• 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.

Will ML be a differentiator between the consoles?

AgentP

Thinks mods influence posters politics. Promoted to QAnon Editor.
The context is shader cores hardware MORE suitable to machine learning

Again, your one source is a CEO quote who has no technical background. Explain how a shader core is used for ML and how they are different.
 

geordiemp

Member
I don't think it's reaching when it's literally on the fucking hot chips presentation deck Geordie. It doesn't matter anyway. It just annoys me sometimes that you try to speak with such authority on matters you're not an expert in.

Where on the hotchips does it mention you need dedicated ML to do AMD new open source upscaling method ?

Where in the AMD 6800 XT reveal does it go on about ML capability and the new upscaling - no where !

Ps5 can do ML functions on its hardware, is it designed for deep learning high performance like tensor cores, hell no.

The whole thread is a massive assumption that AMD is going to release an open source upscaling and it will need lots of dedicated ML that only XSX has and ps5 has not to fuel console wars - that is a reach and you know it -

Its a FUD thread about ps5 lacking in something with no sources before reveal - what a surprise. Posters see it for what it is.

Heck even 6800 has not revealed dedicated ML yet.

I am not an expert in this, not even close I have no backgrund in ML, but I can read simple english.
 
Last edited:

ZywyPL

Banned
I don't thinks so, it'll probably end just like Rapid Packed Match in PS4 Pro, which is basically the same tech, a.k.a. unused. Even if MS figures out how to actually implement the tech suitable for games, it'll have to sacrifice CUs to do so, and I don't think they'll be willing to do it. Especially there's already a difference in the raw GPU power, plus VRS that by itself can boost the performance by 5-20%, so I don't really think they'll ever need the DirectML
 

Ps5ProFoSho

Member
My point is that neither console has some super secret amazing technological advantage from the other. Xbox has better GPU, PS5 has better SSD and I/O. That's IT. It will be up to developers to take the most advantage of these machines and both companies made an amazing console in different ways for them to achieve an amazing next gen.
 
Last edited:
Ps5 is not Microsofts definition of RDNA2 as it has its own apis

Ps5 and XSX are not AMDs true definition of RDNA2 as no infinity cache, but they are not giving full details for now.

At the moment RDAN2 is a marketing thing until AMD release what a RDNA2 CU really looks like from this slide. Then you will know.

XSX does not have passive fine grain clock gating or any other of the other 2 points below fine glock gating for that matter.

Let the RDNA2 blurb and mis understandings continue for now, it will make the dissapointment even more fun when AMD explains this slide properly :messenger_beaming:.


mdsmnrB.png
RDNA 2 is what AMD say it is. It's their product.
If they are happy to say the XSX is full RDNA 2, then it is.
 
Last edited:

IntentionalPun

Ask me about my wife's perfect butthole
Nobody knows... all we know is MS is advertising DirectML but we don't really know if/when any games will use it on XSX and Sony has patents that mention ML that indicate they are working on their own software.

Hardware wise we have 2 GPUs that are configured quite differently, and one may have a hardware advantage.

We won't know for some time whether either system really uses ML as a differentiator. At this point it's all talk from MS and only guesses from Sony based on patents.
 

SatansReverence

Hipster Princess
Ps5 is not Microsofts definition of RDNA2 as it has its own apis

Ps5 and XSX are not AMDs true definition of RDNA2 as no infinity cache, but they are not giving full details for now.

At the moment RDAN2 is a marketing thing until AMD release what a RDNA2 CU really looks like from this slide. Then you will know.

XSX does not have passive fine grain clock gating or any other of the other 2 points below fine glock gating for that matter.

Let the RDNA2 blurb and mis understandings continue for now, it will make the dissapointment even more fun when AMD explains this slide properly :messenger_beaming:.


mdsmnrB.png
XSX is not RDNA2 CU, if you understand the 3 points, the first one is dead easy...

fR97g0o.png
So you clearly dont undertand the slide then which tells you what RDNA2 CU is. Do you know what pervasive means ?

What your babbling on about L3 which is sort of fabric slower dense cache but not quite, is not what the slide says and is a different subject.

This is about RDNA2 COMPUTE UNITS....Read again


y4ZXZpe.png
All with fine gated freqency control and less than 10 CU per shader array.



