• 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.

Digital Foundry-PS5 Pro: How Did Sony Beat Microsoft In Machine Learning For Consoles?

Fabieter

Member
Not sure how that has anything to do with them being biased towards MS though, they could talk about it and simply be negative? Pretty sure that would get them even more clicks.

Meanwhile they are being really optimistic about a technology that no one has tested in the real world and compared with its competitors (DLSS XESS etc).

You have to read between the lines to see that they are "biased". But I would say biased is the wrong word. They just don't like sony and PlayStation its not like they prefer other platforms over it well besides alex at least.
 

richdot

Neo Member
Xbox doesn't have much good news now days does it? I do like Xbox as a brand had every generation of console likewise playstation but right now playstation gets my limited gaming time I can't remember last time I switched on my series x :(
 

ZehDon

Gold Member
If I recall correctly, there was an internal push at Microsoft for AI-based upscaling tech for this console generation, but it ultimately wasn't picked up. And that to me is the problem with Xbox: it's forward thinking about the monetisation, not the technology. They're running out of time to correct that.
 

Loxus

Member
The only thing that stopped Sony, was AMD didn't have anything AI/ML solution to put in their consoles at the time.

Now AMD has multiple solutions.
LGa1csv.jpg
 
Last edited:
Sony has being doing ML upscaling in some PSP games running on PS4, and textures ML upscaling on God of War Ragnarok. They have litteraly kickstarted the reconstruction way of doing on consoles in 2016 (hardware + software APIs) and also doing frame generation on all their PSVR games.

They are not new at this game and they are basically the only ones competing against Nvidia and Intel. What prevented MS for starting their own AI upscaling venture since all those years? What are they waiting for to innovate?
 

StereoVsn

Member
Nah ms got some wins. Mostly All on the service side seeing how they are a service first company now:

*Game refunds are easier.
*The way they do digital ownership vs game subs is better on xbox. Seriously I cant buy a game outright if its on psplus extra if I already redeemed it. And yet sometimes I can ?? Clusterfuck of a infrastructure that psn.
* Cloud saves not tied to online membership. Cloud saves that magically appear on pc if it has the Play Anywhere tag which of course ms let it fall by the wayside.

All these Sony can actually copy and improve.
Got to agree, the inability to purchase games you “redeemed” through the sub service is ridiculous. And locking saves behind a paywall at the same time disabling capability to backup saves to USB (without backing up entire game) is greedy and obnoxious.

Finally, Sony’s position on refunds is absolutely anti consumer and should be illegal.
 
Last edited:

yurinka

Member
Pretty sure sony uses Microsoft tech for cloud gaming.

No, Sony uses their own patented cloud gaming stuff and MS pays Sony to use these patents.

Sony's cloud gaming servers use modified PS3, PS4 and PS5 hardware, that obviously MS can't build.

I hate to break it to you but PS Studios aren't exactly setting the world on fire currently. However with Xbox bringing games to PS you should hope Xbox lifts their game, I think they just might 🤞
Sony console exclusives, including PS Studios releases, are in the top selling rankings and win GOTY awards, unlike the MS console exclusive games.

Even in Microsoft's Windows, last week Sony had 2 games in the top 3 best selling paid games:
image.png


I mean, right now when including F2P they also have Destiny 2 in the top 15, above CoD.
 
Last edited:

SHA

Member
So Ellie will watch what she says and no longer talk shit cause she will learn from her mistakes? I don't understand machine Learning at this point in video games other than NPCs getting a little smarter, just a little.
 

SHA

Member
news_20240108_MS_pc_(1).jpg


When heavens divide
I will see
The choices within my hands
How can we ever protect
And fight with our tiny souls
Let me shine
Like the sun
Through the doubts and fear
Do you feel the storm
Approach as the end draws near
When heavens divide
Time will come
To softly lay me down
Then I can see a face
That I long to see
And for you,
Only you
I would give anything
Leaving a trace
For love
To find a way
When heavens divide
I will dive into
The fire
Spilling the blood of
My desire
The very last time
My name
Scorched into
The sky
When heavens divide
I will see
The choices within my hands
How can we ever protect
And fight with our tiny souls
Let me shine
Like the sun
Through the doubts and fear
Do you feel the storm
Approach as the end draws near
When heavens divide
Time will come
To softly lay me down
Then I can see a face
That I long to see
And for you,
Only you
I would give anything
Leaving a trace
For love
To find a way
When heavens divide
 
Last edited:

FireFly

Member
Sony has being doing ML upscaling in some PSP games running on PS4, and textures ML upscaling on God of War Ragnarok. They have litteraly kickstarted the reconstruction way of doing on consoles in 2016 (hardware + software APIs) and also doing frame generation on all their PSVR games.

