Support NeoGAF

[Red_Driver85]

Sony went for a clock budget, which is also shared between CPU and GPU, based on activity counters and MS wanted fixed clock rates.

Hm, maybe power budget?

 

[MadchesterManc]

Why do some folk assume that if these features are not done in the same manner as XBOX & AMD then it can't possibly be supported? Turing has been on the market since 2018 with hardware support for the full DX12U feature set, yet I'm quite sure Nvidia are doing these things their own way and not following AMD doctrine. Isn't it feasible that this could also be possible of Sony too considering they seem to have diverged on their GPU design earlier?

Food for thought at least, no one here really knows anyway

 

[Lysandros]

 

[longdi]

thats the thing, no?
rDNA2 have sw vrs and primitive shaders, and it is good ps5 and SX can make use of them.
primitive shaders iirc was already in rx vega

But only SX have addiitonal hardware vrs and mesh shaders.

 

[FatKingBallman]

1.) The PS5 likely doesn't support the hardware based VRS solution which the Xbox Series and RDNA2 include.
It's possible to implement similar techhniques in software, even quite efficient or with advantages but it's not universally better and is already achieved on the last gen consoles.
Further one can use and combine both solutions, the Xbox Series has this option the PS5 likely hasn't.
How large is the advantage? Nobody knows without benchmarks, so claiming it's in general much better, the same or worse is pure fantasy without good data and arguments backing it up.

2.) I said it already on this forum but you really need to know the specifications and features from "Primitive Shaders" on the PS5 to know what's the difference, if there is any, in comparison to "Mesh Shaders".
Those terms are just arbitrary names otherwise.
Personally I don't think there is a significant difference.

3.) You can support Machine Learning on nearly every advanced processor but you can't train or execute ML networks fast enough for real time rendering on every hardware.
There are many ML models which don't need high precision mathematics, so one can use FP16, INT8, INT4 or even INT1.
What's important for real time rendering is that the throughput is high enough.
It wouldn't be surprising to me if the PS5 only has packed math for FP16 and INT16, that's quite a bite worse than the mixed precision dot-product instructions which the Xbox Series and RDNA2 GPUs can offer.
They can multiply 2xFP16, 4xINT8 or 8xINT4 data elements and add the results to a FP32 or INT32 accumulator in one execution step.
This is more precise and/or a lot faster than just working with single precision and packed math operations.


You can precisely tell that the Floating-Point Register File was cut in half, that's enough to know that the PS5 can't execute FP256-Instructions as fast as vanilla Zen2 cores.
What's harder to tell or impossible is, if and in what way the digital logic on the execution paths was cut.
I think Nemez's fifth tweet is talking about theoretical TFLOPs and that in real world terms more factors are important, putting both closer together in practise.

The raster pipeline on the PS5 is from a high level view structured as on RDNA1 GPUs.
This may be even a performance advantage, since AMD's own RDNA2 GPUs rebalanced execution resources and made some cut downs.
They have less Primitive Units and Depth ROPs per Shader Engine.
From a Compute Unit perspective, for most operations, there isn't a real difference between RDNA1 or RDNA2.

Smart Shift is basically "firmware magic".
Since multiple generations AMD has the necessary hardware built in to control and set voltage levels, clocks, thermal limits, etc.
It's all very programmable and can be set as desired.
Sony went for a power budget, which is also shared between CPU and GPU, based on activity counters and MS wanted fixed clock rates.
AFAIK the way Sony is doing it, is very similar to how AMD first implemented variable frequencies under Cayman (6970/50, 2010/11 GPUs), where AMD opted for no clock differences between the same GPU SKUs.
That's different now, AMD is using real power and thermals measurements and let's each chip optimally behave under his unique attributes.
Every chip is different in quality, they all behave a bit differently, something you certainly don't want for a gaming console which is why every PS5 is modeled after a reference behaviour based on activity counters.

That's a high level diagram from GCN2, showing how the power/thermal/clock control is laid down, with programmable software layers.

Thanks for the thorough explanation

Good job! GOT EMMMMMMMMMM!