So the AMD slide is technobable now ? You should inform AMD of your expert analysis

W1BLdIj.png


Also aligns with the RDNA2 leakers from a while ago



Jesus fucking autism christ.

Has quite literally zero evidence to support theory.

Spams image that doesn't say what you claim to browbeat and shut down any conversation.

Quotes random twitter bullshit.

But is protected autist on neogaf.

***

As for the topic, only one of the consoles has direct confirmation of hardware support of ML algorithms/acceleration.

The other has at least 2 first hand reports from developers/engineers for the console saying it does not support it while not having any direct claims from sony saying it supports it.

After seeing DLSS 2.0 in action, there is clear evidence that it could indeed make a difference. Whether or not it will be utilised is another story.
 
Last edited:

AgentP

Thinks mods influence posters politics. Promoted to QAnon Editor.
After seeing DLSS 2.0 in action, there is clear evidence that it could indeed make a difference. Whether or not it will be utilised is another story.

DLSS is a Nvidia IP. AMD does not have the equivalent and if they do come up with some upscaling tech, I doubt it will be as efficient or as good. So with AMD silicon, let's not expect miracles, everything is a trade off. If you spend cycles on a ML scaling solution, you will have less to spend in other areas.
 

AgentP

Thinks mods influence posters politics. Promoted to QAnon Editor.
DLSS 2.0

Has what relevance? The technology is proven. It requires both local hardware and server computing to work. MS has both.

Right, so when your Xbox gets a Nvidia based GPU then we can celebrate. Until then the Xbox and PS5 are based on AMD and AMD has not shown any ML scaling solution and likely it will not be as good/fast when it appears. That's assuming Nvidia doesn't have a patent around the secret sauce, which they likely do. Nvidia is also on the 2nd gen of this tech and has designed their latest GPU on it (in part).

So keep hope alive, but in the end I doubt you will see anything competitive with DLSS.
 

SatansReverence

Hipster Princess
Right, so when your Xbox gets a Nvidia based GPU then we can celebrate. Until then the Xbox and PS5 are based on AMD and AMD has not shown any ML scaling solution and likely it will not be as good/fast when it appears. That's assuming Nvidia doesn't have a patent around the secret sauce, which they likely do. Nvidia is also on the 2nd gen of this tech and has designed their latest GPU on it (in part).

You are really struggling to detach the idea of DLSS from Nvidias DLSS aren't you?

Again, there is no evidence that the PS5 supports ML on the same level as the XSX. If anything, there is more evidence to the contrary. AMD is again, irrelevant. It has been confirmed that the XSX has the hardware customisations to support accelerated ML. They have confirmed they have an API for it in the form of DirectML. Whether or not AMDs desktop GPUs support it is irrelevant.
So keep hope alive, but in the end I doubt you will see anything competitive with DLSS.
It's a question of it being utilised by devs or not.
 
I think CB rendering on PS4 Pro was a smart decision from Sony, and I wonder if they have advanced the tech for the PS5.
I haven't heard Cerny mention it, and so would have thought they would continue to use it.
 
Last edited:

Riky

$MSFT
Jesus fucking autism christ.

Has quite literally zero evidence to support theory.

Spams image that doesn't say what you claim to browbeat and shut down any conversation.

Quotes random twitter bullshit.

But is protected autist on neogaf.

***

As for the topic, only one of the consoles has direct confirmation of hardware support of ML algorithms/acceleration.

The other has at least 2 first hand reports from developers/engineers for the console saying it does not support it while not having any direct claims from sony saying it supports it.

After seeing DLSS 2.0 in action, there is clear evidence that it could indeed make a difference. Whether or not it will be utilised is another story.

He's been posting that slide ever since he's much hyped Sony being the real RDNA2 theory turned out to be completely wrong. So now babbles on about Infinity Cache and shared Cache instead as proof that XSX isn't RDNA2 and that all will again be revealed soon.....more tick tock crap basically.

He totally ignored what MS and AMD actually said,

"Xbox Series X|S are the only next-generation consoles with full hardware support for all the RDNA 2 capabilities AMD showcased today...........including hardware accelerated DirectX Raytracing, Mesh Shaders, Sampler Feedback and Variable Rate Shading."