They are not new at this game and they are basically the only ones competing against Nvidia and Intel. What prevented MS for starting their own AI upscaling venture since all those years? What are they waiting for to innovate?
The XSX only has 49 TOPS and 24 TF FP16, making it half a 2060. I doubt they can get close to the 2ms upscaling time that you need to make using it worthwhile. And they've (seemingly) decided not to launch a mid-gen refresh. But the leaked FTC slides claimed they were working on AI based Super Resolution for their next console and planning on looking into Frame Generation.
 
Last edited:

Mr.Phoenix

Member
What? Where they used it?

Microsoft was talking about using ai part in Xbox series for ML and image reconstruction but they didn't do anything with it (so far).
Well... not ML reconstruction per say, but Sony was first to push the technologies of which we can say DLSS is its final form. With the PS4pro. In truth,, it even goes further back than that, with Killzone shadowfall MP mode. But come the PS4pro in 2016, Sony had built-in modified GPU hardware that "accelerated"? CBR. And CBR was the first mainstream image reconstructor (not upscaler) that used temporal data from previous frames and motion vectors.

Some people and DF... talk about AI reconstruction as if it is, or was the first and only form of temporal data image reconstruction tech out there. It isn't, and it wasn't. It was just the best way to ultimately do something that sony pretty much started.
 
Last edited:

Radical_3d

Member
Right now I guess MS is either betting for full Blizzard Activision exclusives for next console including all past, present and future catalog, or just stop bleeding and go full third party.
Yeah, so they are not getting me in neither of those scenarios XD
 
The XSX only has 49 TOPS and 24 TF FP16, making it half a 2060. I doubt they can get close to the 2ms upscaling time that you need to make using it worthwhile. And they've (seemingly) decided not to launch a mid-gen refresh. But the leaked FTC slides claimed they were working on AI based Super Resolution for their next console and planning on looking into Frame Generation.
The problem was always RD, will and software, not hw. Sony has being doing ML textures upscaling with only 20TF of FP16 on PS5. If they wanted MS could have come up with something on XSX. But they produced nothing for 4 years. They are all talk and PR BS.
 

midnightAI

Member
news_20240108_MS_pc_(1).jpg


When heavens divide
I will see
The choices within my hands
How can we ever protect
And fight with our tiny souls
Let me shine
Like the sun
Through the doubts and fear
Do you feel the storm
Approach as the end draws near
When heavens divide
Time will come
To softly lay me down
Then I can see a face
That I long to see
And for you,
Only you
I would give anything
Leaving a trace
For love
To find a way
When heavens divide
I will dive into
The fire
Spilling the blood of
My desire
The very last time
My name
Scorched into
The sky
When heavens divide
I will see
The choices within my hands
How can we ever protect
And fight with our tiny souls
Let me shine
Like the sun
Through the doubts and fear
Do you feel the storm
Approach as the end draws near
When heavens divide
Time will come
To softly lay me down
Then I can see a face
That I long to see
And for you,
Only you
I would give anything
Leaving a trace
For love
To find a way
When heavens divide
Why did I read that in my head in a french accent like a perfume advert?
 

FireFly

Member
The problem was always RD, will and software, not hw. Sony has being doing ML textures upscaling with only 20TF of FP16 on PS5. If they wanted MS could have come up with something on XSX. But they produced nothing for 4 years. They are all talk and PR BS.
The system used in Ragnarok takes 9.5 ms (!) to upscale a single texture to 4K. It would be worthless as a general purpose Super Resolution feature, and is instead used intelligently when there are free resources.

MS could be doing more to use A.I within games but there is no silver bullet when you are upscaling the entire frame. If there were, Sony would not have needed to increase their A.I capabilities by an order of magnitude with the PS5 Pro.
 
The system used in Ragnarok takes 9.5 ms (!) to upscale a single texture to 4K. It would be worthless as a general purpose Super Resolution feature, and is instead used intelligently when there are free resources.

MS could be doing more to use A.I within games but there is no silver bullet when you are upscaling the entire frame. If there were, Sony would not have needed to increase their A.I capabilities by an order of magnitude with the PS5 Pro.
It's not the point. They are doing real-time texture AI uspcaling on PS5 so they are gaining experience that they will use on PS5 Pro (with developers likely giving feedback to Cerny). They have also years of doing reconstruction with CBR, that will also help them with PSSR.