I'm not native english speaker nor i don't understand english phrases. But what you said, this sounded like he owned me. Actually he owned me in thorough explanations of some things, but Locuza basically confirmed what i've said, what others saying too : PS5 does support VRS ( so what if it is software based but it works great, better than HW in some cases ), Machine Learning ( if PS4 did support it, why PS5 wouldn't, also there are lots of patents for it ). and Mesh Shading/Primitive Shaders which you, people from green camp so furious trying to negate. I trust Locuza ( i believe others too ),, not because it fits my narrative, but because he is very level headed guy, unlike others here. I'm quite happy that supposedly a "much weaker" machine outperforms a stronger one. So much that there was a need for articles and excuses why XSX failed to show it.

Why is the PS5 outperforming the 'world's most powerful console'?

PS5 beating Xbox Series X in performance — here's why

 

[Locuza]

This is how I got something described - please note that I am an interested layman here and not an expert - nor do I sit on insider knowledge but I have had talks with someone that works in the PS5 development environment. The functionalities of the GE in PS5 (what is transistor, i.e. hardware driven, and what is software driven is unknown to me) allows for culling as well as merging of geometries. These two function as such create a similar functionality as mesh shaders in terms of culling but it happens earlier in the pipeline as well as VRS by allowing for control of the number of pixels that enters the final shader step through merging of geometries (i.e. the same functionality that VRS plays in Nvidia and current AMD cards).

It is not the same as the mesh shader and VRS but the functionality is the same.

This might of course be erroneous information but would be interested in what you have heard about the same. For me it lines up fairly well with the Cerny talk so I have in general taken the information at face value.

I'm not sure if the PS5 hardware can cull geometry earlier than RDNA1/2 from AMD and the Xbox Series.
There is fixed function hardware which can waste time and energy on assembling geometry data which gets culled right after and there are programmable options to cull geometry earlier, be it indirectly through the software compiler or explicitly through the game developer using Mesh/Primitive-Shaders.

But the world space pipeline (usually handling 3D geometry in a virtual space) isn't providing VRS style control, it's a step happening before it:

Wheter it's fixed function hardware or programmable Shaders like Vertex-, Task&Mesh-Shaders, geometry data is generated and handed of to the fixed-function Rasterizer.
This HW unit generates the corosponding pixel fragments based on the supplied geometry data (with the rasterizer the screen space pipeline starts, which transform the virtual/world space data to a 2D screen which your monitor displays).
HW VRS is now offering the ability to control the shading rate of that pixel surface, instead of shading each pixel individually it's possible to shade just 2x2 pixels or other combinations (RDNA2 supports just a few, for example 4x4 is not supported but in practise it appears not to be very useful anway).

Hm, maybe power budget?

I changed it to power budget, which I also wrote originally but later on I thought clock budget would describe it more directly to some.
Anyway, I hope the explanation was more or less clear. ^^

 

[Fafalada]

No - it happens at the rasterization stage. There's no two ways about it.

Define 'happening'.
Rasterizer generates pixel/sample quads, but the sampling parameters based on that screenspace mask have to come from somewhere first.

 

[Elog]

I'm not sure if the PS5 hardware can cull geometry earlier than RDNA1/2 from AMD and the Xbox Series.
There is fixed function hardware which can waste time and energy on assembling geometry data which gets culled right after and there are programmable options to cull geometry earlier, be it indirectly through the software compiler or explicitly through the game developer using Mesh/Primitive-Shaders.

But the world space pipeline (usually handling 3D geometry in a virtual space) isn't providing VRS style control, it's a step happening before it:

Thanks for this! Could you comment on Matt's twitter here in the context of your first point? Is he not in essence claiming that they have added logic to the GE in the PS5 in light of how you are describing a more normal GE function above? He is a good guy and was part of the PS5 development team.

 

[SenjutsuSage]

The question is not what they thought, but how they actually did it. The implementation of data management in their systems is quite different, for example. Whose approach will be better - we will see over time.

This is probably going to be controversial to some, but I've thought this from the get go. I think Xbox Velocity Architecture is superior to Sony's setup overall, Sony's setup just appears easier to utilize since it seems to hinge more on raw SSD speed backed up by kraken decompression to seemingly meet its potential, and "just works" according to Cerny I guess once you use it, whereas for Series X in addition to a fast SSD and hardware decompression, with their own gpu texture specific compression, Sampler Feedback Streaming appears a more complete rethinking of how GPUs handle memory, and as such it's a more complex (shouldn't be confused to mean overly complex) endeavor than Sony's setup to design into a game.

