598
fair point
>>
>>71830599
The leaves are much smaller than the bolts on that tank, plus the vegetation is a large mass of uneven colors and a whole lot of high-frequency information/fine detail which DLSS destroys. Those bolts on the tank however are fairly large in comparison and quite clearly distinct on top of a pretty even surface without a lot of other detail around them, like the texture on the glove which does look worse but isn't basically removed. It doesn't actually look much like what DLSS does in my game.
>>
>>71830749
That's the beauty of deep learning. It doesnt behave consistently at all, it has to be trained how every single shape angle surface nut and bolt is supposed to look, individually.
>>
>>71830877
I don't know what they're doing with DLSS in particular, but the other neural network upscalers I played with do not behave this way (madVR NGU variants, NNEDI, waifu2x, some others I forgot the name of). They can all produce the effect you see in my DLSS image when they are not used properly, but I've never seen one as selective as DLSS supposedly is. I've never seen DLSS produce results like in that video either, even though the results it produces are obviously shit anyway so this is largely academic, it's not like DLSS looks good otherwise.
>>
Radeon sharpener + Navi11 APU.
This sounds interesting as fuck
>>
>>71829250
>>71829290
You can see in many videos people saying the same thing. DLSS is bad. If you don't believe just do it yourself, get an RTX card, take a screenshot at 4k them take a screenshot 4k DLSS.
>>
>>71828594
Will the user have access to the sharpening features in the driver?
>>
File: 1561761181908.png (102 KB, 757x676)
102 KB
102 KB PNG
>>71829237
>he posts unironically on 4chan
>>
Sharpened looks too artificial.
>>
>>71828594
But when will it be able to do integer scaling?
>>
>>71829806
DLSS makes it look like aliased mush.
>>
>>71828602
>LOOK AT ME GUISE!!
>I'M SO MATURE THAT I DON'T ENJOY THE THINGS YOU DO!!!
>AREN'T I SO FUCKING IMPRESSIVE YOU GUISE!?
please die
>>
File: Untitled.jpg (1.69 MB, 2560x1440)
1.69 MB
1.69 MB JPG
I can't cap at 4K as I have a 1440p monitor. So here is 4K screencap (played back in 4K) at 1440p with 99% JPG quality to fit the 2MB image limit.
>>
Fuck I'm gonna have to pick up an 5700xt
>>
>>71832155
The upscale looks better than the original. Good job AMD.
>>
>>71832245
In some cases its better. I really like the global setting, now they just need to implement game specific option. As well as DX11 support.
>>
>>71832155
all the details, the scratches on top of the tank, the back of the guys coat, the sandbags, the rivets and the helmet on the back all look better than native 4k lmao.
>>
>>71828648
>>71828594
Wow, that looks amazing. Going 4k 120Hz+ might not be completely retarded in the future.
>>
>>71832364
Anon you're clearly deluded, there's no way you can see a difference at such a low resolution. To appreciate the majesty of Nvidia's latest technology, pictures of which I have safely secured on my hard drive, you need only appreciate it for yourself. Of course, I cannot share these fabled pictures, for they are too large. Clearly though, there is no difference between the three pictures; one might even be tempted to say the light gleams in third, Nvidia picture, a little more realistically than the others. We might chalk that up to the tensor cores, or perhaps the superior frameworks upon which all Nvidia cards are built. Anyone that tries to sell you otherwise is merely that - a poor salesman. You can see for yourself, can't you, that at such low resolution that can be no palpable difference? I can see the glint of rationality in those keen eyes, do not be deceived.
>>
File: 1561502035491.png (222 KB, 617x466)
222 KB
222 KB PNG
Lads, we know how next gen consoles will have 4K 60fps natively.
>>
>>71832604
"Natively"
>>
>>71832662
Exactly.
>>
>>71832604
Well, PS4 alreadly uses the checkerboard rendering which works more like a Temporal upscaling which in most part is very competent. If they use that on the new consoles they easily can reach 60fps if not more than that.
>>
>>71828730
>bespoke
Stopped reading there.
>>
What games have Radeon Image Sharpening available?
>>
>>71832793
All DX9, 12 and Vulcan. If you have Navi that is.
DX11 support coming Soon™
>>
Because I'm a fucking brainlet I still don't fucking understand DLSS.
You have a 1440 display
You have DLSS
You now have a 1800 pixel display? But your monitor still only displays 1440?
What?
Should I be looking at 4k monitors or 1440? Or 4k scaled down to 1440 with DLSS to get near 4k?
>>
>>71832851
It takes a 1440p image and upscales it (poorly) to 2160p, to output it on a 2160p display. If you have a 1440p display you don't use DLSS.
>>
>>71832662
better than DLSSatively
>>
>>71832604
Or as always developers will prioritize adding more graphical bells and whistles and stick to 30fps instead.