They never said the XSX is identical to the new AMD GPUs, they said it has hardware support for the features they revealed. These features obviously don't need Infinity Cache.

As for Machine Learning they said this in the very same statement,

" we have gone even further introducing additional next-generation innovation such as hardware accelerated Machine Learning capabilities for better NPC intelligence, more lifelike animation, and improved visual quality via techniques such as ML powered super resolution."

It's all there in the statement but it doesn't fit his agenda so he ignores it.
 

Panajev2001a

GAF's Pleasant Genius
Custom debate from PS4 and PS4 Pro amounts to BS PR relative to R7 265 and RX 470 respectively.

I'm not buying the "custom" debate. I'm not going to fall for it again!

Sure, you be you :). Not sure why you need to bring up the drama and what BS there is with regards to PS4 and PS4 Pro, but magic 8 ball says you did not like them much ;).
 

Panajev2001a

GAF's Pleasant Genius


stop spreading FUD, do you guys ever tire of being wrong?


Do you even read what people type and think of context before seeing red in your eyes and typing angry replies :)?

The actual argument debated is meaningless it was the attitude at the same point in time when there was unknown data about either console: in the XSX case that poster was hopefully despite lack of data (“it is custom surely, we do not know, but that does not mean it is not there”) and the same person in another thread about PS5 was hyper cynical and pessimistic because lack of data (“if they have not explicitly called it out, it is because it is not there, not there!!!”).
 
Ok so going through and reading all of you guys comments i can safely say that the PS5 does not have AMD Machine Learning capability right now but that not to say that in the future that they might implement it of some kind.
 
Last edited:

Eternal21

Member
Remember when cloud processing was supposed to be the differentiator last gen? What happened to Crackdown 3 and the "power of the cloud"? Same shit, different pile. At the end of the day one platform has Naughty Dog and masterpieces that come with it, and the other has Craig.
 
Remember when cloud processing was supposed to be the differentiator last gen? What happened to Crackdown 3 and the "power of the cloud"? Same shit, different pile. At the end of the day one platform has Naughty Dog and masterpieces that come with it, and the other has Craig.
MS was just too early with the cloud stuff.
Now we actually have cloud gaming, and things like ML and DLSS have the majority of the compute done on the cloud.

And the whole Craig thing is not going to age well when Starfield, TES, Fallout, Doom, Wolfenstein, Forza, Gears, Quake etc are all exclusive to MS.
 

MrFunSocks

Banned
Remember when cloud processing was supposed to be the differentiator last gen? What happened to Crackdown 3 and the "power of the cloud"? Same shit, different pile. At the end of the day one platform has Naughty Dog and masterpieces that come with it, and the other has Craig.
Did you even play crackdown 3 MP? The cloud destruction was incredible, completely revolutionary.
 

rnlval

Member
Right, so when your Xbox gets a Nvidia based GPU then we can celebrate. Until then the Xbox and PS5 are based on AMD and AMD has not shown any ML scaling solution and likely it will not be as good/fast when it appears. That's assuming Nvidia doesn't have a patent around the secret sauce, which they likely do. Nvidia is also on the 2nd gen of this tech and has designed their latest GPU on it (in part).

So keep hope alive, but in the end I doubt you will see anything competitive with DLSS.
AMD's CDNA has discrete tensor cores.
 

rnlval

Member
Again, your one source is a CEO quote who has no technical background. Explain how a shader core is used for ML and how they are different.
Like NVIDIA's CEO, AMD's CEO is an engineer which is better than a certain pure MBA CEO governed tech companies.

Being a CEO doesn't remove the person from being an engineer. A tech company should be lead by engineers.

Both Atari's and Commodore's upper management was non-engineers.
 
Last edited:

rnlval

Member
I don't thinks so, it'll probably end just like Rapid Packed Match in PS4 Pro, which is basically the same tech, a.k.a. unused. Even if MS figures out how to actually implement the tech suitable for games, it'll have to sacrifice CUs to do so, and I don't think they'll be willing to do it. Especially there's already a difference in the raw GPU power, plus VRS that by itself can boost the performance by 5-20%, so I don't really think they'll ever need the DirectML

Newer hardware provides advances in ML performance through the use of FP16 precision and designated ML space on chips. DirectML’s metacommands provide vendors a way of exposing those advantages through their drivers to a common interface. Developers save the effort of hand tuning for individual hardware but get the benefits of these innovations.

