- Status
Vaztu
Member
Don't know if they got the information from independent sources or sharing each others' sources.
But, this dude posted this on Oct 17, 2019.
Radical_3d
Member
Meh. The Mac-head in me preferred when they used G5 power stations.that's how m$ rolls. nothing newi still remember E3, when they said that it is running on xbox.. until "xbox" crashed to Windows window
p.s. i checked your link later time after my comment
I call it BS. The diminishing returns of making such a model for in-game purposes wouldn't be worth it. I bet the main difference will be in the ground detail.
Last edited:
FranXico
Member
Because it would be ridiculously expensive compared to using discs.
Last edited:
Bonsaibäumchen
Member
Or its for ps5's PS3 Backward compatibility which in turn is backward compatible for PS2 and PS1 games
dottme
Member
Cost, and especially in mass production. In mass production, pressing a disc is really cheap. Writing to an SSD or a rom is much more expensive (it takes more time and time is money).
ArcaneNLSC
Member
xacto
Member
Famous words "I've been told". He makes it sound like he's the only available and true insider in the Universe. So sad.
"Millions of polys!"
SMH.
Snowdonhoffen
Member
Honestly I think this will be done but just in close up shots
I doubt devs really care that much about polygon counts anymore. We've reached a point of diminishing returns where millions of polygons in one character is just a waste of power that could be used somewhere else.
My guess is the "leak" was made by someone with very basic understanding of game graphics and thinks that more polygons = more graphics.
Kusarigama
Member
This is exactly what came to my mind as well.Or its for ps5's PS3 Backward compatibility which in turn is backward compatible for PS2 and PS1 games
Radical_3d
Member
The NeoGeo of the XXI century!
farmerboy
Member
FeiRR
Banned
It's quite understandable why they were running PCs instead of devboxes: for security and financial reasons. But telling everyone that it ran on consoles is a whole different thing.that's how m$ rolls. nothing new
p.s. i checked your link later time after my comment
Doesn't have to be like that. You create one high-poly model for all purposes: gameplay, cutscenes, etc. If your LOD scaling works perfectly well, there's no problem with using the same model. In closer takes the additional polys just load in place while with camera further away they just disappear. This requires a new way of dealing with LOD, not tier switching but smooth transitions. If they managed to tackle that, we're going to see wonders., like in the Crysis intro but in realtime.
I'd have no problem with switching to digital if I could still sell games and if digital copies weren't more expensive than disks. Until that happens, I'll support physical media and hope they don't disappear.
FranXico
Member
What the hell? Is he getting his leaks from f-ing Twitter?
Radical_3d
Member
What I'm saying is, do you need a more poly starring (I don't remember the name of the girl). Better shaders. Better textures. Ok. But, more polygons? Other gaffer suggested that the leak was made by someone with little knowledge of tech. I can believe that. There is a lot of staff in a studio and only so many of them understand what better graphics means.
Snowdonhoffen
Member
Who knows, either he read it from Twitter or he really heard it from one of his sources. But as always, we'll have to wait and see if it's true.
FeiRR
Banned
I explained why you might need more polys, even gave a video example...
3liteDragon
Member
Why can't it be both? This is how I understood it since I wasn't familiar with the tech before, I watched NVIDIA's mesh shading video and their approach is based on LOD, which is what PS5's primitive shaders are based on as well.
So if you look at it from that way, the closest object to the player (LOD 10) will have the largest number of triangles, and objects that are considerably far away from the player (LOD 4, 2, 1) will have FAR less triangles compared to even an LOD 6 object. When VRS comes into play, it's shading the low-tier objects (LOD 4, 2, 1) at a much smaller rate in order to save on CPU cycles, BUT, the vertices are still being processed for the lower-tier objects thus resulting in more draw calls from the CPU for objects the player can barely even see. The point Matt Hargett's trying to make is, if we cull the geometry processing (stop drawing triangles/processing vertices) for the lower-tier objects, then you're doing the CPU a huge favour by saving on unnecessary cycles AND draw calls, so he's talking about VRS and geometry culling combined (They're two different things). So as you're getting closer and closer to the low-tier object, that same object is transitioning to a high-LOD object as geometry's being instantly synthesized by the Geometry Engine block on the GPU, while the objects behind you (off-screen) are slowly transitioning to a low-LOD object to the point where (let's say starting at LOD 4) geometry processing is being culled at the same time geometry processing is going on for the objects in front of you as you're getting closer to them.
During The Road to PS5 stream, Mark Cerny was talking about how simple usage of the Geometry Engine means only focusing on culling geometry processing off-screen and how complex usage meant utilizing primitive shaders (most likely both processing AND culling at the same time). So I'm pretty sure doing both are possible with the Geometry Engine.
Please correct me if I'm wrong on anything.
