• 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.

Sony custom RDNA2 have their own VRS and mesh shading - Cerny and Naughty dog

BigLee74

Member
Do you know what the 3 points in the slide actually mean :messenger_beaming:
]
They are important, they detail the RDNA2 CU and what its doing for high frequencies > 2 Ghz.

I understand if you want to pretend it does not exist. But the hardware is fixed now.

Do I know what they mean? Not in the slightest! And to be honest, I don't particularly care.

Just passing comment that you are posting that picture so often, I think it's a forum advert! 😂
 

longdi

Banned
Do you know what the 3 points in the slide actually mean :messenger_beaming:
]
They are important, they detail the RDNA2 CU and what its doing for high frequencies > 2 Ghz.

I understand if you want to pretend it does not exist. But the hardware is fixed now.

You keep troll baiting freely. Rnda2 cu runs between 1.8~2ghz in games, from between 250~300w. :messenger_weary:

Do you think Sony will dedicate 300w to the gpu part?
 

Three

Member
It’s the first time I hear that VRS has anything to do with the Halo Infinite demo. You should let Digital Foundry know. And SFS is exactly the opposite approach of the PS5 SSD. it would still multiply the texture bandwidth. The point you could make is that the GPU bottleneck made it an overkill but Sony hasn’t stated that.
HaloVRS.jpg


DF should have been the ones telling you about this but they concentrated on the time of day in the game.
 
Last edited:

Elog

Member
Someone should tell him that variable rate shading has nothing to do with geometry.

Geometry is handled by mesh shading.

Geometry Engine is a hold over from GCN which AMD dropped in favour of supporting Microsofts solution.

You miss the whole point.

All the buzzwords used are about one thing and one thing only: How can we reduce the number of calculations to create a frame with minimum fidelity loss?

VRS downgrades the shading rate for parts of the 3D space that matters less for how the frame looks (e.g. periphery, obscured objects etc).

What Matt refers to is what Mark Cerny stated in 'Road to Ps5'. If you have an advanced enough GE you can a) cull all geometries that are obscured in the frame and b) annotate priorities to the areas of the geometries that are seen (e.g. highest priority to the middle of the frame, higher priority to objects closer to the view-point compared to those further away). Then you can apply primitive shaders the most to the highest priority areas, less to lower priority areas/objects and not at all to culled objects.

The result is identical - significant reduction in # of calculations with minimal visual fidelity loss. What is most efficient? I do not know but based on the rumour mill, the front loading of the whole pipeline like this seems like tomorrow's technology and not yesterday's which you imply.
 
Last edited:

TrebleShot

Member
Hoping this means the XSX has some sort of visible advantage and not just marketing speak, I am FAR from technical on these subjects but my interpretation is the Sony decided to do their own implementation via the G.E for VRS due to the enhanced throughput of the SSD which is highly custom also.

Not sure on the other advantage. Something about clocking CUs higher? my understanding is the PS5 CPU is clocked higher anyway, would that have any impact on performance of the GPU? no idea.

Hoping Assassin's Creed Valhalla uses RDNA2 VRS on XSX as thats what I went with for launch.
 

NEbeast

Member
It’s the first time I hear that VRS has anything to do with the Halo Infinite demo. You should let Digital Foundry know. And SFS is exactly the opposite approach of the PS5 SSD. it would still multiply the texture bandwidth. The point you could make is that the GPU bottleneck made it an overkill but Sony hasn’t stated that.

Vrs was clearly visible in screenshots posted on here around MS games showcase. Maybe I'm understanding it wrong but doesn't SFS just load what the eyes can see, therefore saving memory? It's exactly what cerny talked about in the road to ps5, " the ssd is fast enough that you can move assets into memory as the player turns" (it's not an exact quote). Maybe I have my wires crossed about SFS.
 
HaloVRS.jpg


DF should have been the ones telling you about this but they concentrated on the time of day in the game.

What's good for us gamers is that they no longer hold a monopoly on the market. And while their competition is relatively small when it comes to notoriety - they're gaining steam. They keep making "mistakes" and they'll keep digging their own graves. The biggest market share in traditional gaming is held by Sony. They're tip toeing around land mines. Not too bright..... it's a long game.
 
