Support NeoGAF

[Topher]

---

-Variety of diagrams showing flowcharts of what this upscaler does, example images
-First flowchart sounds a lot like checkerboarding
Uses one of two upscaling solutions:
1) Spacial upscaling
2) Machine learning
-Technique might be leveraging PS4 Pro/PS5 hardware for pixel ID
-The machine learning process is more selective to reduce strain on hardware
-Making machine learning practical for RDNA 3 without needing something like a tensor core
-Patent sounds like a "grab bag of different buzzwords"
-2x upscaling vs 4x with DLSS
-Alex wondered if this patent were for some non-gaming use -Edit

Summarized best I could but a lot of this is above my head so sorry for any mistakes.

 

[Fake]

AMD is too slow doing this kinda of stuff. IMO was a mistake from Sony for waiting AMD to move foward upscaling.

Sony did a good job with PS4pro cb, so I would expect the same.

 

[lh032]

im wondering....if Sony did invest some sort of custom crazy dlss/fsr tech on playstation, hope 60fps will be the standard mode.

 

[Imtjnotu]

im wondering....if Sony did invest some sort of custom crazy dlss/fsr tech on playstation, hope 60fps will be the standard mode.

No but they have been doing alot of ML stuff in their other sectors.

About time playstation catches up

 

[winjer]

This is the Patent: https://www.freepatentsonline.com/WO2021214485A1.pdf

This doesn't seem like a form of upscaling technology for videogames.
More like a tech to recover parts of corrupted images in video.
Basically, the video is divided into tiles. If one is damaged, maybe an error during streaming or a damaged disk, an AI will try to infer the missing hole, by interpolating previous and forward frames.

 

[Gameplay Gods Bless]

---

-Variety of diagrams showing flowcharts of what this upscaler does, example images
-First flowchart sounds a lot like checkerboarding
Uses one of two upscaling solutions:
1) Spacial upscaling
2) Machine learning
-Technique might be leveraging PS4 Pro/PS5 hardware for pixel ID
-The machine learning process is more selective to reduce strain on hardware
-Making machine learning practical for RDNA 3 without needing something like a tensor core
-Patent sounds like a "grab bag of different buzzwords"
-2x upscaling vs 4x with DLSS

Summarized best I could but a lot of this is above my head so sorry for any mistakes.

What? I thought the PS5 Pros RDNA3.5 architecture came with its own tensor core like, AI dedicated hardware. Wtf dude it's 2024 & AMD and Sony still can't catch up to Nvidia from 6 years ago. Can you believe Nvidia launched all this the same year God of War 2018 and Spiderman launched? Yet it's 2024 and consoles are still playing catch up, unbelievable.

 

[NEbeast]

What? I thought the PS5 Pros RDNA3.5 architecture came with its own tensor core like, AI dedicated hardware. Wtf dude it's 2024 & AMD and Sony still can't catch up to Nvidia from 6 years ago. Can you believe Nvidia launched all this the same year God of War 2018 and Spiderman launched? Yet it's 2024 and consoles are still playing catch up, unbelievable.

Ofc they're playing catch up, they're consoles. If you want the best tech then buy a PC.

 

[Sleepwalker]

It doesn't need to be better or equal to DLSS (altho that would be nice of course) it just needs to be better than the shitty FSR2 we have.

 

[Thief1987]

This is the Patent: https://www.freepatentsonline.com/WO2021214485A1.pdf

This doesn't seem like a form of upscaling technology for videogames.
More like a tech to recover parts of corrupted images in video.
Basically, the video is divided into tiles. If one is damaged, maybe an error during streaming or a damaged disk, an AI will try to infer the missing hole, by interpolating previous and forward frames.

They are really talking about this patent?
Because how is this even relevant to an upscale tech

Digital images can contain regions of missing or corrupted image data. Missing or corrupted regions are referred to in the art as "holes". Holes are normally undesirable, and methods of inferring what information is missing or corrupted are employed to fill the holes. Filling holes in images is also referred to as image completion or inpainting.

A variety of processes exist for filling holes in images. Machine learning inference techniques, which rely on trained processes, can fill holes in images with high-quality results. However, machine learning techniques are performance intensive, requiring powerful computer hardware and a large amount of time.