>>71832703
Considering how checkerboarding still looks softer than native 2160p, pairing it with some non-shit sharpening should have good results.
>>
>>71833293
So what's the difference between DLSS and AMDs image sharpening?
>>
>>71833755
One isn't completely garbage
>>
>>71832586
See an optician
>>
>>71833293
You can use it at 1440p too I think? Not sure now but if so might be useful for upscaling 1800p (maybe not 1080p but who knows) too if the game is shit FPS wise.
>>
>>71828613
Of course you can't, OP is a retard that compares 4k renders on a 1080p screenshot of a 720p overcompressed video
>>
>>71828756
>that loss of detail isn't possible without lower resolution textures
Take a wild guess what DLSS is
>>
>>71833962
If you can't see the difference in the DLSS shot then I don't know what to tell you
>>
>>71833962
>And yet the middle screenshot still looks better.
Are you retarded?
>>
>>71828934
Yeah, that's retarded when the game shot for stock PS4 and runs in standard definition on it
>>
Can i use that sharpening to scale from whatever resolution i chose? Like can i scale 1080p to 1440p on 1440p screen for more fps? Would that look bad? Would that be just plain stupid?
>>
>>71833755
Well for starters DLSS is strictly an upscaling solution, image sharpening is not. You can sharpen an image and keep it at the same resolution, however sharpening an image after it's been upscaled with some other generic method (which is what's being done here) will make the end result look better.
DLSS is based on machine learning shit and relies on Nvidia's tensor cores to run, that means a) it needs to be trained and implemented on a per-game basis and b) it has some performance impact. Image sharpening is much simpler and basically just a post processing filter, meaning it has much better compatibility and only a minor/nonexistent performance hit. Granted Navi does seem to have some kind of hardware acceleration for it, but even without it should be faster.

>>71833952
Yeah, the source and target resolutions depend on the game. I just used those two for the sake of simplicity.
>>
>>71830948
900 upscaled to 1080 on a mobile apu maybe?
>>
>>71834061
In other words can i lower the res and sharpen to my native monitor res for better fps?
>>
>>71834069
Thanks for the answer.
>>
>>71834077
You can, but the lower the starting res, the shittier the final result. In order to sharpen detail that detail needs to exist in the first place.
>>
>>71834098
I mean will it look better with the sharpening?
>>
Full fat 4K lossless PNG for your perusal.
https://drive.google.com/file/d/1YMVFKsaiovVKXKOVImFEum7xtTf_n41r/view?usp=sharing
>>
The problem with DLSS is it does not know what to do with flat surfaces.
>>
>>71832851
It's the other way around. If it's labeled 4k, that's actually upscaled 1440p. Yes, you need 4k monitor for that, it's pointless at 1440p (well it's still possible to upscale 960p to 1440p but it will be ugly).
The difference with AMD, their sharpening filter has almost no performance hit (like DLSS) and no resolution shenanigans. You get the resolution you selected. But you can purposely lower rendering resolution (in some games, at least) and apply sharpening to boost performance while keeping the image quality (you may even get superior quality like demonstrated in the video).
DLSS at "4k" will perform like rendering 78% of 4k (I suppose that's the reason why the guy in the OP video used 78% to test AMD's sharpening) but the quality is significantly worse as it's in fact upsaceled from 67%, not 78%.
>>
>>71828613
This is a poor image extraction by op. It's most obvious on the flat surfaces of the tank. If you look at the original video there's a lot of detail (mostly tool-marks) in the native and basic sharpened footage but dlss just turns it into smudge.
>>
>>71834128
Yeah. But basically what I'm saying is that while doing 1800p-->2160p it looks almost as good as native 2160p, doing 1080p-->1440p should look better but not so much better it can pass for native 1440p.
>>
>>71834152
The asphalt looks oversharpened.
>>
>>71834220
5700XT has enough performance for running 1440p natively anyway, but could be interesting with APUs.