I think Sony needed that nearly two times faster SSD just to keep up. I believe Xbox Velocity Architecture is superior, but more work to get fully implemented into a game. Sony's approach I think is less new in its thinking overall, but is likely to be much easier to just get operation for a game because nothing related to Sony's I/O architecture requires as significant a direct change to game code as I believe Sampler Feedback Streaming does.

 

[SenjutsuSage]

I beg to differ.

With more info coming out, it seems MS was right about their earlier chest beating.

SX is the more graceful, thoughtfully customed soc whilst ps5 feels more like just using the basic Amd rDNA2 IP but with a big chunk of their custom I/O block attached to it.

Even from the exterior designs, SX just seemed to be assembled more gracefully than ps5.

Mostly agree. Xbox Series X just plain looks better designed overall than Playstation 5 based on all information presented from official sources. This isn't to be taken that PS5 is a bad system or anything, but it doesn't appear anywhere as thoughtfully constructed as Xbox Series X has been from a SoC architecture viewpoint. Xbox Series X just plain seems a lot more advanced. The most impressive looking piece of PS5 is it's SSD and overall I/O architecture. All else seems very plain jane.

 

[Physiognomonics]

This is probably going to be controversial to some, but I've thought this from the get go. I think Xbox Velocity Architecture is superior to Sony's setup overall, Sony's setup just appears easier to utilize since it seems to hinge more on raw SSD speed backed up by kraken decompression to seemingly meet its potential, and "just works" according to Cerny I guess once you use it, whereas for Series X in addition to a fast SSD and hardware decompression, with their own gpu texture specific compression, Sampler Feedback Streaming appears a more complete rethinking of how GPUs handle memory, and as such it's a more complex (shouldn't be confused to mean overly complex) endeavor than Sony's setup to design into a game.

I think Sony needed that nearly two times faster SSD just to keep up. I believe Xbox Velocity Architecture is superior, but more work to get fully implemented into a game. Sony's approach I think is less new in its thinking overall, but is likely to be much easier to just get operation for a game because nothing related to Sony's I/O architecture requires as significant a direct change to game code as I believe Sampler Feedback Streaming does.

This is not controversial. This is delusional.

 

[FatKingBallman]

This isn't to be taken that PS5 is a bad system or anything,

But you are always trying to paint PS5 into a bad light in every possible way and category where the is no way that PS5 can have a single advantage over XSX. Always. You saying that PS5 isn't a bad system it's like you trying to play a nice, but you actually have a poker face.

Xbox Series X just plain seems a lot more advanced.

It is the best of the best, bestest, and beyond everything. So much beyond that i'm pretty lost in speaking the best epithets for XSX. :/

 

[longdi]

Mostly agree. Xbox Series X just plain looks better designed overall than Playstation 5 based on all information presented from official sources. This isn't to be taken that PS5 is a bad system or anything, but it doesn't appear anywhere as thoughtfully constructed as Xbox Series X has been from a SoC architecture viewpoint. Xbox Series X just plain seems a lot more advanced. The most impressive looking piece of PS5 is it's SSD and overall I/O architecture. All else seems very plain jane.

yep seems that way. i think when cerny was incepting the ps5 at the early stages, he probably looked at what amd has instore, and simply work with that.

besides the I/O, the next big hype of ps5 is the GE. from what we know, it's pretty much the primitive shaders amd intro with vega. and said primitive shaders fall in the time line with what i said, cerny took it and further thought how to make better use of it, to not be limited by what happened with vega/PC adoption, and that's probably how we got that I/O IP and resulting micro polygons rendering.

so to say it just feels like sony selects and takes what IP amd has to sell, while MS co-worked with amd to develop new IP.

 

[Panajev2001a]

Mostly agree. Xbox Series X just plain looks better designed overall than Playstation 5 based on all information presented from official sources. This isn't to be taken that PS5 is a bad system or anything, but it doesn't appear anywhere as thoughtfully constructed as Xbox Series X has been from a SoC architecture viewpoint. Xbox Series X just plain seems a lot more advanced. The most impressive looking piece of PS5 is it's SSD and overall I/O architecture. All else seems very plain jane.

