• 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.

Next-Gen PS5 & XSX |OT| Console tEch threaD

Status
Not open for further replies.


RedGamingTech says he was contacted by a developer who works in a studio and has a PS5 developer kit, he gave some information however RGT did state that this was not one of his regular "AMD and Sony" sources so take it with a grain of salt! However RGT did say a lot of the things he was told by this developer corroborated with stuff he heard from more reliable sources.

  • Why was Sony hesitant on revealing a lot of deep technical information about the PS5? Apparently this was because Jim Ryan wanted to focus less on the technical side of the PS5 and instead focus more on showing the games and experiences of the PS5. They believe that most customers don't really care about the technical aspects of the console but rather focus on games and pricing
  • Is it difficult to code for the PS5? they said no, according them it only took a few weeks to get code up and running on the PS5 despite the lack of initial optimisation. The developers also saw significant rises in frame rate when they made use of the CPU's SMT
  • PS5 is capable of the same lower precision operations as the Series X and would be capable of upsampling techniques in line with RDNA 2 dGPU's and Series X
  • With the game the developer was coding, the GPU was hitting it's maximum clock frequency 95% of the time with the more demanding sections dropping to around 2.1 Ghz however this only happened with very intense CPU usage and the frequencies could shift within a single frame multiple times.
  • Developer thinks devs will optimise their code better in the future when more tools are available and the architecture is understood better
  • Developer says CPU is Zen 2 based however the cache is 8MB and unified the same as Zen 3
  • There is NOT an additional cache pool on the GPU (infinity cache) however there is a lot of cache coherency which mitigates a lot of bandwidth issues
  • Memory addressing is much faster on the PS5 compared to the PS4
  • The Geometry Engine is the "Pièce de résistance" of the console and is totally custom, the developer states that the graphics pipeline strength of the PS5 comes from the GE
  • The developers are being actively encouraged to investigate the Geometry Engine with robust code samples and demos as well as information on how it functions
  • Very few games right now leverage the Geometry Engine to it's fullest potential because it does take a long time to learn. It's going to be most likely second and third wave games when developers start fully utilising it (this includes 3rd party)
  • VR and open world games will benefit heavily from the GE as well
  • The functionality of the Geometry Engine closely emulates a mesh shader of the DX12 U, however it's better in some ways because it allows the geometry to be culled much earlier and allows more control than a mesh shader, this is through a compute based primitive shader. However it is much more tricky to code for than a DX12 Mesh Shader, although this is from the perspective of this developer only
  • VRS is handled with the PS5's Geometry Engine and is handled much earlier in the pipeline, VRS with GE allows very high levels of accuracy
I havent quoted everything verbatum because it's way to long so just check out the video for yourselves. I might have missed 1 or 2 things. Again take this information with a grain of salt, take it or leave it. The developer said that more information "might" be revealed in the coming days because most of the NDA's will be lifted however this is not certain and will depend on how open developers want to talk about it.


It's pure ROAD TO PS5. Nothing more, nothing less.

Oh, wait, there's something more: Zen3 8MB cache CCX and int4 and int8 precision. These low-precision operators are in RDNA 1 white paper. Obviously PS5 would implement.
 
Last edited:

Lethal01

Member


RedGamingTech says he was contacted by a developer who works in a studio and has a PS5 developer kit, he gave some information however RGT did state that this was not one of his regular "AMD and Sony" sources so take it with a grain of salt! However RGT did say a lot of the things he was told by this developer corroborated with stuff he heard from more reliable sources.