Anyone can confirm if it works with any game if you use GPU scaling. I mean, will the filter apply before or after scaling? Or is it mandatory to have a setting in the game for 3d resolution?
>>
File: 1511720487614.png (307 KB, 1300x2000)
307 KB
307 KB PNG
>>71829851
wew lad
>>
File: scalex.jpg (99 KB, 500x200)
99 KB
99 KB JPG
>>71833755
One is driven by virgin tech deep learning. (DLSS)
The other is driven by an undisclosed method (probably a chad heuristic).
It's just different implementations of the same idea. Take a low resolution image. Do image post processing. Get something that looks like the high resolution image.
This costs way less in gpu resources since you're doing something relatively simple compared to rendering a high resolution scene render.
Naturally it won't ever be as good as the full native render. That's just ground truth. But the existence of this shows there's at least some tangible benefits.

I'd be interested in seeing official support for other upscale algorithms. Like waifu2x or scale2x.
>>
>>71829851
This deserves a response.
Approximately as much. It's really cheap on both platforms and the resolution change is just as effective on both platforms.
>>
>>71834201
see
>>71834152
>>
>>71834267
Still looks better than DLSS
>>
>>71834406
>Still looks better 'overall' than DLSS
fixed
>>
File: DLSS ON vs OFF.jpg (908 KB, 1597x1043)
908 KB
908 KB JPG
DLSS is useless shit and sharpening is piss easy to implement
>>
>>71834473
Holy shit the left looks bad.
>>
>>71834348
It's funny that Nvidia is going this backasswards way to implement upscaling that ends up being worse just to justify them adding tensor cores for data center use.
>>
>>71828594
I can't believe Nvidia keeps getting away with scamming people like this.
>>
>>71834594
I don't think the tensor cores are relevant. I'd be extremely surprised if they did training on our GPUs. That's just absurdly dumb. They probably took the trained algorithm and put it in there.
They also promised they'd update (train, was their wording I believe) it though. Just to make it sound like what you're picturing. But it's still absurdly stupid.
>>
>>71834635
Think positive. Thanks to nvidia we got AMD to implement superior sharpening for everyone.
>>
>>71828847
>DLSS doesnt alter game textures its a post process
The person you're replying to never said that.
It's post process that alters the WHOLE SCREEN.
It has an "oil painting" effect in a lot of games, and in the best case is blurry like in that picture.

>>71829337
I can see the blurriness of 4K in OPs image as well.
Meanwhile the 1800p with CAS looks 95% as good as 4K native but at 30-35% better perf

>>71829806
reminds me of Deep Mind image generation lmao
>>71829854
Metro DLSS is definitely better, but still worse than simply upscaled and sharpened and DLSS runs worse.

Anyway it's up to the developer to implement and improve. No one wants to use that shit because even in the best case scenario it's garbage.

>>71834473
>sharpening is piss easy to implement
I hope more games implement it where the UI/HUD is still rendered in native 4K over top of the sharpened game render.
>>
>>71834684
Be it ours or theirs in the cloud, it's all to justify their investments in deep learning
>>
So does someone want to explain to me how CAS is any different from ReShade? Because it checks for contrast?
>>
>>71834779
It seems more then 30%, closer to 70% actually. So yea...pretty fucking good.
>>
>>71834832
Yes. It varies it's strength according to contrast or some higher level idea of surface boundaries. AMD says its based on contrast. The width of the kernel is probably pretty intelligent though.
>>
>>71834856
30-35% better perf when it's 1800p upscaled.
70% when it's 1440p upscaled.