Holes in images arise in image-based rendering systems. For example, where there are two or more images representing perspectives of the same environment, there may be no image data corresponding to an intermediate perspective that a user would like to see. Alternatively, there may be some image data missing from one of the perspectives. Machine learning processes may be used to infer the intermediate perspective and to infer the missing image data. Executing machine learning processes to obtain missing data is computationally costly and time consuming.

Hitting a new low, though it's df so nothing surprising.

 

[Mahavastu]

They are really talking about this patent?
Because how is this even relevant to an upscale tech

Yes, at 1:52 in the video you can see a screenshot of the patent and the numbers seem to match

 

[winjer]

They are really talking about this patent?
Because how is this even relevant to an upscale tech

As far as I can tell, it's the same patent.
But what is described there is not an upscaler. But some sort of error correction.

 

[Zathalus]

Patent reads like it might see used in VR.

 

[S0ULZB0URNE]

What? I thought the PS5 Pros RDNA3.5 architecture came with its own tensor core like, AI dedicated hardware. Wtf dude it's 2024 & AMD and Sony still can't catch up to Nvidia from 6 years ago. Can you believe Nvidia launched all this the same year God of War 2018 and Spiderman launched? Yet it's 2024 and consoles are still playing catch up, unbelievable.

Nvidia are Legends

 

[Raonak]

Sony are damn good at building on top of amd tech to add cool features.
Built in checkerboard system in the PS4 pro.
The SSD IO subsystems and the tempest engine in the PS5.

Cannot wait to see what wizardry mark cerny had in store for PS6.

They're the best at squeezing the most performance per dollar.

 

[Mahavastu]

-2x upscaling vs 4x with DLSS

2x upscaling was the old checkerboard rendering, the new solution should be better then that.

This doesn't seem like a form of upscaling technology for videogames.
More like a tech to recover parts of corrupted images in video.

Ok, from what I understand (not sure if I know what I talk about ) it is a repair method to fix problems from streaming.
It is intended for VR headsets which get the image for both screens streamed somehow, e.g. like the current PSVR2 via USB cable or via cloud service or whatever.
It first checks for streaming related problems (holes) and if it find some it tries to fix them as good as possible.
It also stores areas without problems so it can be used to fix later images.

yeah, seems no upscaling...

 

[Puscifer]

AMD is too slow doing this kinda of stuff. IMO was a mistake from Sony for waiting AMD to move foward upscaling.

Sony did a good job with PS4pro cb, so I would expect the same.

I gotta say CB is amazing and I'm sad it's not used anymore, probably the only real competitor to DLSS I've seen.

 

[Crayon]

This is the Patent: https://www.freepatentsonline.com/WO2021214485A1.pdf

This doesn't seem like a form of upscaling technology for videogames.
More like a tech to recover parts of corrupted images in video.
Basically, the video is divided into tiles. If one is damaged, maybe an error during streaming or a damaged disk, an AI will try to infer the missing hole, by interpolating previous and forward frames.

Wth.

Playing Tekken. I'll be back when you guys figure out what's going on here lol.

 

[Buggy Loop]

This is the Patent: https://www.freepatentsonline.com/WO2021214485A1.pdf

This doesn't seem like a form of upscaling technology for videogames.
More like a tech to recover parts of corrupted images in video.
Basically, the video is divided into tiles. If one is damaged, maybe an error during streaming or a damaged disk, an AI will try to infer the missing hole, by interpolating previous and forward frames.

For better cloud streaming quality?

 

[Rocco Schiavone]

Sony are damn good at building on top of amd tech to add cool features.
Built in checkerboard system in the PS4 pro.
The SSD IO subsystems and the tempest engine in the PS5.

Cannot wait to see what wizardry mark cerny had in store for PS6.

They're the best at squeezing the most performance per dollar.

Yeah Wizardry.

Such a Wizardry that all games in performance mode are a blurry mess.

 

[ChiefDada]

God of War Ragnarok ML TAAU remains the best 1440p game I've ever seen on console. If PS5 Pro can go even further to ensure all games even from less capable developers look just as clean or better then it will be a success.

 

[Fake]

I gotta say CB is amazing and I'm sad it's not used anymore, probably the only real competitor to DLSS I've seen.