Considering actual performance, how critical the I/O revolution is for both consoles (especially to alleviate concerns from devs who would have wanted more RAM), and how much tweaking throughout the system you need to ensure you are not creating bottlenecks for yourself I am not sure where you are taking this idea from.

It is your personal view that XSX is a lot more advanced and PS5 is a plain jane system nowhere near as thoughtfully constructed as XSX . It seems that what you do not know about PS5 must be worse, but that is also how they talk about tech: look at the controller and latency and accuracy improvements. They have accomplished quite a lot without shouting anywhere near as much as MS did. I see a console with clear forward looking goals and understanding of their cost, performance, power consumption, and developer experience constraints... and those are reflected in the customisations they performed (that we know of), the components they selected, the transistors budget they dedicated to what, and their clocking strategy especially for the GPU (some people still see it as a last minute overclock though, but that is them...).

 

[Panajev2001a]

yep seems that way. i think when cerny was incepting the ps5 at the early stages, he probably looked at what amd has instore, and simply work with that.

besides the I/O, the next big hype of ps5 is the GE. from what we know, it's pretty much the primitive shaders amd intro with vega. and said primitive shaders fall in the time line with what i said, cerny took it and further think how to make the best use of it, to not be limited by what happened with vega/PC environment, and that's probably how we get that I/O IP and micro polygons rendering.

so to say it just feels like sony took what amd has to offer, while MS co-work with amd to develop new IP.

I think you are being very generous with one party (as if MS invented VRS and SF) and a bit condescending with the other side, but I do not get that. It should be enough to sing the praises of your HW of choice that excite you without having to try to widen the gap, just makes people look more insecure than they are (not saying you are).

 

[Red_Driver85]

Sony's setup just appears easier to utilize since it seems to hinge more on raw SSD

I think Sony needed that nearly two times faster SSD just to keep up.

Sony's approach I think is less new in its thinking overall

nothing related to Sony's I/O architecture requires as significant a direct change to game code as I believe Sampler Feedback Streaming does.

I thought this was a technical thread. You are so wrong and you write such nonsense, but it doesn't even bother you? Really? I don't know what else to say, but any technical conversation with you is useless and pointless.

 

[Panajev2001a]

This is not controversial. This is delusional.

Also because developers dealing with SSD's are very focused on much bigger problems than pure HW speed but how to actually make use of these low level API's and rethink how they store geometry, textures, sound assets, etc... and stream them in and out of the SSD (see the massive work UE5 is doing in this area).

SF and SFS (the latter sees purely texture focused work) are good evolutions, but they are being oversold IMHO almost making it look like Sony panicked and packed together as many parallel flash chips to somehow brute force that bandwidth in. SF support in the GPU is now replacing all the effort Sony did from the flash memory arrangement, custom SSD controller and additions priority levels, to the CPU's and memory and other co-processors they dedicated to enable efficient data transfer and reduce memory bandwidth and cache miss issues.

 

[longdi]

I think you are being very generous with one party (as if MS invented VRS and SF) and a bit condescending with the other side, but I do not get that. It should be enough to sing the praises of your HW of choice that excite you without having to try to widen the gap, just makes people look more insecure than they are (not saying you are).

But it does feel that way, no? What does your instincts tell you after more info is out?

I do think sony' tech team need credits for their super sdk/api. Imo it is also the same 'not limited by vega/PC' problems that mark cerny wanted to solve for developers.

 

[longdi]

Also because developers dealing with SSD's are very focused on much bigger problems than pure HW speed but how to actually make use of these low level API's and rethink how they store geometry, textures, sound assets, etc... and stream them in and out of the SSD (see the massive work UE5 is doing in this area).

SF and SFS (the latter sees purely texture focused work) are good evolutions, but they are being oversold IMHO almost making it look like Sony panicked and packed together as many parallel flash chips to somehow brute force that bandwidth in. SF support in the GPU is now replacing all the effort Sony did from the flash memory arrangement, custom SSD controller and additions priority levels, to the CPU's and memory and other co-processors they dedicated to enable efficient data transfer and reduce memory bandwidth and cache miss issues.