>>71834832
Performance. ReShade has awful performance gimping.
And yes, it's a new algo that only sharpens areas with certain contrast. It tries to just sharpen edges and details. I've seen artifacts with like depth of field in some games, though, where it shouldn't sharpen stuff blurred by DoF.
It'll be a lot better when developers start implementing this directly into games. But it's already better than DLSS as is.
>>
>>71828602
>meanwhile esport players make more than you do in gaymes
>>71828613
where are you located ? so i can send you to the nearest eye doctor
>>
>>71828613
Same and I bet in motion there would be even less of a difference to see. Gameplay over graphics win again!
>>
>>71835368
Then you're really gonna love this.
Because you get more fps.
>>
Dlss is horrid I got more performance and better results using downsampling and reshade
>>
>>71828594
OK here is the real question: why they refused to implement a quality re-scaler before? its not that hard algo- or performance -wise and monitor scaling is usually garbage but even some expensive TVs come with a good saling/post-processing chip that can make typically poor tv signal or old dvd into a decent picture. What were GPU maker thinking?
What AMD has done with upscaler and CAS should've become a standard years ago
>>
>>71835611
Well they had to counter Nvidia with something and Navi probably had the architecture to support it with a performance hit.
>>
>>71835611
Pretty sure scaling only really works with higher resolutions so it wasn't really worth the investment until 4k became more common. You try to scale with lower res and the artifacts are more noticeable.
>>
>>71835611
Performance at 4K is still shit and will remain shit for the foreseeable future, so hardware makers suddenly realized making use of upscaling and other shortcuts might be useful. That's not just AMD and Nvidia but also console makers, Sony arguably started this trend with the push for checkerboard rendering on PS4 Pro.
>>
>>71835611
There is this integer upscaling that doesn't blur the image and tends to look good in high dot pitch screens. I think it alreadly exists in Linux and people are asking for AMD and NVidia to implement it on driver level.
AMD did this pool asking what people wanted to see in future drivers and the Integer upscaling was the most voted feature.
>>
doesn't DLSS also learn from customers GPU data? so Nvidia are basically using their customers GPU to improve their own shitty features. kek
>>
File: images.jpg (40 KB, 739x415)
40 KB
40 KB JPG
>DLSS
Anons please explain DLSS to me. Does DLSS mean that I can make non gaming software like waifu 2x better, or is it strictly for vidya?
>>
>>71835985
No. The DL part is deep learning and it means Nvidia has to "train" things to work with it.
>>
>>71836054
So DLSS is using the extra tensor, and cuda cores in rtx series?
>>
>>71835985
>layering upscale filters
It's just not effective. You want to have the filter as well adapted to the content you're processing as possible. So for waifu2x that's probably among the worst pairings. You want to feed it stuff that looks like anime. Sharpening isn't conducive to that. You could probably clean up the image if it has some sort of artifact and get better results. But that's not what dlss or AMDs upscale is for.
>can I use dlss for offline processing tasks
I don't know. But worst case you could write the most basic renderer in the world that just draws a quad with your image and then enable dlss on that and take a screenshot. I don't see why that wouldn't work.
If you're lazy you could probably just find a full-screen program that allows you to show an image.
>explain it for me
It just takes the image the game rendered and scales it up while applying the filter. It's like applying waifu2x in real time to the frame buffer. Nothing more complicated.
>>
>>71836099
See
>>71834684
>>
>>71834779
>I hope more games implement it where the UI/HUD is still rendered in native 4K over top of the sharpened game render.
This can be achieved with integer scaling
>>
>>71835784
>>71836702
Fuck you and fuck your integer scaling
>>
>>71835773
Works on my 2080ti
>>
>>71837062
What do you have against beauty, anon?
>>
>>71828594
>>71828613
>1080p screenshot comparing 4k content
There's a good reason why the difference isn't visible
>>
>>71837355
see
>>71834152
>>
>>71828594
>sharpening
AVSforum is laughing
>>
>>71837086
>$1200 for a vidya gpu