Last edited:

geordiemp

Member
Vrs was clearly visible in screenshots posted on here around MS games showcase. Maybe I'm understanding it wrong but doesn't SFS just load what the eyes can see, therefore saving memory? It's exactly what cerny talked about in the road to ps5, " the ssd is fast enough that you can move assets into memory as the player turns" (it's not an exact quote). Maybe I have my wires crossed about SFS.

PRT+ or SF does the MIP level stuff already on current gen including ps4, SFS is a final blending of the updated data which ps5 does not sue or will not need.
 

SatansReverence

Hipster Princess
You miss the whole point.

All the buzzwords used are about one thing and one thing only: How can we reduce the number of calculations to create a frame with minimum fidelity loss?

VRS downgrades the shading rate for parts of the 3D space that matters less for how the frame looks (e.g. periphery, obscured objects etc).

The tweet is trying to compare 2 completely different things.

VRS is NOT about geometry culling.

Mesh shading is an actual comparison.

Mesh shading culls geometry long before it reaches any shading operation
 

Elog

Member
The tweet is trying to compare 2 completely different things.

VRS is NOT about geometry culling.

Mesh shading is an actual comparison.

Mesh shading culls geometry long before it reaches any shading operation


You miss the point again.

ALL these techniques are about reducing the # calculations with minimal visual downgrade. That is it.

Sony has chosen to change how this is done in the PS5 by creating a new GE that is the main driver to achieve this. You imply that this is somehow worse - and we do not know that. Given that Sony sits on the largest group of world-class 3D engine software engineers I highly doubt it is a bad solution.

And please note that one goal for Sony here was to allow for the use of much more polygons/vertices, i.e. almost down to single pixel sized. The normal pipeline approach with methodologies that you mention starts to break down then. Have they succeeded? Who knows.

Point is that you claim advantage when there is so far no indication that is the case.
 

geordiemp

Member
You miss the point again.

ALL these techniques are about reducing the # calculations with minimal visual downgrade. That is it.

Sony has chosen to change how this is done in the PS5 by creating a new GE that is the main driver to achieve this. You imply that this is somehow worse - and we do not know that. Given that Sony sits on the largest group of world-class 3D engine software engineers I highly doubt it is a bad solution.

And please note that one goal for Sony here was to allow for the use of much more polygons/vertices, i.e. almost down to single pixel sized. The normal pipeline approach with methodologies that you mention starts to break down then. Have they succeeded? Who knows.

Point is that you claim advantage when there is so far no indication that is the case.

Its funny, even with Sony custom Geometry and VRS, and MS RDNA2 standard mesh and VRS, none of those techniques will be used in Nanitie anyway.
 

SatansReverence

Hipster Princess
You miss the point again.

ALL these techniques are about reducing the # calculations with minimal visual downgrade. That is it.

Sony has chosen to change how this is done in the PS5 by creating a new GE that is the main driver to achieve this. You imply that this is somehow worse - and we do not know that. Given that Sony sits on the largest group of world-class 3D engine software engineers I highly doubt it is a bad solution.

And please note that one goal for Sony here was to allow for the use of much more polygons/vertices, i.e. almost down to single pixel sized. The normal pipeline approach with methodologies that you mention starts to break down then. Have they succeeded? Who knows.

Point is that you claim advantage when there is so far no indication that is the case.

I'll simplify this,

Both consoles have advanced Geometry culling systems. (not even going into AMD dropping GE for Mesh Shading), XSX has VRS in addition to mesh shading.

The tweet being paraded around is trying to compare 2 entirely different functions of a rendering pipeline.
 
Last edited:

Elog

Member
I'll simplify this,

Both consoles have advanced Geometry culling systems. (not even going into AMD dropping GE for Mesh Shading), XSX has VRS in adition to mesh shading.

The tweet being paraded around is trying to compare 2 entirely different functions of a rendering pipeline.

My last post. The idea that a geometry engine is a geometry engine that you describe above is the same as stating that a shader is a shader.

