• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

The PS5 Pro Proves AI Upscaling Is the Future

winjer

Gold Member
It wasn't that deep and Playstation absolutely has a ton of visual data to train the AI.

Don't worry I wasn't insulting Nvidia....

Visual data is aplenty in every gaming platform.
The first part is to develop a proper temporal upscaler, this is done in an heuristic form, with code written by humans.
The AI part is having a convolution network to run on top of that temporal upscaler, to detect and eliminate artifacts, like disocclusion, ghosting, pixel shimmer, etc.

And what do you mean it wasn't that deep?
Do you know how many parameters (weights) Nvidia used with DLSS? Or Sony used with PSSR?
Or do you think that Deep actually refers to some measurement of vertical distance?
 
Last edited:
We need more articles like this
It started with checkerboard rendering. Now they’re unto AI upscaling. I don’t see the issue.
Somehow this is lost in the conversation, Sony started this path of hardware upscaling before everyone AI was just a natural progression and recently a Sony executive mentioned PSSR took a lot of time and that's why they couldn't have it ready at PS5's launch. Thinking about it, I'm not sure if RDNA 4 will have any type of AI upscaling and if that's the case then it makes total sense why Microsoft didn't plan to make a mid-gen refresh because they're also developing a solution AMD seems to be way behind but we'll see.
 

lh032

I cry about Xbox and hate PlayStation.
Why? There was checkerboard render on the PS4,with Sony have an AI base super resolution. The only weird thing is Guerilla Games coming up with another super resolution type of AA.
checkerboard? that was ass.Did you see the Pssr preview? looks incredible, i preorderd ps5 pro and cant wait to see the result
 
Last edited:

Imtjnotu

Member
So glad sony brought AI upscaling to the main stream world... The innovation of the industry never stops

Donald Trump Applause GIF
 

DeepEnigma

Gold Member
Something unclear about A.I.? If you wanna just talk about upscalers, they're much older than checkerboard rendering.
People are pointing out temporal in 2018. R&C used temporal prior to that on the 4Pro.

While this article is clickbait, it did mention the key advantage that they were able to have a faster to market solution in such a short time, is due to their closed platforms. Their TVs have been running AI scaling chips for years now as well.

It mentioned it's a good thing for all consumers, but only we gamers and nerds can arguing about who and what was first.
 
Last edited:

James Sawyer Ford

Gold Member
Absolutely, but this also reminds me that we have now no less than 3 AI-based upscaling solutions and 2 hardware-agnostic ones. I can't imagine devs being happy with this many. Microsoft might also develop their own eventually or go with whatever solution AMD comes up with in the future. That's going to be yet another one added onto the pile.

I doubt it’s difficult to implement with the SDK, so likely not a major issue
 

winjer

Gold Member
PC master retards was saying those PS4 Pro games were fake 4k (CB) in 2016. The same people embraced DLSS some years after.

That is not even correct to say that CB is an upscaler. It's just interpolation.
DLSS, especially DLSS2 is a very different tech from interpolating pixels with every other frame.
 

Gaiff

SBI’s Resident Gaslighter
PC master retards was saying those PS4 Pro games were fake 4k (CB) in 2016. The same people embraced DLSS some years after.
They also shat on DLSS1. Almost as if they didn't care for low-quality solutions.
People are pointing out temporal in 2018. R&C used temporal prior to that on the 4Pro.
Cool, it's not an AI upscaler.
While this article is clickbait, it did mention the key advantage that they were able to have a faster to market solution in such a short time, is due to their closed platforms. Their TVs have been running AI scaling chips for years now as well.

It mentioned it's a good thing for all consumers, but only we gamers and nerds can arguing about who and what was first.
The solution that proved that AI upscalers were the future was DLSS, not checkerboard rendering that was seemingly difficult to implement and only saw the light of day in a select titles. What are we doing here exactly? Trying to discredit DLSS or trying to credit Sony with coming up with AI upscalers for games because they had it in their TVs? The AI upscaling revolution in games really took off with DLSS2. DLSS1 was trash and laughed at.
 

XXL

Member
And what do you mean it wasn't that deep?
Do you know how many parameters (weights) Nvidia used with DLSS? Or Sony used with PSSR?
Or do you think that Deep actually refers to some measurement of vertical distance?
You don't know what "it wasn't that deep" means?
 

Elios83

Member
It's obvious that nVidia was first but they also beta tested the whole thing with DLSS1 which didn't work really well.
IGN is completely dumb but I guess the point is that once technologies are adopted within consoles it means they have become fully viable and mainstream hence the "future".
 
Last edited:

vkbest

Member
That is not even correct to say that CB is an upscaler. It's just interpolation.
DLSS, especially DLSS2 is a very different tech from interpolating pixels with every other frame.
CB is not a upscaler, it’s a reconstruction technique. You are the perfect example why I call them PC Master retards.
 
What a realization!
And it's only several years too late.
UE4 has Temporal upscaling since 2018. DLSS2 since 2020.
And then we got more upscalers like FSR2, XeSS, TSR, etc...

No Shit Sherlock GIF by Team Coco
Seriously. My first thought was PC space is doing this for years and spawned several well established standards in the meantime. All of a sudden when Sony introduces the technology, IGN loses it's mind :messenger_tears_of_joy:
 

winjer

Gold Member
CB is not a upscaler, it’s a reconstruction technique. You are the perfect example why I call them PC Master retards.

It's neither. It's interpolation. Something that has been done for decades. Even in videogames.
It's just interlacing one set of pixels in one frame, then another set of pixels in another frame. It's not that difficult to understand.
You can try to insult as much as you want, but the fact remains you don't understand the tech.
 
Last edited:

DeepEnigma

Gold Member
They also shat on DLSS1. Almost as if they didn't care for low-quality solutions.

Cool, it's not an AI upscaler.

The solution that proved that AI upscalers were the future was DLSS, not checkerboard rendering that was seemingly difficult to implement and only saw the light of day in a select titles. What are we doing here exactly? Trying to discredit DLSS or trying to credit Sony with coming up with AI upscalers for games because they had it in their TVs? The AI upscaling revolution in games really took off with DLSS2. DLSS1 was trash and laughed at.
No shit, I've been a huge proponent of DLSS for years and years now. Been always saying that native high resolutions are a waste of resources with all the refined scaling tech that's coming out.

Once again, refer to my posts. Clickbait? Yes. Only us nerds will argue about what is "first" when all companies have contributed in some way to the scaling revolution.

With that said, consoles having it will drive it more to the mainstream normies as well, which we will all benefit from in the long run. Which was the crux of the article outside of the panty arguing off the clickbait title.
 
Last edited:

ap_puff

Member
That is not even correct to say that CB is an upscaler. It's just interpolation.
DLSS, especially DLSS2 is a very different tech from interpolating pixels with every other frame.
disagree, generating pixels that weren't there is upscaling regardless of method. hell a bilinear or even nearest neighbor pixel upscaler is upscaling. It's just the output quality is bad. Interpolating half the frame's pixels is not the same as purely generated frames such as by DLSS3. Clearest indication is that DLSS3 generated frames add input latency because the frame is something that is external to the game engine.
 

winjer

Gold Member
disagree, generating pixels that weren't there is upscaling regardless of method. hell a bilinear or even nearest neighbor pixel upscaler is upscaling. It's just the output quality is bad. Interpolating half the frame's pixels is not the same as purely generated frames such as by DLSS3. Clearest indication is that DLSS3 generated frames add input latency because the frame is something that is external to the game engine.

But there are no pixels being generated with CB.
The raster is rendering one group of pixels in one frame, then another set on another frame.
 
Top Bottom