At the beginning, DLSS was not good and the hardware barely powerful enough. But Nvidia learned and improved it along the years.
 
Last edited:

jroc74

Phone reception is more important to me than human rights
ML temporal upscalers and DVD upscaling done by your Bravia TV are two completely seperate things though. No Bravia TV had dedicated silicon that was utilizing ML in realtime, heck TAA didn't even exist until 10 years ago. I'm sure there is some overlap in research and skillset but let's not pretend that Sony were pioneers in this field for gaming. Even Nvidia completely screwed the pooch with the initial push into ML based upscaling.

Cool, sorta like how many assumed MS because of Azure would just take over and dominate cloud gaming.

Ironic...isnt it?

Cant have it both ways.

I hate to break it to you but PS Studios aren't exactly setting the world on fire currently. However with Xbox bringing games to PS you should hope Xbox lifts their game, I think they just might 🤞
main-qimg-a2c1d137c7dd6de3ccd9eab59a81bb07
 

Bojji

Member
Listen to what I'm actually saying: AI/ML based upscaling has been in Sony's R&D pipeline for years now. So for DF to frame PS5 Pro implementation of ML upscaling as "Sony eating Microsoft's lunch" is another case of them proving their own ignorance and refusal to do their homework. Again, these guys are supposed to be "tech experts" but the laziness and incessant cowtailing towards Microsoft is becoming less amusing and more unbearable for someone like me who would like to understand more about gaming tech at a deeper level beyond resolution and frame rate.


G8YdduR.jpg


hNqwD5t.jpg

Never heard of it when I heard about MS talking about ML in image reconstruction. Interesting stuff but Sony didn't announce anything to gaming audience.

Maybe DF guys didn't know?

Sonys tv division? ML is present in any modern TV.

Yes but at the same time it's not the same thing as ML in games rendering.

Well... not ML reconstruction per say, but Sony was first to push the technologies of which we can say DLSS is its final form. With the PS4pro. In truth,, it even goes further back than that, with Killzone shadowfall MP mode. But come the PS4pro in 2016, Sony had built-in modified GPU hardware that "accelerated"? CBR. And CBR was the first mainstream image reconstructor (not upscaler) that used temporal data from previous frames and motion vectors.

Some people and DF... talk about AI reconstruction as if it is, or was the first and only form of temporal data image reconstruction tech out there. It isn't, and it wasn't. It was just the best way to ultimately do something that sony pretty much started.

Since PS4 with checkerboard rendering.



I agree that CR was first wide spread image reconstruction technique, grandfather to FSR, TSR and TAAU but not exactly to DLSS without ML element.

Insomniac was doing image reconstruction even in PS3 days (using MSAA), before CB there also was Killzone SF MP (like you mentioned) and Quantum Break.
 

Reckheim

Member
So Ellie will watch what she says and no longer talk shit cause she will learn from her mistakes? I don't understand machine Learning at this point in video games other than NPCs getting a little smarter, just a little.
its meant to increase performance.

As an example DLSS 3.5 uses AI to literally double the frames in games; doubling the performance of games.

Having a good AI upscaling technique will improve performance beyond what the spec is on paper. So you don't have to have top tier hardware to run your games.
 

kensama

Member
I agree that CR was first wide spread image reconstruction technique, grandfather to FSR, TSR and TAAU but not exactly to DLSS without ML element.

Insomniac was doing image reconstruction even in PS3 days (using MSAA), before CB there also was Killzone SF MP (like you mentioned) and Quantum Break.

Yes but ML was not there until PS5 and Series X/S launch in consoles area.

DLSS first version exist only since 2019 an,d was not with AI or ML:
Deep learning super sampling (DLSS) is a family of real-time deep learning image enhancement and upscaling technologies developed by Nvidia that are exclusive to its RTX line of graphics cards,[1] and available in a number of video games. The goal of these technologies is to allow the majority of the graphics pipeline to run at a lower resolution for increased performance, and then infer a higher resolution image from this that approximates the same level of detail as if the image had been rendered at this higher resolution. This allows for higher graphical settings and/or frame rates for a given output resolution, depending on user preference.[2]
https://en.wikipedia.org/wiki/Deep_learning_super_sampling
As of September 2022, the 1st and 2nd generation of DLSS is available on all RTX branded cards from Nvidia in supported titles, while the 3rd generation unveiled at Nvidia's GTC 2022 event is exclusive to Ada Lovelace generation RTX 40 series graphics cards.[3] Nvidia has also introduced Deep learning dynamic super resolution (DLDSR), a related and opposite technology where the graphics are rendered at a higher resolution, then downsampled to the native display resolution using an AI-assisted downsampling algorithm to achieve higher image quality than rendering at native resolution.[4]