There are multiple approaches to achieve this with very different data-paths. Sony has a new GE that - if the available information is correct - can do both advanced culling and prioritisation - with those priorities applied in all parts of the downstream pipeline to minimise GPU workloads. This cannot be done that way with any other GPU in the market.

I find it very interesting with such a different approach and we have already seen an amount of geometry on the screen in for example Demon's Souls that we have not seen before.

Let's wait for benchmarks but it is pretty rich to make assumptions that this is not beneficial for the platform based on what Mark Cerny, ND, Bluepoint, Guerrilla etc has achieved in the past. Developers seem very happy with the solution.
 
You miss the point again.

ALL these techniques are about reducing the # calculations with minimal visual downgrade. That is it.

Sony has chosen to change how this is done in the PS5 by creating a new GE that is the main driver to achieve this. You imply that this is somehow worse - and we do not know that. Given that Sony sits on the largest group of world-class 3D engine software engineers I highly doubt it is a bad solution.

And please note that one goal for Sony here was to allow for the use of much more polygons/vertices, i.e. almost down to single pixel sized. The normal pipeline approach with methodologies that you mention starts to break down then. Have they succeeded? Who knows.

Point is that you claim advantage when there is so far no indication that is the case.

Nothing that you say will convince him that he has no idea what he's talking about. Maybe he's realized he tripped but will double down to save face quoting stuff, videos and buzzwords he doesn't understand. At least he's behaving and hasn't thrown the word clown at others yet.

Show me how all this translates into games.

That's all I care about. The result.

All that ultimately really matters.

fd42eac2e4410280628d0cc0b1b1da2ff132b171.gifv
 
Last edited:

Krisprolls

Banned
Stop with all that buzzword talk everybody. You're - not - devs. You don't understand what you're talking about. You don't know the technical implications.

It's bad enough fighting for 1 pixel and 2 fps difference, let's not start the tech buzzword console war where nobody knows what they're talking about.

Wait for comparison videos, they're coming pretty soon.
 

Lort

Banned

SatansReverence

Hipster Princess
My last post. The idea that a geometry engine is a geometry engine that you describe above is the same as stating that a shader is a shader.

There are multiple approaches to achieve this with very different data-paths. Sony has a new GE that - if the available information is correct - can do both advanced culling and prioritisation - with those priorities applied in all parts of the downstream pipeline to minimise GPU workloads. This cannot be done that way with any other GPU in the market.

I find it very interesting with such a different approach and we have already seen an amount of geometry on the screen in for example Demon's Souls that we have not seen before.

Let's wait for benchmarks but it is pretty rich to make assumptions that this is not beneficial for the platform based on what Mark Cerny, ND, Bluepoint, Guerrilla etc has achieved in the past. Developers seem very happy with the solution.
You're still failing to provide any rationale for why that tweet is trying to compare geometry culling to variable rate shading or why it is attempting to claim that XSX isn't culling said geometry.
 

Njocky

Banned
You implying one of two things here (or both): AMD knows better than Sony, and Sony cant do better than AMD.
It’s not about knowing better. At the end of the day Sony is not designing its own GPU. They went to AMD for a reason.
You implying one of two things here (or both): AMD knows better than Sony, and Sony cant do better than AMD.
I'm implying that a conglomerate like Sony probably wouldn't hang the future of their biggest cash cow (PlayStation) on inventions by a single engineer. That's precisely why they hired AMD's R&D department and didn't design the PS5 GPU internally like their TV or camera chips. The advantage of this is obvious. AMD was able during their presentation to display exactly how their technology fares against Nvidia in actual games. What tangible data could Cerny have presented to his bosses to prove the merit of his inventions? Just because Cerny filed a patent does not mean it was the blueprint for the PS5.

At the end of the day we should be able to rely on Sony's communication instead of posts by geordiemp about what's inside the PS5. At the very least we know what's NOT in the PS5, as MS has now made sure of that. We also know that Sony is far from silent when they think they have the superior features. Just look at the controller coverage.
 