Last edited:
cragarmi
Member
Indeed, this can also reduce Dev time, as you only need to create one object, rather than several variants of the same one, for LOD transition. This should be a powerful tool going into next gen, for both MS and Sony.Why can't it be both? This is how I understood it since I wasn't familiar with the tech before, I watched NVIDIA's mesh shading video and their approach is based on LOD, which is what PS5's primitive shaders are based on as well.
So if you look at it from that way, the closest object to the player (LOD 10) will have the largest number of triangles, and objects that are considerably far away from the player (LOD 4, 2, 1) will have FAR less triangles compared to even an LOD 6 object. When VRS comes into play, it's shading the low-tier objects (LOD 4, 2, 1) at a much smaller rate in order to save on CPU cycles, BUT, the vertices are still being processed for the lower-tier objects thus resulting in more draw calls from the CPU for objects the player can barely even see. The point Matt Hargett's trying to make is, if we cull the geometry processing (stop drawing triangles/processing vertices) for the lower-tier objects, then you're doing the CPU a huge favour by saving on unnecessary cycles AND draw calls, so he's talking about VRS and geometry culling combined (They're two different things). So as you're getting closer and closer to the low-tier object, that same object is transitioning to a high-LOD object as geometry's being instantly synthesized by the Geometry Engine block on the GPU, while the objects behind you (off-screen) are slowly transitioning to a low-LOD object to the point where (let's say starting at LOD 4) geometry processing is being culled at the same time geometry processing is going on for the objects in front of you as you're getting closer to them.
During The Road to PS5 stream, Mark Cerny was talking about how simple usage of the Geometry Engine means only focusing on culling geometry processing off-screen and how complex usage meant utilizing primitive shaders (most likely both processing AND culling at the same time). So I'm pretty sure doing both are possible with the Geometry Engine block on the GPU.
Please correct me if I'm wrong on something.
FranXico
Member
It also eliminates another source of storage bloat (alongside duplication intended to reduce seek times).
Handy Fake
Member
Aye, that's how I understand it too.
My brother does this 3d modelling witchcraft and - again as I understand it - he'd start off with a high quality model and then reduce it in polygons* down to the limit given to him by the requirements of his task.
*if this is outdated terminology, I'm old.
Last edited:
Thirty7ven
Banned
So is the GE as explained by Mark Cerny in the road to PS5 a custom bit that Sony worked with AMD and is "exclusive" to one of the consoles and later PC Cards?
I ask because back then GE was said to be a completely standard feature, yet we are now learning that it's not the same as before. I don't want to go the way of "MS didn't mention in therefore they don't have it", just trying to understand.
I ask because back then GE was said to be a completely standard feature, yet we are now learning that it's not the same as before. I don't want to go the way of "MS didn't mention in therefore they don't have it", just trying to understand.
Shmunter
Member
Mesh/primitive shaders are baked into rdna2 and different to the geometry engine. Seemingly only Sony has the geometry engine.
3liteDragon
Member
I could be flat out wrong on this but I think it's custom to AMD's PC cards and Sony. Sony probably came up with the GE, cause if it was made by AMD, it would've been a standard RDNA 2 feature and in that case, both the PS5 and the Series X can it.
Cerny did also mention that some of the features they came up with could be used in AMD'S upcoming PC GPU's and vice versa.
Last edited:
Shmunter
Member
Could be an RDNA3 feature
SgtCaffran
Member
I haven't seen any confirmation that the PS5 GE is anything more than mesh/primitive shaders, have you? I'm actually keen to learn even more about these PS5 features.
Shmunter
Member
Anything is possible but why would they 'rebadge' a stock feature? Did they describe it doing like for like with mesh shaders? Can't remember.
THE:MILKMAN
Member
The extreme streamlining/efficiency/de-bottlenecking and supercharged I/O of these next-gen consoles is the real secret sauce.
Can anyone estimate how much die area the I/O block in the PS5 APU might take up and how it compares to other examples?
Can anyone estimate how much die area the I/O block in the PS5 APU might take up and how it compares to other examples?
ZywyPL
Banned
THE:MILKMAN
Member
If Cerny's slides are any indicator, about 2/3rds of the APU:
That is why I was asking because surely those graphs were nowhere near to scale or the APU would be >500mm2!
It definitely takes a relatively significant chunk of die space for sure. I could see the final die size close to the OG PS4's (348mm2) and not much less than XSX' 360mm2.
M1chl
Currently Gif and Meme Champion
~36USD just for normal NVMe SSD, stacked chip would be at least 10x, see HBM2 memory.
If Cerny's slides are any indicator, about 2/3rds of the APU:
Just 3900x that is without any GPU, just a chiplets and that SoC, which are memory controllers are really huge.