Sony has participated greatly in solutions in gaming under open GL:
The architecture of the PS3 ---> https://en.wikipedia.org/wiki/Vulkan. If you see why Vulkan was created and you know how Playstation's first party studios work (and with the existence of the ICE team), it should indicate you why.

PS4 and PS4 Pro served as the basis for the RDNA
FSR I think owes its existence to the checkerboard ----> https://en.wikipedia.org/wiki/GPUOpen#FidelityFX_Super_Resolution

Since AMD equiped the console (Xbox and PS) of their APU Sony was directly implicated with AMD in work of this one and give AMD right to implement what Sony wanted for their APU --->
 

Zathalus

Member
Cool, sorta like how many assumed MS because of Azure would just take over and dominate cloud gaming.

Ironic...isnt it?

Cant have it both ways.
What had this got to do with what I said? I didn't mention Azure or Cloud gaming at all. It has nothing to do with this topic.
 
Last edited:

Gaiff

SBI’s Resident Gaslighter
Yes but ML was not there until PS5 and Series X/S launch in consoles area.

DLSS first version exist only since 2019 an,d was not with AI or ML:
Deep learning super sampling (DLSS) is a family of real-time deep learning image enhancement and upscaling technologies developed by Nvidia that are exclusive to its RTX line of graphics cards,[1] and available in a number of video games. The goal of these technologies is to allow the majority of the graphics pipeline to run at a lower resolution for increased performance, and then infer a higher resolution image from this that approximates the same level of detail as if the image had been rendered at this higher resolution. This allows for higher graphical settings and/or frame rates for a given output resolution, depending on user preference.[2]
https://en.wikipedia.org/wiki/Deep_learning_super_sampling
As of September 2022, the 1st and 2nd generation of DLSS is available on all RTX branded cards from Nvidia in supported titles, while the 3rd generation unveiled at Nvidia's GTC 2022 event is exclusive to Ada Lovelace generation RTX 40 series graphics cards.[3] Nvidia has also introduced Deep learning dynamic super resolution (DLDSR), a related and opposite technology where the graphics are rendered at a higher resolution, then downsampled to the native display resolution using an AI-assisted downsampling algorithm to achieve higher image quality than rendering at native resolution.[4]

Sony has participated greatly in solutions in gaming under open GL:
The architecture of the PS3 ---> https://en.wikipedia.org/wiki/Vulkan. If you see why Vulkan was created and you know how Playstation's first party studios work (and with the existence of the ICE team), it should indicate you why.

PS4 and PS4 Pro served as the basis for the RDNA
FSR I think owes its existence to the checkerboard ----> https://en.wikipedia.org/wiki/GPUOpen#FidelityFX_Super_Resolution

Since AMD equiped the console (Xbox and PS) of their APU Sony was directly implicated with AMD in work of this one and give AMD right to implement what Sony wanted for their APU --->

Huh, yes, DLSS used ML out of the gate with its crappy 1.0 version that was trained on a per-game basis using 16K image samples to train their AI. They completely ditched this model with 2.0 and are now using a generalized network (NGX) and motion vectors.
 

kensama

Member
Huh, yes, DLSS used ML out of the gate with its crappy 1.0 version that was trained on a per-game basis using 16K image samples to train their AI. They completely ditched this model with 2.0 and are now using a generalized network (NGX) and motion vectors.

Ok
That's not what said Wiki because of this:

As of September 2022, the 1st and 2nd generation of DLSS is available on all RTX branded cards from Nvidia in supported titles, while the 3rd generation unveiled at Nvidia's GTC 2022 event is exclusive to Ada Lovelace generation RTX 40 series graphics cards.[3] Nvidia has also introduced Deep learning dynamic super resolution (DLDSR), a related and opposite technology where the graphics are rendered at a higher resolution, then downsampled to the native display resolution using an AI-assisted downsampling algorithm to achieve higher image quality than rendering at native resolution.[4]
 
Last edited:
Top Bottom