Last edited:

Yoboman

Member
It’s not about knowing better. At the end of the day Sony is not designing its own GPU. They went to AMD for a reason.

I'm implying that a conglomerate like Sony probably wouldn't hang the future of their biggest cash cow (PlayStation) on inventions by a single engineer. That's precisely why they hired AMD's R&D department and didn't design the PS5 GPU internally like their TV or camera chips. The advantage of this is obvious. AMD was able during their presentation to display exactly how their technology fares against Nvidia in actual games. What tangible data could Cerny have presented to his bosses to prove the merit of his inventions? Just because Cerny filed a patent does not mean it was the blueprint for the PS5.

At the end of the day we should be able to rely on Sony's communication instead of posts by geordiemp about what's inside the PS5. At the very least we know what's NOT in the PS5, as MS has now made sure of that. We also know that Sony is far from being silent when they think they have the superior features. Just look at the controller coverage.
Even if they are bespoke features, they are still codeveloped with AMD. Nobody suggested otherwise.
 
Its no good explaining anything when it comes to the ps5 it's not going to break through to certain people here.

From what I understand are people actually claiming the ps5 is less than rDNA 2 because of a Microsoft affiliated tweet is that where all the confidence is coming from ? That's weak if it is .
 
Last edited:

RoboFu

One of the green rats
Idk if I have to point this out but a patent doesn’t mean it’s actually in there. Probably is but it’s not a confirmation.
 

TGO

Hype Train conductor. Works harder than it steams.
Sony took hardware and customised it?
That's crazy talk, they've never done that before
And they never built computer hardware before to have the knowledge to do that!
What do you think they are electronic engineers!
 

JonnyMP3

Member
Wait? What?
We arguing over shaders again now?
Didn't we establish months ago that the PS5 uses Primitive Shaders because of the Geometry Engine and Scrubbers and Pipeline allows for advanced Primitives because of the SSD saturating the GPU?


Or did I miss something new in the last 2 months?
 

Tomeru

Member
It’s not about knowing better. At the end of the day Sony is not designing its own GPU. They went to AMD for a reason.

I'm implying that a conglomerate like Sony probably wouldn't hang the future of their biggest cash cow (PlayStation) on inventions by a single engineer. That's precisely why they hired AMD's R&D department and didn't design the PS5 GPU internally like their TV or camera chips. The advantage of this is obvious. AMD was able during their presentation to display exactly how their technology fares against Nvidia in actual games. What tangible data could Cerny have presented to his bosses to prove the merit of his inventions? Just because Cerny filed a patent does not mean it was the blueprint for the PS5.

At the end of the day we should be able to rely on Sony's communication instead of posts by geordiemp about what's inside the PS5. At the very least we know what's NOT in the PS5, as MS has now made sure of that. We also know that Sony is far from silent when they think they have the superior features. Just look at the controller coverage.

Nvm, not going any further...
 
Last edited:

Elog

Member
You're still failing to provide any rationale for why that tweet is trying to compare geometry culling to variable rate shading or why it is attempting to claim that XSX isn't culling said geometry.

I do not know what to write to make you understand.

Let's make an oversimplification of a current pipeline (XSX, 20XX/30XX cards etc).

You create your basic geometry, culling is done by the mesh shaders (that also do some of the geometry work in a feed-back loop) on the meshlet level, your VRS function determines how much shaders work that is conducted on each meshlet and then you rasterise and you are good to go. Oversimplification but ok for this discussion.

This is a good approach but has a few challenges: 1) Each meshlet cannot contain too many primitives (triangles etc), so if your geometry is complex you start to bottle-neck the system and 2) both the mesh-shader and the VRS function independently of each other analyse the geometry to conduct culling and to allocate shader work. The more advanced your geometry to the more this duplication of processing becomes a bottle-neck.

In short: If you go for small or pixel sized primitives you potentially give your GPU a myocardial infarction.