I do not think sony panicked per se, but they chose a brute force approach around primitive shaders. Cerny must have liked primitive shaders, which was a new IP of rx vega, but totally unusable because of the PC environment back then.

 

[Panajev2001a]

I dont think sony panicked per se, but they just chose brute force approach around primitive shaders back then.

I disagree, but let the one with a much bigger die and many more CU's speak up .

 

[x@3f*oo_e!]

Define 'happening'.
Rasterizer generates pixel/sample quads, but the sampling parameters based on that screenspace mask have to come from somewhere first.

If you roll back this thread you'll see I was responding to someone who straight out stated on twitter 'geometry engine VRS' quote : "Is the geometry engine VRS done by SW or HW?" .. '"Geometry Engine is doing it via SW and sometimes can beat the HW"

..if you really want to defend that, or show how it's possible for that statement to be true then do.

 

[Snake29]

Mostly agree. Xbox Series X just plain looks better designed overall than Playstation 5 based on all information presented from official sources. This isn't to be taken that PS5 is a bad system or anything, but it doesn't appear anywhere as thoughtfully constructed as Xbox Series X has been from a SoC architecture viewpoint. Xbox Series X just plain seems a lot more advanced. The most impressive looking piece of PS5 is it's SSD and overall I/O architecture. All else seems very plain jane.

When the competition is beating the so called "most powerful console", And you have to do everything in your power to make it right for yourself.

 

[Panajev2001a]

But it does feel that way, no? What does your instincts tell you after more info is out?

My understanding is not changing much from the Road to PS5 to be honest, I was not prepared for the two consoles to be so close and for PS5's BC not to be anything but very basic... so pleasantly surprised .

I did gain more appreciation for how MS assembled and designed their consoles and the pragmatic choices they made. Some of their fans have this obsession with winning all of the metrics wars by a mile and act as if their designers had no limitations, but hey... .

 

[Garani]

This is probably going to be controversial to some, but I've thought this from the get go. I think Xbox Velocity Architecture is superior to Sony's setup overall, Sony's setup just appears easier to utilize since it seems to hinge more on raw SSD speed backed up by kraken decompression to seemingly meet its potential, and "just works" according to Cerny I guess once you use it, whereas for Series X in addition to a fast SSD and hardware decompression, with their own gpu texture specific compression, Sampler Feedback Streaming appears a more complete rethinking of how GPUs handle memory, and as such it's a more complex (shouldn't be confused to mean overly complex) endeavor than Sony's setup to design into a game.

I think Sony needed that nearly two times faster SSD just to keep up. I believe Xbox Velocity Architecture is superior, but more work to get fully implemented into a game. Sony's approach I think is less new in its thinking overall, but is likely to be much easier to just get operation for a game because nothing related to Sony's I/O architecture requires as significant a direct change to game code as I believe Sampler Feedback Streaming does.

Ok, folks, I think we can close the thread.

 

[quest]

Considering actual performance, how critical the I/O revolution is for both consoles (especially to alleviate concerns from devs who would have wanted more RAM), and how much tweaking throughout the system you need to ensure you are not creating bottlenecks for yourself I am not sure where you are taking this idea from.

It is your personal view that XSX is a lot more advanced and PS5 is a plain jane system nowhere near as thoughtfully constructed as XSX . It seems that what you do not know about PS5 must be worse, but that is also how they talk about tech: look at the controller and latency and accuracy improvements. They have accomplished quite a lot without shouting anywhere near as much as MS did. I see a console with clear forward looking goals and understanding of their cost, performance, power consumption, and developer experience constraints... and those are reflected in the customisations they performed (that we know of), the components they selected, the transistors budget they dedicated to what, and their clocking strategy especially for the GPU (some people still see it as a last minute overclock though, but that is them...).

Depending on how you look at it. Sony and Microsoft each had different goals. Sony wanted a bad ass io at all costs and since they have a closed system and their first party only release on playstation platforms they could use proprietary stuff. Microsoft wanted enough io to feed the apu but any gains had to be transferred to another platform the pc. Since their first party release games on 2 platforms day 1. So any uber fast proprietary io was out the window because of trying for adoption with multiple hardware vendors. Sony did a incredible job making a super fast low latency io. They also didn't have to worry about any other platforms. Microsoft did a bang up job to bring lots of improvements to 2 platforms working with many different vendors who have their own input. That is not an easy task.