  • Why was Sony hesitant on revealing a lot of deep technical information about the PS5? Apparently this was because Jim Ryan wanted to focus less on the technical side of the PS5 and instead focus more on showing the games and experiences of the PS5. They believe that most customers don't really care about the technical aspects of the console but rather focus on games and pricing
  • Is it difficult to code for the PS5? they said no, according them it only took a few weeks to get code up and running on the PS5 despite the lack of initial optimisation. The developers also saw significant rises in frame rate when they made use of the CPU's SMT
  • PS5 is capable of the same lower precision operations as the Series X and would be capable of upsampling techniques in line with RDNA 2 dGPU's and Series X
  • With the game the developer was coding, the GPU was hitting it's maximum clock frequency 95% of the time with the more demanding sections dropping to around 2.1 Ghz however this only happened with very intense CPU usage and the frequencies could shift within a single frame multiple times.
  • Developer thinks devs will optimise their code better in the future when more tools are available and the architecture is understood better
  • Developer says CPU is Zen 2 based however the cache is 8MB and unified the same as Zen 3
  • There is NOT an additional cache pool on the GPU (infinity cache) however there is a lot of cache coherency which mitigates a lot of bandwidth issues
  • Memory addressing is much faster on the PS5 compared to the PS4
  • The Geometry Engine is the "Pièce de résistance" of the console and is totally custom, the developer states that the graphics pipeline strength of the PS5 comes from the GE
  • The developers are being actively encouraged to investigate the Geometry Engine with robust code samples and demos as well as information on how it functions
  • Very few games right now leverage the Geometry Engine to it's fullest potential because it does take a long time to learn. It's going to be most likely second and third wave games when developers start fully utilising it (this includes 3rd party)
  • VR and open world games will benefit heavily from the GE as well
  • The functionality of the Geometry Engine closely emulates a mesh shader of the DX12 U, however it's better in some ways because it allows the geometry to be culled much earlier and allows more control than a mesh shader, this is through a compute based primitive shader. However it is much more tricky to code for than a DX12 Mesh Shader, although this is from the perspective of this developer only
  • VRS is handled with the PS5's Geometry Engine and is handled much earlier in the pipeline, VRS with GE allows very high levels of accuracy
I havent quoted everything verbatum because it's way to long so just check out the video for yourselves. I might have missed 1 or 2 things. Again take this information with a grain of salt, take it or leave it. The developer said that more information "might" be revealed in the coming days because most of the NDA's will be lifted however this is not certain and will depend on how open developers want to talk about it.


Sounds about right, No need for any "secret sauce", people just haven't payed attention to all the great things that Sony have been saying for months.
 
T

Three Jackdaws

Unconfirmed Member
It's pure ROAD TO PS5. Nothing more, nothing less.

Oh, wait, there's something more: Zen3 8MB cache CCX and int4 and int8 precision. These low-precision operators are in RDNA 1 white paper. Obviously PS5 would implement.
Yeah funnily enough, RGT said in the video to go back and look at the PS5 talk because there was so much information in it which people overlooked, he actually says that in every video when discussing PS5 leaks.
 

geordiemp

Member
It's pure ROAD TO PS5. Nothing more, nothing less.

Oh, wait, there's something more: Zen3 8MB cache CCX and int4 and int8 precision. These low-precision operators are in RDNA 1 white paper. Obviously PS5 would implement.

The unified Cache is a big deal though, instead of Zen 2 with 4mB and 4MB split, its Zen2 with 8MB

So if data is needed that is in the wrong half of the cache, with a split 4 + 4 MB cache it will be much slower to process it.

Unified cache is a Zen3 optimisation.
 

Bojanglez

The Amiga Brotherhood
This is good. Much better to have a single pool than multiple. Much better than the XSX, as it has the split pool just like the regular Zen 2 processors.

Some reason I doubt this is accurate though.



Not surprising that there is no infinity cache, as others have indicated.

Amazing amount of info in that post.
I think Matt Hargett may have actually been teasing this principle when Tom Warren was saying the XB CPU was faster.

 
Last edited:

DeepEnigma

Gold Member
Infinity cache for ps5 is dead.

Clocks unpredictable not 100% gpu clocks even when in non cpu intensive sections.

Mesh shaders better now, may change later.

No confirmation of higher throughput of custom low bit code.

No confirmation of vrs 1.2

No confirmation of sfs

Some ridiculous statement that they didnt want to be too technical when the main reveal was a super technical talk .. obviously this is PR BS and they have nothing more to announce technically.

Once again redtechgaming walking back his predictions ..

Can always rely on Lort Lort to shit in your Wheaties and lack total context or nuance (on purpose?) of the GDC talk, and who it was really meant for under normal non COVID circumstances.
 

geordiemp

Member

We also inquired as to the different implementations of ray-traced reflections across PC, PlayStation 5, and Xbox Series X. PC has a slight advantage.

"The fundamentals of the ray-tracing are the same across all platforms that support it. For RTX on PC, we do ray-tracing for rougher surfaces in more situations. Ray-tracing relies on a simplified version of the world that we trace rays into. This has been optimised individually for each platform to maximise the number of rays we can process. Overall, the look is similar on all platforms supporting ray-tracing, but side-by-side, only experts will be able to spot differences"
 

PSX

Member
Dont be surprised. Give them a year the series S will push the PS5 over the edge.

:messenger_sunglasses:



IpjG8WR.jpg
 

onesvenus

Member

How Big is Black Ops Cold War?