What Mark Cerny talked about is that the culling (mes shader) and prioritisation (VRS) is done at the GE level. That is different - it is a very different GE. His idea - as he described it - is that this means that all functions downstream of that can utilise the now already culled geometry and apply work in accordance with the set priorities. This can be used for texture work, shader work, RT etc. And this calculation is only done once as opposed to more than once per the traditional pipeline. He also claims that this new GE can work with very small primitives without clogging.

We have to wait for benchmarks to understand how this system performs. Mark has made some bold claims here and we know that for example the PS3 did not fare well with all its proprietary technology. My assumption has been that this will penalise the PS5 on multiplats but let it shine in 1st party titles. For the multiplat penalty it of course boils down to the PS5 API and how well traditional pipeline work is translated to this new way of working. I am pleasantly surprised by the third party performance so far so it seems that development environment and the APIs are very well made.

You just off the bat think that the PS5 lacks something - that is what bothers me. It has something new and shiny. Is it good? Time will tell but I love people such as Mark Cerny that tries new things - that takes risks. So I hope he has succeeded.
 

SatansReverence

Hipster Princess
I do not know what to write to make you understand.

Let's make an oversimplification of a current pipeline (XSX, 20XX/30XX cards etc).

You create your basic geometry, culling is done by the mesh shaders (that also do some of the geometry work in a feed-back loop) on the meshlet level, your VRS function determines how much shaders work that is conducted on each meshlet and then you rasterise and you are good to go. Oversimplification but ok for this discussion.

This is a good approach but has a few challenges: 1) Each meshlet cannot contain too many primitives (triangles etc), so if your geometry is complex you start to bottle-neck the system and 2) both the mesh-shader and the VRS function independently of each other analyse the geometry to conduct culling and to allocate shader work. The more advanced your geometry to the more this duplication of processing becomes a bottle-neck.

In short: If you go for small or pixel sized primitives you potentially give your GPU a myocardial infarction.

What Mark Cerny talked about is that the culling (mes shader) and prioritisation (VRS) is done at the GE level. That is different - it is a very different GE. His idea - as he described it - is that this means that all functions downstream of that can utilise the now already culled geometry and apply work in accordance with the set priorities. This can be used for texture work, shader work, RT etc. And this calculation is only done once as opposed to more than once per the traditional pipeline. He also claims that this new GE can work with very small primitives without clogging.

We have to wait for benchmarks to understand how this system performs. Mark has made some bold claims here and we know that for example the PS3 did not fare well with all its proprietary technology. My assumption has been that this will penalise the PS5 on multiplats but let it shine in 1st party titles. For the multiplat penalty it of course boils down to the PS5 API and how well traditional pipeline work is translated to this new way of working. I am pleasantly surprised by the third party performance so far so it seems that development environment and the APIs are very well made.

You just off the bat think that the PS5 lacks something - that is what bothers me. It has something new and shiny. Is it good? Time will tell but I love people such as Mark Cerny that tries new things - that takes risks. So I hope he has succeeded.
You're still trying to compare 2 entirely different parts of the rendering pipeline.
 
Its funny, even with Sony custom Geometry and VRS, and MS RDNA2 standard mesh and VRS, none of those techniques will be used in Nanitie anyway.

Is that really the case or was Epic trying to be vague in order not to divulge NDA'd features of the PS5? Mark Cerny did not explain in detail those features so maybe Epic was trying to avoid discussing those features.
 

Handy Fake

Member
Wait? What?
We arguing over shaders again now?
Didn't we establish months ago that the PS5 uses Primitive Shaders because of the Geometry Engine and Scrubbers and Pipeline allows for advanced Primitives because of the SSD saturating the GPU?


Or did I miss something new in the last 2 months?
Why no. It's just folk shouting about things they don't understand because it seemingly one ups the opposition as per usual.
 

Thirty7ven

Banned
You're still trying to compare 2 entirely different parts of the rendering pipeline.

variable rate shading is a process that allows different parts of the fov to be rendered at different resolutions. It’s an optimization process. It’s in Sony and naughty dog patents already, they just don’t call it VRS.

This is a another marketing stunt. What the Sony guy is saying is that through the geometry engine they already have a path for those optimizations. The flexibility is already there.
 
Last edited:
Top Bottom