-----

So, how do I access NVIDIA RTX's Tensor cores that would work with XSX GPU?
 

rnlval

Member
DLSS is a Nvidia IP. AMD does not have the equivalent and if they do come up with some upscaling tech, I doubt it will be as efficient or as good. So with AMD silicon, let's not expect miracles, everything is a trade off. If you spend cycles on a ML scaling solution, you will have less to spend in other areas.
Note that RDNA 2 GPUs are already running "smart" pixel re-construction de-noise pass for hardware real-time raytracing DXR workloads WITHOUT NVIDIA's discrete Tensor cores.

Certain ML competency is inherent in RDNA 2 due to DXR's needs.

While XSX GPU's shaders are more suited for ML, it doesn't mean PS5 doesn't have basic ML competency when it's needed for "smart" pixel re-construction de-noise pass for hardware real-time raytracing
workloads.

We don't know how DXR's denoise pass is handled on RDNA 2 GPUs e.g. is it fix function hardware solution just for denoise pass or using RPM shaders. NAVI compute unit has two known implementation i.e.
1. basic RPM with double rate FP16 compute unit
2. optional dot4/dot8 capable compute unit.


Tom Warren from The Verge received more information on this super-resolution technology:
While AMD is promising to go head to head with Nvidia in 4K gaming and more, the one big missing piece of this battle is a lack of an equivalent to Nvidia’s DLSS. Nvidia’s AI-powered super sampling technology has been transformative for the games that support it, bringing great image quality and higher frame rates by simply toggling a game setting.
AMD tells me it has a new super sampling feature in testing, which is designed to increase performance during ray tracing. The company is promising its super sampling technology will be open and cross-platform, which means it could come to next-gen consoles like the Xbox Series X and PS5. AMD is working with a number of partners on this technology, and it’s expecting strong industry support. Unfortunately, this won’t be ready for the launch of these three new Radeon RX 6000 Series cards.
 
Last edited:

Clintizzle

Lord of Edge.
Where on the hotchips does it mention you need dedicated ML to do AMD new open source upscaling method ?

Where in the AMD 6800 XT reveal does it go on about ML capability and the new upscaling - no where !

Ps5 can do ML functions on its hardware, is it designed for deep learning high performance like tensor cores, hell no.

The whole thread is a massive assumption that AMD is going to release an open source upscaling and it will need lots of dedicated ML that only XSX has and ps5 has not to fuel console wars - that is a reach and you know it -

Its a FUD thread about ps5 lacking in something with no sources before reveal - what a surprise. Posters see it for what it is.

Heck even 6800 has not revealed dedicated ML yet.

I am not an expert in this, not even close I have no backgrund in ML, but I can read simple english.
I smell a classic Geordiemp meltdown incoming. Get yer popcorn, lads.
 
My source is AMD DirectML is using DirectX feature which the PS5 does not use since it is Microsoft technology did you even read the posting now show me where it shows that PS5 have AMD DirectML?

You've confused the fact that DirectML is a software API that can run on any hardware supported.

PS5 doesn't use DirectX or DirectML, true, but PS5 will use it's own graphics and ML API specific for Sony's own console platform.

It's pretty much the same as how PS5 doesn't have RTX (an NVidia hardware technology) or DirectRT (a MS software API) for raytracing. It has both its own RT hardware and its own RT API that exposes the hardware features to graphics programmers for them to more easily leverage the features in their games.

People also need to stop trying to equivocate DirectML and NVidia's DLSS technology. Any ML algorithm written using DirectML will use standard compute resources on an AMD GPU (and the same is true for XSX). Yes MS has confirmed the inclusion of Rapid Packed Math for integer instructions on their RDNA2 XSX/S GPUs for accelerating ML computation, but it's very unlikely to be anything close to NVidia's DLSS 2.0 tech; which boasts the luxury of utilising Tensor cores which are up to an order of magnitude faster at ML computations, even with int RPM on AMD's newest GPUs (which NVidia's cards have had for a generation already at least).