Good goy
>>
>>71837468
It's OK. They're spending money on acoustic cable rocks.
>>
Have a ReShade injector of it.
https://www.reddit.com/r/Amd/comments/cc0575/i_ported_fidelityfx_cas_to_reshade_so_anyone_can/
>>
>>71828594
completely pointless to do the comparison in a youtube video.
>>
>>71837776
How so? All those images left to right are compressed equally so you can see the same equally. It's not like the DLSS on the right was compressed more than the two images next to it.
>>
>>71837793
first, the source is compressed (recording the video)
second, the recorded video is compressed, again (in adobe premier, when cutting the video)
third, the video is compressed AGAIN (by youtube, the worst quality loss from all the steps)
when comparing 4K upscaling to NATIVE 4K, one should use png's and write a blog post instead of these disgusting yt videos.
>>
>>71834984
>meanwhile you're still a virgin
>>
>>71829304
sounds like a game issue then
>>
Imagine not being able to enjoy what will probably be the greatest game of our lifetime and appreciate each and every last bit of the world CDPR created through amazing lighting all because you were a poorfag nigger fighting "Ngreedia" for another corporation that doesnt give a flying fuck about you. Inb4 5700XT gets like 5 fps in the game as well.
>>
>>71839150
I'd rather not play the game at 1080p and 30fps
>>
>>71828838
>>71828724
>>71828648
>>71828613
The detail isn't gone, it was never there.
Algorithms can't invent information out of nothing.
DLSS pulls information out of a library of historical data, lossily compressed into a statistical model.
Either there simply isn't enough data to replace the lost information, or the initial resolution was too low to make the leap to 4K, hence the lack of detail.
>>
>>71837973
How is it not compressed equally?
Or do you not seeing the difference?
>>
>>71837973
Sure it's not the best way to compare a videogame but to say it's pointless is absurd. Some of us have good eye you know.
>>
File: aKCAW5x.jpg (27 KB, 600x315)
27 KB
27 KB JPG
>>71839176
>Its afraid
pottery
>>
>people already started post purchase rationalisation of Navi cards they haven’t even bought yet
>sour grapes over Nvidia for good measure

Yikes
>>
>>71840870
>DLSS
>Sour grapes
im glad my gpu doesnt support a feature that makes all my games look like blurry shit
>>
File: DtTnfFH2Hjc.webm (424 KB, 640x360)
424 KB
424 KB WEBM
DLSS was shat on since its release.
Did they at least try to improve it?
>>
>>71841004
it's supposed to be constantly improving by the nature of deep learning but I don't know what patches have been released
>>
>>71828594
nigger like you don't even know what DLSS IS. what it actually is is something that allows for ray-tracing without the huge computational toll due to the games being pre-anaylzed by nvidia's super computers and the output of that analysis being coupled with each new driver release.
>>
>>71828594
how in the actual FUCK does image sharpening have ANYTHING to do with ray-tracing? as far as i'm concerned the words radeon image sharpening and dlss shouldn't even be uttered in the same breath.
>>
>>71841106
>>71841136
this is /g/ now, full of retards who can't even read marketing pamphlets
>>
>>71829806
>DLSS is obviously blurry compared to native but it doesn't shit on textures
WAT! are you blind? those textures looks shitty as fuck, just look at the wall right next to the camera, _all_ detail is gone and whats left is just the 'artifacting' from the upscaler. even that squiggly black line below the S in shoot moved from one place onto another, how does dlsshit not change textures? its literally a screen space filter, you idiot!
>>
>>71836099
no, the training is done by nvidia on nvidia headquarters and the resulting STATIC network is shipped with the drivers, so you only get "better" results after updating drivers. there is no training done on your computer, when jensen said it learns and adapts as you go, he was not speaking the truth, and nvidiots gobbled it up.
>>
Never heard of dlss.
>>
>>71834072
On medium settings with this sharpener. I can see it looking decent enough, and having pretty solid fps.
>>
>>71841106
>>71841136
They're the same. CAS allows you to save ~30% fps, exactly what ray tracing needs.
The difference is, CAS actually works unlike DLSS.
>>
>>71843772
only problem is that radeon DOESNT HAVE RAY TRACING
>>
>>71843804
Neither does nvidia. Unless you have two 2080ti.
>>
File: 1471937810179.png (846 KB, 990x518)
846 KB
846 KB PNG
>>71828602
>>
>>71828594
>mfw 1080 owner happy to see Nvidia finally getting BTFO by competition
who's with me bros?
>>
>>71839176
We're talking about a Navi feature that lets it out better than 4K quality despite only rendering in 2K (where AMD BTFOs Nvidia, so novideo couldn't compete even if DLSS wasn't complete shit)
>>
>>71844789
970 cuckold here, I'm happy AMD is competing again. honestly considering the 5700XT when they go down in price.
>>
>>71843804
Who cares? DLSS is doing the same function, saving frames while looking crisp. Ray tracimg will only come to AMD when the right time comes.
>>
>>71832245
I mean, if you want to sharpen your games you can always use Reshade and you get way more control over what you use and how you tune it. NVIDIA has also had their Freestyle shit for a long time now, that can also sharpen your games, but I don't know if that works on all games or not.
>>
>>71841691
>how does dlsshit not change textures? its literally a screen space filter, you idiot!
That's literally how you retard, it works on the final image, not on the fucking textures.
>>
File: image.jpg (690 KB, 1280x960)
690 KB
690 KB JPG
>>71828602
fpbp