CB, for the time and money, was the best upscaler around. I don't really understand why Sony don't build a upcaling team based on CB for PS5.

PS5 is like at the mercy of AMD to get what they can do. Even RT perfomance is poor.

 

[winjer]

For better cloud streaming quality?

Yes. Could be one of the use cases.

 

[Thief1987]

Patent reads like it might see used in VR.

Yeah, and seeing that the patent is basically filled by the people from now closed London Studio this is most likely the intended use of it.

 

[Fafalada]

They are really talking about this patent?
Because how is this even relevant to an upscale tech

It's describing image reconstruction - which is what every upscaler also does. It 'is' a functional superset (patent is generalized to reconstructing 'any' type of gaps - not just the ordered grid you get from eg. 1080p->2160p upscales) but to say the two aren't in closely-related solution space would be disingenuous.

However...
1) The patent is concerned with a particular reconstruction needs - and indeed - not really resolution upscales (and definitely not AA, which the popular upscalers also perform). What I read in the patent - this was targeted at VR (specifically wireless/remote rendered). Basically what a few other posters mentioned above.
2) DF for some reason looked at it as 'the' resolution upscaling patent - which it's clearly not. And it's very clearly not a derivation of CBR either.

With that in mind - regarding things DF stated:

-2x upscaling vs 4x with DLSS

Nothing in the patent is suggesting reconstruction of 2x (or CB or even rectangular grid). It actually implies reconstructing arbitrary sized blocks (which is relevant for the VR usecase more so than upscaling, but it applies either way).
Also - CBR was never limited to 2x - it was just the ratio chosen at the time that gave best IQ/perf tradeoff. In fact - 4x CBR is basically identical to standard 4x upscaler just with added benefit of ID/depth buffer at full resolution (so slightly more expensive to render).

-First flowchart sounds a lot like checkerboarding

So would any other upscaler if you broke it down into a similar flow-chart. Basically this is 'reader sees what they want to see' thing.

-Patent sounds like a "grab bag of different buzzwords"

That's just standard patent language. I thought DF would have been familiar with that by now.

 

[CityHunter1981]

Sony are real wizards when it comes to upscaling. No television at the moment produces a better upscsling result than Sony TVs. Even my 10 year old Sony 4K tv has a great upscaler.

 

[Topher]

This is the Patent: https://www.freepatentsonline.com/WO2021214485A1.pdf

This doesn't seem like a form of upscaling technology for videogames.
More like a tech to recover parts of corrupted images in video.
Basically, the video is divided into tiles. If one is damaged, maybe an error during streaming or a damaged disk, an AI will try to infer the missing hole, by interpolating previous and forward frames.

I meant to note that Alex did wonder if this were not for use in video games. Added that to the OP

 

[shamoomoo]

God of War Ragnarok ML TAAU remains the best 1440p game I've ever seen on console. If PS5 Pro can go even further to ensure all games even from less capable developers look just as clean or better then it will be a success.

I wish there was more info on SSM technique outside of the IGN interview.

 

[Mahavastu]

Sony are real wizards when it comes to upscaling. No television at the moment produces a better upscsling result than Sony TVs. Even my 10 year old Sony 4K tv has a great upscaler.

The PS3 from 2006 already had very high quality upscaling from DVD to full HD.

 

[Gameplay Gods Bless]

Ofc they're playing catch up, they're consoles. If you want the best tech then buy a PC.

What're you talking about did you not read my comment? This hw was available in 2018, i.e the year God of War, Spiderman and RDR2 launched i.e when the PS4 and Xbox One ruled the day. This isn't some crazy tech that Nvidia launched in 2023, hell even Intel has this tech and they created on their first attempt at a GPU competitor. Hell even Apple has this now on their hardware despite Apple being way behind Nvidia because guess what graphics tech isn't something Apple is innately experienced on.


The 2060 is ancient man yet that card has all of this. At least when the PS4 and PS4 Pro launched they had all the latest graphics bells and whistles. This was not the case for the PS5 and according to this somehow not even the PS5 Pro is gonna beat the PS5 in this department.

 

[solidus12]

---

why does Alex Bugaga look like he's about to cry?