 

[Panajev2001a]

Depending on how you look at it. Sony and Microsoft each had different goals. Sony wanted a bad ass io at all costs and since they have a closed system and their first party only release on playstation platforms they could use proprietary stuff. Microsoft wanted enough io to feed the apu but any gains had to be transferred to another platform the pc. Since their first party release games on 2 platforms day 1. So any uber fast proprietary io was out the window because of trying for adoption with multiple hardware vendors. Sony did a incredible job making a super fast low latency io. They also didn't have to worry about any other platforms. Microsoft did a bang up job to bring lots of improvements to 2 platforms working with many different vendors who have their own input. That is not an easy task.

MS designed a brilliant system overall, this generation has some common goals (both decided that expanding RAM was not the way to go anymore for example and a lot followed from that and other decisions), but very interesting differentiating traits too.

We all have different ideas and some preferences in terms of what we see and understand of the HW at hand, but both consoles deliver HW their designers should be proud of.

 

[Red_Driver85]

I do not think sony panicked per se, but they chose a brute force approach around primitive shaders.

 

[longdi]

I disagree, but let the one with a much bigger die and many more CU's speak up .

Imo its a matter of time. But im not sure apu die size is the be all.

Iirc ps4p-onex have a smaller die size difference than ps5-sx.

how many trannies does ps5 have?

also from these ps5 die shots, a rather big part of them are unidentified, possibly its the I/O IP.

i wonder if it seems unbalanced by committing this big a portion for just the custom I/O?

 

[REDRZA MWS]

It's not a gotcha posting, just my commentary you may agree or disagree with.
Since I'm just a layman I may be also wrong in multiple regards, corrections are welcomed.

I'm not here for fanboy wars.

Me neither, at all. It's the most ridiculous thing in this hobby. I'm a GAMER. Was just applauding someone actually taking their time to even debate this insanity. I could care less, own PS5, series x, and a solid pc.

 

[longdi]

Depending on how you look at it. Sony and Microsoft each had different goals. Sony wanted a bad ass io at all costs and since they have a closed system and their first party only release on playstation platforms they could use proprietary stuff. Microsoft wanted enough io to feed the apu but any gains had to be transferred to another platform the pc. Since their first party release games on 2 platforms day 1. So any uber fast proprietary io was out the window because of trying for adoption with multiple hardware vendors. Sony did a incredible job making a super fast low latency io. They also didn't have to worry about any other platforms. Microsoft did a bang up job to bring lots of improvements to 2 platforms working with many different vendors who have their own input. That is not an easy task.

Thats pretty much how i felt, but elegantly put.

MS and Amd continues the progress of creating new IP that needs to be more advanced and efficient, because PC environment is not really efficient.

Sony just took what was previously created and try to hull down and squeeze the potential out from it.

 

[Garani]

i wonder if it seems unbalanced by committing this big a portion for just the custom I/O?

Under what standard does the PS5 APU seem unbalanced? Actually it's the one that is closer to AMD's GPU layout and design.

 

[darthvaderer]

No, you beat your chest after you deliver the goods... Some specs sheet is nice and all, but what counts is the output.

No receipts no chest pumping.

That is fanboy talk.

I kind of agree with you on that front.

I think exactly the same things as you about that post

 

[TheThreadsThatBindUs]

Well this thread went to shit fast.

 

[Clear]

If you look at how MS are presenting their stuff what you see are DX12 API elements and Velocity API elements, they have to split them because when they are rolled out onto PC they need to be able to handle scenarios where either or both cannot be completely supported fully. For instance say the user has only an older, slower SSD but a modern but low-to-mid-range GPU.

Sony's advantage is that they don't need to worry about any of this beyond disbarring the use of SSD's that cannot match or exceed the onboard drive's performance. They know what the minimum capabilities are on both sides and can balance based on assurance of that level of performance.