The reality is, both AMD's DLSS-like solution, or MS's DirectML super resolution, may still be too slow to be effectively used in games, because at their most rudimentary level, they're still reliant on more general purpose "sparse data"-based computing devices than the dedicated "dense data"-based Tensor cores NVidia GPUs have that are just many times faster at inferencing computation.
 

rnlval

Member
You've confused the fact that DirectML is a software API that can run on any hardware supported.

PS5 doesn't use DirectX or DirectML, true, but PS5 will use it's own graphics and ML API specific for Sony's own console platform.

It's pretty much the same as how PS5 doesn't have RTX (an NVidia hardware technology) or DirectRT (a MS software API) for raytracing. It has both its own RT hardware and its own RT API that exposes the hardware features to graphics programmers for them to more easily leverage the features in their games.

People also need to stop trying to equivocate DirectML and NVidia's DLSS technology. Any ML algorithm written using DirectML will use standard compute resources on an AMD GPU (and the same is true for XSX). Yes MS has confirmed the inclusion of Rapid Packed Math for integer instructions on their RDNA2 XSX/S GPUs for accelerating ML computation, but it's very unlikely to be anything close to NVidia's DLSS 2.0 tech; which boasts the luxury of utilising Tensor cores which are up to an order of magnitude faster at ML computations, even with int RPM on AMD's newest GPUs (which NVidia's cards have had for a generation already at least).

The reality is, both AMD's DLSS-like solution, or MS's DirectML super resolution, may still be too slow to be effectively used in games, because at their most rudimentary level, they're still reliant on more general purpose "sparse data"-based computing devices than the dedicated "dense data"-based Tensor cores NVidia GPUs have that are just many times faster at inferencing computation.
You haven't factored in memory bandwidth bound issues which is a separate topic for RTX.

Your NVIDIA discrete tensor cores argument is a red herring for XSX vs PS5

Repeating, Source 1
From https://wccftech.com/xbox-series-xs...ning-powered-shader-cores-says-quantic-dream/

"The shader cores of the Xbox are also more suitable to machine learning, which could be an advantage if Microsoft succeeds in implementing an equivalent to Nvidia’s DLSS (an advanced neural network solution for AI)."

--------------


Also, DirectML Meta Commands enables vendor architecture-specific hardware access.

"Metacommands expose a black box of functionality up to DirectML, which allows hardware vendors to provide DirectML access to vendor hardware-specific and architecture-specific optimizations" . From https://docs.microsoft.com/en-us/windows/win32/direct3d12/dml-intro

Remember this?

 
Last edited:
You haven't factored in memory bandwidth bound issues which is a separate topic for RTX.

Your NVIDIA discrete tensor cores argument is a red herring for XSX vs PS5

Repeating, Source 1
From https://wccftech.com/xbox-series-xs...ning-powered-shader-cores-says-quantic-dream/

"The shader cores of the Xbox are also more suitable to machine learning, which could be an advantage if Microsoft succeeds in implementing an equivalent to Nvidia’s DLSS (an advanced neural network solution for AI)."

--------------


I'm not really focused on the comparison between XSX and PS5, but rather the difference between what we can expect from ML inferencing on consoles versus NVidia's DLSS solution (as is the OP).

As far as I know, Wccftech is speculating on the above quote, as there is no confirmation at all on the presence or lack of int RPM (which is what the quote is referring to) for PS5. So the jury is still out.

Also, DirectML Meta Commands enables vendor architecture-specific hardware access.

"Metacommands expose a black box of functionality up to DirectML, which allows hardware vendors to provide DirectML access to vendor hardware-specific and architecture-specific optimizations" . From https://docs.microsoft.com/en-us/windows/win32/direct3d12/dml-intro

Remember this?



This is all API related. If Sony designed the PS5 to accelerate ML computation with the inclusion of int RPM, they will expose those hardware features in the same way in their own bespoke API. The above doesn't address that, so I'm sure of the intended relevance.
 

rnlval

Member
I'm not really focused on the comparison between XSX and PS5, but rather the difference between what we can expect from ML inferencing on consoles versus NVidia's DLSS solution (as is the OP).