Last edited:
Gamernyc78
Banned
Netbooks are fine as long as you don't use Windows which can bring anything to its knees because Microsoft thinks that "optimization" is a term for accountants, not programmers.
In-engine may mean from PS4 but not in real time. Engine renders frames at any pace and you put them together to form a super-smooth video. But do you really doubt in Sucker Punch? Anytime in their release history when they didn't deliver a well made game?
ICE tech is shared between all proprietary studios, not only those that have ICE members. This means ICE team can help any developer who makes games for Playstation. For example, my friend's small studio got help from Santa Monica for their project and they also published it.
If you want to learn more about call draws and how the rendering pipeline works, there's a quite approachable presentation made not for programmers but artists, so less techno jargon. In general, call draws are a real pain in ass of realtime graphics and their optimization means a lot more than any teraflop count. Please bear in mind that the presentation is for traditional rendering setups, i.e. PCs and new consoles are going to cut a lot of corners in that process.
Render Hell 2.0
🇨🇳 Chinese Version by Leon He 🇻🇳 Vietnamese Version by Vu Phuong Hoang Added a whole new book covering the pipeline in detail Added 2 new videos and 32 new links to great articles, whitepapers, ... Extended the section copying data from HDD to graphicsimonschreibt.de
Thnks for the info. Also trust me I'm well aware that ice team was created to be the goto specialists amongst all Sony game studios. I just know that some people on Ice are made up of one or two people from those top studios such as guerilla and Naughty. As I stated if anyone can pull it off it's one of those dev studios as they are the cream of the crop so much so individuals from those studios have been tasked with being on Ice and helping others in time of need.
geordiemp
Member
Just 3900x that is without any GPU, just a chiplets and that SoC, which are memory controllers are really huge.
The IO is 12 nm vs 7nm for the chiplet (CPU cores)
ZywyPL
Banned
As per below:
Yup, the CCX themselves are made with 7nm process, but the controllers are still 14nm, hence their size. But your post made me wonder - maybe Sony will indeed opt for such chiplet design to cut the production costs as much as possible, whereas MS went with traditional monolithic design?
THE:MILKMAN
Member
I think it is clearly a monolithic chip as it even says 'main custom chip'. I think the graph just separates each section out for clarity.
Saying that I guess it can't be ruled out but I would be really surprised if it is a chiplet design.
Agnostic2020
Member
Didn't cerny in road to ps5 say big chunk of apu is devoted to IO which has equivalent of 11 zen 2 cpu cores working to ensure main cpu is not needed and run its game task at peak all the times ?
Last edited:
FranXico
Member
Conceptual diagrams are not drawn to scale. WTHIf Cerny's slides are any indicator, about 2/3rds of the APU:
I do remember him saying that the co processor for the I/O block is the equivalent to a certain amount of zen 2 cores
M1chl
Currently Gif and Meme Champion
Sure, but it still would be at least size of the one of CCX. It's not insignificant.
Well we see, would be smart, but we don't know how much into development these consoles started when chiplets came along.
ResilientBanana
Member
But the GPU can only produce 10TF of information so if you have LOD 10 objects in the background, surely they would load quickly off the SSD but now you have 5 MIllion+ triangles for every asteroid on screen which doesn't make sense and would make your game engine grossly inefficient. The point of the LOD system isn't because it's difficult pulling up assets, but to ease the strain of the GPU on distant objects. THe only place the SSD has in all this is changing that object from a LOD1 to a LOD10 depending on how close it is to the player. Having an SSD doesn't magically give the PS5 more GPU power to display more triangles, it simply allows the loading of many different types of assets quickly, but you're still held back by the GPU and what it can render.
If this is true than Nvidia would basically double their CUDA Cores and Tensor Cores and even tripple their RT-Cores and additionaly boost the clock speed with 200MHz. And than there is the die shrink from 12nm to 7nm. That would be a massive performance increasement
ethomaz
Banned
Geometry Engine is an standard feature since GCN cards because the render pipeline needs to pass thought the Geometry Engine.
Last edited:
TheAmazingSpiderMan
Member
unfortunately whatever this guy reports it's always the opposite of what he says. I don't think anything he's said in the past have ever come true.
Agnostic2020
Member
Ssd is not the main thing here. The io complex makes data access instant for the system and remove bottlenecks like gpu flushing of all caches which lowers gpu performance . Ultimately it means cpu and gpu can run at their peak performance . And yes it magically doesnt mean it has 12 tf now. But it can run at 10.2 tf much more often than say a solution that doesn't have this io complex and bottlenecks cpu with flood of instant data and gpu with full cache flushing .
Thirty7ven
Banned
Geometry Engine is an standard feature since GCN cards because the render pipeline needs to pass thought the Geometry Engine.
Then why the focus? This is so weird.
- Status