 

[REDRZA MWS]

Thanks for the thorough explanation



I'm not native english speaker nor i don't understand english phrases. But what you said, this sounded like he owned me. Actually he owned me in thorough explanations of some things, but Locuza basically confirmed what i've said, what others saying too : PS5 does support VRS ( so what if it is software based but it works great, better than HW in some cases ), Machine Learning ( if PS4 did support it, why PS5 wouldn't, also there are lots of patents for it ). and Mesh Shading/Primitive Shaders which you, people from green camp so furious trying to negate. I trust Locuza ( i believe others too ),, not because it fits my narrative, but because he is very level headed guy, unlike others here. I'm quite happy that supposedly a "much weaker" machine outperforms a stronger one. So much that there was a need for articles and excuses why XSX failed to show it.

Why is the PS5 outperforming the 'world's most powerful console'?

PS5 beating Xbox Series X in performance — here's why

Was never about who got "owned". Good luck with that. Keep stanning, great warrior! I hope you aren't over the age of 12. Arguing about which plastic encased bits are better than another is ridiculous.

 

[Locuza]

Thanks for this! Could you comment on Matt's twitter here in the context of your first point? Is he not in essence claiming that they have added logic to the GE in the PS5 in light of how you are describing a more normal GE function above? He is a good guy and was part of the PS5 development team.

Yes, you may come to that conclusion.
But it's hard to tell without further explanation.

When I paraphrase his statement a bit, he is actually saying that VRS is a nice tool to reduce shading cost and to increase performance but it doesn't come close to the potential performance benefits of the Geometry Engine.

Let's construct a hypothetical performance case.
We are going to use VRS for our game and we get 5-10% better performance, depending on the scene.
Now on the other console we don't have VRS but the Geometry Engine is handling geometry processing more efficiently, less geometry data ends up being processed in further stages.
Performance may be over 15% better.

The question is now, has Sony really a customized Geometry Engine which is better than on the Xbox Series and AMD's vanilla RDNA1/2 Hardware?
If so what is required for the performance benefits?
Does it work automatically or do developers have to use it explicitly?
Are those only fixed-function unit improvements for the legacy geometry pipeline or do they only(or also) apply to the new and more programmable geometry pipeline?

WIthout further transparency it's hard to judge.

 

[Red_Driver85]

Sony just took what was previously created and try to hull down and create some more from it.

 

[StreetsofBeige]

Going by the recent posts looks like PS5 has no HW VRS, which is what people have been saying from the beginning.

 

[DJ12]

Thats pretty much how i felt, but elegantly put.

MS and Amd continues the progress of creating new IP that needs to be more advanced and efficient, because PC environment is not really efficient.

Sony just took what was previously created and try to hull down and squeeze the potential out from it.

You sir, really have no clue what you're talking about. Rumour had it, Sony and AMD basically collaborated on RDNA from the start. Kind of backed up by this quote from Cerny.

"After confirming that Sony's PS5 will house "a custom AMD GPU based on their RDNA 2 technology," Cerny added, "if you see a similar discrete GPU available as a PC card at roughly the same time as we release our console, that means our collaboration with AMD succeeded in producing technology useful in both worlds. It doesn't mean that we at Sony simply incorporated the PC part into our console.

Hmm, whats coming out soon? Oh yeah, a 40 CU high clocked AMD desktop part....

 

[jroc74]

so to say it just feels like sony selects and takes what IP amd has to sell, while MS co-worked with amd to develop new IP.

How did this get flipped? I swore some folks were claiming it was the other way around...

Just wanna know what angle to debate from, its getting so confusing now.

What in the world happened...things seems like a complete 180 now.

Another narrative shift? Maybe my reading comprehension is bad...but from the die shot explanations from informed folks, Road to PS5 references....I don't see why it's shifting.

 

[thicc_girls_are_teh_best]

This is probably going to be controversial to some, but I've thought this from the get go. I think Xbox Velocity Architecture is superior to Sony's setup overall, Sony's setup just appears easier to utilize since it seems to hinge more on raw SSD speed backed up by kraken decompression to seemingly meet its potential, and "just works" according to Cerny I guess once you use it, whereas for Series X in addition to a fast SSD and hardware decompression, with their own gpu texture specific compression, Sampler Feedback Streaming appears a more complete rethinking of how GPUs handle memory, and as such it's a more complex (shouldn't be confused to mean overly complex) endeavor than Sony's setup to design into a game.