To pre-load Black Ops Cold War, players will need to ensure they have the right amount of available storage to launch the game. Here are the storage space numbers required for each system.
  • PS4: 95GB
  • PS5: 133GB
  • Xbox One: 93GB
  • Xbox Series X|S: 136GB
  • PC 125GB (Full Game on Ultra Graphics), 82GB (Full Game), 35GB (Multiplayer Only)
Are we expecting higher quality assets in PS5 vs XSX? The way some posters were talking about kraken and oodles it seemed like assets size would be 20% less than XSX

The functionality of the Geometry Engine closely emulates a mesh shader of the DX12 U, however it's better in some ways because it allows the geometry to be culled much earlier
I don't understand how the geometry can be culled "much earlier" when it can be performed in the Amplification
Shader which is literally the first step in the pipeline
 

martino

Member
I'm sure you know more than a former Principal Engineer at Sony.
authority argument on a tweet you don't understand.
but your intelligence can only reach what needed for you to war...
what he says is correct though
but vrs doesn't solve the problem he is talking about.
Mesh shader exists and do and happen before pixel shader phase (where vrs happen)
 
Last edited:

geordiemp

Member
authority argument on a tweet you don't understand.
but your intelligence can only reach what needed for you to war...
what he says is correct though
but vrs doesn't solve the problem he is talking about.
Mesh shader exists and do and happen before pixel shader phase (where vrs happen)

If you read some of the patents, there is a compression between pixel vertices and pixel shaders and some optimisation where the offsets are done differently which is more efficient for the whole process.

I am sure the actual implementaion is even more complex than the patents which just outlines the basic, which are not light reading.
 
Last edited:

geordiemp

Member
oh I didn’t catch that

He made 2 comments, first about differences he could see, and second about PC settings high vs medium and diferent filtering of shadows / effect of roughness on reflections.

It will be puddlegate 2.0, 2 gens ago it was grass gate, now we are zoomed in on puddles and poppin counting :messenger_beaming:

Gives dictator something to talk about I guess - to me the models and animations on watchdogs was what needed attention but hey....
 
Last edited:

oldergamer

Member
Matt is a disgenous one.
VRS alone you don't cull
but since you have amplication/mesh shader. if you use them when you are at pixel shader step the useless vertices are already long gone

201104105050342219.jpg
Exactly. Microsoft demonstrated using this to cull geometry months back. There is too much BS going on with these features.
 


RedGamingTech says he was contacted by a developer who works in a studio and has a PS5 developer kit, he gave some information however RGT did state that this was not one of his regular "AMD and Sony" sources so take it with a grain of salt! However RGT did say a lot of the things he was told by this developer corroborated with stuff he heard from more reliable sources.

  • Why was Sony hesitant on revealing a lot of deep technical information about the PS5? Apparently this was because Jim Ryan wanted to focus less on the technical side of the PS5 and instead focus more on showing the games and experiences of the PS5. They believe that most customers don't really care about the technical aspects of the console but rather focus on games and pricing
  • Is it difficult to code for the PS5? they said no, according them it only took a few weeks to get code up and running on the PS5 despite the lack of initial optimisation. The developers also saw significant rises in frame rate when they made use of the CPU's SMT
  • PS5 is capable of the same lower precision operations as the Series X and would be capable of upsampling techniques in line with RDNA 2 dGPU's and Series X
  • With the game the developer was coding, the GPU was hitting it's maximum clock frequency 95% of the time with the more demanding sections dropping to around 2.1 Ghz however this only happened with very intense CPU usage and the frequencies could shift within a single frame multiple times.
  • Developer thinks devs will optimise their code better in the future when more tools are available and the architecture is understood better
  • Developer says CPU is Zen 2 based however the cache is 8MB and unified the same as Zen 3
  • There is NOT an additional cache pool on the GPU (infinity cache) however there is a lot of cache coherency which mitigates a lot of bandwidth issues
  • Memory addressing is much faster on the PS5 compared to the PS4
  • The Geometry Engine is the "Pièce de résistance" of the console and is totally custom, the developer states that the graphics pipeline strength of the PS5 comes from the GE
  • The developers are being actively encouraged to investigate the Geometry Engine with robust code samples and demos as well as information on how it functions
  • Very few games right now leverage the Geometry Engine to it's fullest potential because it does take a long time to learn. It's going to be most likely second and third wave games when developers start fully utilising it (this includes 3rd party)
  • VR and open world games will benefit heavily from the GE as well
  • The functionality of the Geometry Engine closely emulates a mesh shader of the DX12 U, however it's better in some ways because it allows the geometry to be culled much earlier and allows more control than a mesh shader, this is through a compute based primitive shader. However it is much more tricky to code for than a DX12 Mesh Shader, although this is from the perspective of this developer only
  • VRS is handled with the PS5's Geometry Engine and is handled much earlier in the pipeline, VRS with GE allows very high levels of accuracy
I havent quoted everything verbatum because it's way to long so just check out the video for yourselves. I might have missed 1 or 2 things. Again take this information with a grain of salt, take it or leave it. The developer said that more information "might" be revealed in the coming days because most of the NDA's will be lifted however this is not certain and will depend on how open developers want to talk about it.

