That’s……. actually a good thing. You Xbox guys should try it sometime.
Yeah this. I'm happy for the console bros, this kind of tech is the shit when it comes to IQ + Framerate.
Iron sharpens Iron...
That's not compiled data, I don't think you and most of the people claiming it is custom FSR4 understand the technology.
He didn't say anything about the AI block, he only said they added custom hardware for machine learning and an AI Library call PSSR.
PSSR is most likely Playstation's version of FSR4 for their API, in which they can update at their own time instead of waiting for AMD to do any updates.
If that custom hardware for machine learning was their own solution, Mark Cerny would of when in to more details about it.
How quickly the PS5 Pro was made pretty much confirm they didn't create their own solution for AI upscaling. They're most likely using RDNA4 AI Accelerators. Which AMD themselves states is similar to the Tensor Cores.
PSSR blurry? hell no
It even beats THE native 4K
But this is deep learning, a training run per firmware release for PlayStation and per version release for FSR4, unless CoPilot is wrong and PSSR is a recurrent neutral network. But either way, PlayStation's cadence to update PSSR will almost be on a per game release basis, where as FSR4 will likely have the same update cadence as DLSS with just 3.7versions across 7years because card manufacturers are selling hardware with architecture changes yearly or longer, whereas PlayStation is selling games and want PSSR looking great for each A-AAA game as it releases so are driving the initiative in software.
I mean he barely even mentioned the custom hw in ps4pro (there was a bunch of stuff that had nothing to do with cbr or AMD gpus at the time).
For (apparently) most posters here it is simply inconceivable that Sony would come with something on their own. They have such a 'high esteem' for this Japanese hardware company. You would think their engineers are inept or something. It has to be AMD or nothing.
I talking about the solution having a game_id as an input to PSSR.
No, I'm meaning the overall solution is trained in a game agnostic way and used with or without tweaks - unless Naughty dog loons overroad every node with tweaking-, then that gets feed into the agnostic solution on the next major update to keep the net node files sizes down.
PSSR looks less blurrier, mainly noticeable on the foliage. Reflections in the helmet look more detailed as well.No. It's showing less aliasing because of how it handles edges. You can still have an aliased image with softer output and the images are also very zoomed in. So the difference in clarity won't be as big when playing.
The two technologies handle edges slightly different and just in this game so far. We'd need to see more examples. In-game, that would hardly be noticeable.
This is one of those common-sense things: Sony and AMD are working together. Sony doesn't just use AMD tech; they share tech, too. I don't know how or what some of you think ML technologies entails, but I am beginning to feel there is a fundamental misunderstanding of what it really is. That what has got people on here thinking that Sony and AMD would have two different performing ML technologies, working with the same data sets and on the same hardware architecture... in the exact same way, but somehow would yield different results.
Think about this for a second... not that I even agree what with you are saying.. but lets pull this thread.
True... but that's not relevant here.
I am not sure to follow you logic. So your reason is that Sony has a tendency to settle with mediocrity contrary to AMD?
Alan Wake 2 preview shows the results of this as the performance mode is rendering under 900p but using PSSR to get to 4k.
PSSR's base solution will be the same, but PlayStation as a pinnacle tier AAA developer of games for image quality on fixed hardware PlayStation will know only too well that you could have effectively identical input scenarios for an ML AI reconstruction algorithm and in one game the image inference should be X when in all other games it should be Y, and only a superset of nodes with game_id for the outlier will catch all situations, and as a platform with a cut of all software, PlayStation can't have a scenario where the outlier has no solution. So out of necessity PSSR will need to handle the situations.
Nvidia and Intel don't understand the problem at the same level because they don't make money from a specific hardware platform and the games software that run on that closed platform hardware.
They wouldn't be doing anything to PSSR's AI, they'd have controls to alter the handling of each image filter and would tweak as desired, that would generate overloaded functions for the game agnostic inference functions that PSSR uses by default. They wouldn't even need to know the algorithms or technical workings of PSSR, just the adjustments and the SDK calls to integrate it into their game engine would be my guess.
Which is what I've been saying.
Humble?
How would making their own bespoke ML hardware upset AMD?
The second Post above yours pretty much confirm what I've been saying all along.
Exactly like that, humble and well spoken making some sense
.
DLSS, XeSS, PSSR and FSR4... all do the exact same thing. Using the exact same data sets and going about reconstructing an image in the exact same way. Only difference between them is how far along they all are. The only difference I see between PSSR and FSR4, I suspect will be that PSSR will work (or works) better with DRS + reconstruction as opposed to FSR which would be better suited for fixed input resolutions.
I won't call it mediocrity... just that Sony doesn't care about some things. As has been made very evident even with the base PS5. Look how long it took them to patch in native 1440p support. And we still don't have LFC for VRR outputs. Or their reluctance to support DV.
They did the same with RT.
When I said PlayStation games I was referring to everything released the entire catalogue, which was mentioned by someone official talking about PSSR.
How would you know how PlayStation approaches the training of their AI to make such a statement?
LFC as catch all frame-rates is the mistake of a company not caring. you shouldn't be catching a dropping 60fps at 2/3rds of its starting point. But, AMD don't make games, so aren't expected to be an authority on "should we offer it like that?", rather than a hardware vendor only attitude of "can we offer that and market it as a feature?" Which is what AMD do with providing LFC
It certainly is going by how PSSR does a version of the same database object recognition and inference as the original Sony X9's X1 chip according to the SE info on how they used PSSR to improve specific objects..
Do you know what an AI data set is? You do realize at the end of the day all this shit is just basic instructions right? What we are calling ML hardware, are just matrix instruction accelerators. Which means they are ALL handling matrix instructions. ML reconstruction is an algorithm-driven matrix operation. Nvidia calls those accelerators tensor cores, AMD calls them AI cores or whatever.
Forget AMD, even Xbox has it. And for a console where you have the possibility of games running below your set VRR window, not having LFC is the kinda gross oversight that sony is actually known for. As I said, they can be very weird when it comes to some things. And as much as I like pissing on MS and Xbox, there are a number of very basic forward-thinking things they do and that sony practically has to be dragged kicking and screaming to implement.
Sigh...
So you don't know how the regression is done to produce the equations for inference then? and are just talking about the hardware it computes on being similar, and presumably have no idea how you can come at this problem in different ways? .sigh
Couldn't have been, or you would've understood my point from the beginning, which you have not.
PSSR, like DLSS and XeSS before it, are all examples of a convolution neural network. And by regression rate you referring to gradient descent? They all do it. Its how the AI model is trained. And not even gonna go into inference since that's simply how the model fills in the gaps to create the higher rez image... which again, they all do. Some better than others no doubt, but like with DLSS, it will always get better.
The friends we made along the way.
I'm seeing sony gaf fighting each other so it seems more like friendships breaking apart
Or people just having a disagreement despite favoring the same piece of plastic.
A good reason to end friendships i'd say
For the time being DLSS surely has a stability advantage, and looks like it could be producing sharper results. But when it comes to the latter it's hard to tell for sure because sharpness setting in Ratchet don't match, and AW2 pictures are sourced differently.
Do you even pay attention?