As far as I know, Wccftech is speculating on the above quote, as there is no confirmation at all on the presence or lack of int RPM (which is what the quote is referring to) for PS5. So the jury is still out.

This is all API related. If Sony designed the PS5 to accelerate ML computation with the inclusion of int RPM, they will expose those hardware features in the same way in their own bespoke API. The above doesn't address that, so I'm sure of the intended relevance.
Again, the context is shader cores hardware MORE suitable to machine learning between the consoles.

Source 1
From https://wccftech.com/xbox-series-xs...ning-powered-shader-cores-says-quantic-dream/

The shader cores of the Xbox are also more suitable to machine learning, which could be an advantage if Microsoft succeeds in implementing an equivalent to Nvidia’s DLSS (an advanced neural network solution for AI).
 

IntentionalPun

Ask me about my wife's perfect butthole
TheThreadsThatBindUs TheThreadsThatBindUs : There was a DLSS implementation for Control that didn't use tensor cores and it was still pretty good I guess, just not as good as actual DLSS 2.0


Of course, this isn't the first DLSS implementation we've seen in Control. The game shipped with a decent enough rendition of the technology that didn't actually use the machine learning Tensor core component of the Nvidia Turing architecture, relying on the standard CUDA cores instead. It still provided a huge performance boost, and generally looked better the higher up the resolution chain you went, but the new 2.0 revision offers a profound improvement.

We really just have no way of knowing how good a DLSS like solution can be on next-gen consoles, but people certainly shouldn't get their hopes up too high. Better to be pleasantly surprised.
 
Last edited:

Jokerevo

Banned
What's the point if you don't have the talented studios and the games? I mean do you not think for one second that 1st party Sony hasn't come up with their own solution.

MS need to show us something that genuinely embodies all these power and architectural advantages because last gen they were also most powerful and yet somehow failed to find a single game to harness that power and say :fuck you this is what XoX flops looks like.

Meanwhile we have studios extracting every last bit of 1.84 shitttyflops on the ps4 and producing stuff like TLoU2 and GoW etc etc that really shouldn't be fucking possible and now they have 4 times that power to play with and you're what? Showing me Halo infinite that looks like it isn't even last gen. Where is all the tech advantages in that showcase? Are MS ever actually going to walk the walk?
 
Last edited:
What's the point if you don't have the talented studios and the games? I mean do you not think for one second that 1st party Sony hasn't come up with their own solution.

MS need to show us something that genuinely embodies all these power and architectural advantages because last gen they were also most powerful and yet somehow failed to find a single game to harness that power and say :fuck you this is what 12 flops looks like.

Meanwhile we have studios extracting every last bit of 1.84 shitttyflops on the ps4 and producing stuff like TLoU2 and GoW etc etc that really shouldn't be fucking possible and now they have 4 times that power to play with and you're what? Showing me Halo infinite that looks like it isn't even last gen. Where is all the tech advantages in that showcase? Are MS ever actually going to walk the walk?
Wow. You seem a little insecure.
Forza and Forza Horizon are two of the best looking racing games the whole generation.
Gears 5 was a super looking game.
Flight Sim is the best looking next gen game I have seen.
Then we have Hellblade 2, Project Mara, Forza 8 as well as all of Bethesda next games. MS owns Idtech engine as well.

Sony will continue to produce super games, and MS will up the anti big time this gen.
 

Jokerevo

Banned
Wow. You seem a little insecure.
Forza and Forza Horizon are two of the best looking racing games the whole generation.
Gears 5 was a super looking game.
Flight Sim is the best looking next gen game I have seen.
Then we have Hellblade 2, Project Mara, Forza 8 as well as all of Bethesda next games. MS owns Idtech engine as well.

Sony will continue to produce super games, and MS will up the anti big time this gen.

Insecure? You're citing 2 racing games as system sellers? Gears 5? 50m sales for the powerful console that couldn't even pass the portable machine with its simple looking games?

So 3 titles in the entire gen for the most powerful console? Can you understand what I'm trying to say?

I was just trying to illustrate how power is nothing without control.

MS needs talent. That talent has to be cultivated. Do 343 look talented? Look at the 360 era.
 
Top Bottom