That's awesome. Do we know if XSX CPU also has 8MB shared between all cores? Or if it's 4MB + 4MB like with Zen 2?
 
if this is true then come November 6th we will see the PS5 versions of games running on par with the xsx versions. No 18% difference.

If the difference is still 18% or bigger then Matt was BSing.

He talks about CPU and 18% difference is about GPU.

Why to say statements like that so confidently, when you dont clearly know what you are talking about if you mix cpu and gpu differences?

And you clearly didn't even consider that launch games may or may not have differences because of hundreds of different reasons and hardware is only one of those.

So differences or lack of them is far from fool proof evidence..

I think someone is bullshitting, but not the twitter dude
 

Mahavastu

Member
This is good. Much better to have a single pool than multiple. Much better than the XSX, as it has the split pool just like the regular Zen 2 processors.

Some reason I doubt this is accurate though.
There was a GDC talk by one of Sony's own devs, in which he said that on the PS4 it took about 190 cycles latency if the data is in the wrong half of the cache. Which was about the same latency as for normal RAM access, so that having the wanted data in cache did not help at all
That could be quite an incentive to use a unified cache, as you have fast access to double the data.
 

IntentionalPun

Ask me about my wife's perfect butthole
If the systems are close to equal that's a good thing power wise.

But I don't quite understand why anyone would really praise Sony for that unless their solution was cheaper? Is it what left room for the more impressive SSD?

They deserve praise for getting a much faster SSD in the same priced box either way though, as well as their $400 DE offering.
 

geordiemp

Member
If the systems are close to equal that's a good thing power wise.

But I don't quite understand why anyone would really praise Sony for that unless their solution was cheaper? Is it what left room for the more impressive SSD?

They deserve praise for getting a much faster SSD in the same priced box either way though, as well as their $400 DE offering.

20 % smaller silicon roughly. If you added 20 % more die area to ps5 is the point......

Pricing is a different discussion, loss leaders both.
 
Last edited:

martino

Member
If you read some of the patents, there is a compression between pixel vertices and pixel shaders and some optimisation where the offsets are done differently which is more efficient for the whole process.

I am sure the actual implementaion is even more complex than the patents which just outlines the basic, which are not light reading.
possibly but this was not the subject
you could wonder why say vrs is a not great without culling when there is a tool to do that : mesh shader.
in fact you can have both gains with xsx and all directx 12U gpu.
also this is what is documented by nvidia for mesh shader :

  • Higher scalability through shader units by reducing fixed-function impact in primitive processing. The generic purpose use of modern GPUs helps a greater variety of applications to add more cores and improve shader’s generic memory and arithmetic performance.
  • Bandwidth-reduction, as de-duplication of vertices (vertex re-use) can be done upfront, and reused over many frames. The current API model means the index buffers have to be scanned by the hardware every time. Larger meshlets mean higher vertex re-use, also lowering bandwidth requirements. Furthermore developers can come up with their own compression or procedural generation schemes.
    The optional expansion/filtering via task shaders allows to skip fetching more data entirely.

scale with more SM/CU (possibly MS/Nvidia using more cu/SM is not only for tf on paper or server use....), can use custom vertex encoding/generation/compression , more reuse....
yeah lot of gains come using mesh shader too and even more can playing with them.
everything become programmable and less dependent to fix functions relying more on CU/SM and solution dev with come up with

so claiming one is better on a patent alone seems more than premature....
 
Last edited:

Tmack

Member
if this is true then come November 6th we will see the PS5 versions of games running on par with the xsx versions. No 18% difference.

If the difference is still 18% or bigger then Matt was BSing.


Even if just slightly worse, thats still a HUGE win for PS5. It`s all about managing expectations.
 
Status
Not open for further replies.
Top Bottom