>>71828672
>>71828940
>>71829168
>>71832117
>>71832545
>>71843856
You HAVE to go back 2 reddit. Fucking off yourselves
>>
>>71839150
screenspace reflections has been used in games since 1998

cope more 24fps R(e)T(ard)X cuck
>>
>>71828602
based
>>
File: 1562854720642.jpg (62 KB, 1200x220)
62 KB
62 KB JPG
>>71828594
lmao wtf nvidia is like PS2 level
>>
File: 1559655854160.jpg (26 KB, 251x242)
26 KB
26 KB JPG
>>71829806
>>71829826
>nvvidia is literally making an oil painting out of your game
how do nvidiots come with this lmao holy shit
>>
File: 1562244892149.jpg (41 KB, 512x564)
41 KB
41 KB JPG
>>71829806
>its a photoshop filter
>>
>>71828594
A more detailed article on this.
https://www.techspot.com/article/1873-radeon-image-sharpening-vs-nvidia-dlss/
>>
>>71829806
DLSS looks like a weird impressionist painting
>>
File: the scream.jpg (220 KB, 800x993)
220 KB
220 KB JPG
>>71849115
this
>>
File: wew.png (2.15 MB, 2560x1440)
2.15 MB
2.15 MB PNG
Fake news.
>>
>>71848699
kek, AMD looks better than native
>>
>>71839150
I'm glad they delayed the launch another (almost) year. 8 years in the making - this better be damn good. I don't care what it takes to run the thing.

But if they release some kind of Grand Cry Witcher nonsense that is just above mediocre in all aspects, and lacks the ability to truly play the game with freedom, I'm out. I don't want to be hampered by 30 hours of story just to unlock everything. In the "real cyberpunk world" if you have cash you get what you want - nothing artificial but the implants, and you can become anything along the way. That's the whole spirit of the genre.

And nethacking better be fucking amazing with great VR implementation or f.u.c.k.i.n.g d.r.o.p.p.e.d
>>
>>71829646
Not just texture quality, the geometry is also different

Seems to be a marketing bullshot paid for by AMD
>>
File: 3467346.png (179 KB, 369x365)
179 KB
179 KB PNG
>>71849202
>tripfag
>uses picture from a video published by nvidia themselves on a benchmark not an actual game
>>
>>71850180
Port Royal is basically a tech demo that showcases Nvidia tech. The benchmark was also ran by a 3rd party review site, its more indicative of the capability of DLSS.
>>
>>71837760
No comments? You can use this sharpening on any Radeon GPU. I tried it on Apex Legends and it works pretty good.
>>
>>71828847
Christ it's MLAA all over again
>>
>>71850212
>more indicative of the capability of dlss
Doubt it, I'd imagine it's easier for the machine learning to make a benchmark look easier cus it's running the exact same way over and over whereas a game doesn't. Also I'm sure Nvidia let their machine learning spend quite some time on this since they were going to publish it, where as they could throw a game into their machine learning for a day and be like "yea thats good enough ship it in the next patch".
>>
>>71850258
Also you can use it on a game by game basis.
>>
>>71850291
Also imagine having to actually wait for patches from nvidia to get this whereas on the amd card you just flick the switch lmao.
>>
>>71844789
same desu
i cant see why i would need another card right now
other than it getting raped in Metro Exodus, had to go to low/medium to even get 100+ fps in some areas



Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.