I think Sony needed that nearly two times faster SSD just to keep up. I believe Xbox Velocity Architecture is superior, but more work to get fully implemented into a game. Sony's approach I think is less new in its thinking overall, but is likely to be much easier to just get operation for a game because nothing related to Sony's I/O architecture requires as significant a direct change to game code as I believe Sampler Feedback Streaming does.

Posting iRoboto's post from B3D here; best summary on showing the apples-to-oranges approach Sony and Microsoft have taken. Neither is inherently superior or inferior to the other, they simply have different approaches suiting their larger product ecosystem and performance targets.

Not being RDNA 2 in itself doesn't make the system deficient however, it just makes it not RDNA 2. They both set out to solve the same problems, but their approaches are different.
a) Solving the memory cost problem. At a maximum of 16GB to keep costs down, they needed to find a way to increase fidelity while keeping within the limits of 16GB.
Sony:
5.5 GB/s SSD + Kraken Compression
This gets them 80% of the way there. The remaining 20%, the memory has to get into cache for processing. So what to do? Cache scrubbers to remove that last bit of random latency to go from textures of SSD directly into frame.

MS:
2.5 GB/ssd + BCPack Compression
This gets them 50% of the way there. The remaining 50%? Sampler Feedback Streaming, only stream in exactly what you need, further reducing what needs to be sent and when. That gets you 30% of the remaining way. You still got to get it through cache, instead of cache scrubbers their SFS will use an algorithm to fill in values for textures being called by SFS. If for whatever reason the texture has not arrived in frame, it will fill in the texture with a calculated value while it waits for the data to actually arrive within 1-2 frames later.

b) Approaching how we redistribute rendering power on the screen?
MS exposes VRS hardware
PS5 in this case will have to rely on a software variant

C) Improving the front end Geometry Throughput, Culling etc.
MS aligns features sets around Mesh Shaders, and XSX also supports (hopefully by now) Primitive shaders
PS5: Primitive shaders and whatever customizations they put towards their GE
* they are not the same, but comparable.

It's a general way of explaining their system designs in the I/O, and even GPU, CPU etc. departments, and how they're looking to achieve similar results with differing approaches.

Yes, you may come to that conclusion.
But it's hard to tell without further explanation.

When I paraphrase his statement a bit, he is actually saying that VRS is a nice tool to reduce shading cost and to increase performance but it doesn't come close to the potential performance benefits of the Geometry Engine.

Let's construct a hypothetical performance case.
We are going to use VRS for our game and we get 5-10% better performance, depending on the scene.
Now on the other console we don't have VRS but the Geometry Engine is handling geometry processing more efficiently, less geometry data ends up being processed in further stages.
Performance may be over 15% better.

The question is now, has Sony really a customized Geometry Engine which is better than on the Xbox Series and AMD's vanilla RDNA1/2 Hardware?
If so what is required for the performance benefits?
Does it work automatically or do developers have to use it explicitly?
Are those only fixed-function unit improvements for the legacy geometry pipeline or do they only(or also) apply to the new and more programmable geometry pipeline?

WIthout further transparency it's hard to judge.

Plus as a general rule, wouldn't geometry culling always be a bigger performance saver than a technique like VRS or even DLSS?

Any idea on what specific things Sony could have theoretically changed in the Geometry Engine to squeeze out additional performance?

 

[Panajev2001a]

MS and Amd continues the progress of creating new IP that needs to be more advanced and efficient, because PC environment is not really efficient.

Sony just took what was previously created and try to hull down and squeeze the potential out from it.

Again, can you appreciate one HW without shitting on the other? Are you seriously trying to just stir up cheap console warring / fishing for hit takes?

 

[martino]

Thanks for this! Could you comment on Matt's twitter here in the context of your first point? Is he not in essence claiming that they have added logic to the GE in the PS5 in light of how you are describing a more normal GE function above? He is a good guy and was part of the PS5 development team.

he is comparing gains done using VRS at pixel shader phase to gains you can have before rasterization using primitive shader (or mesh shader)
to be able to execute primitive shader you need change on geometry engine (especially when you are comparing your hardware to the previous generation like in road to ps5 for example)
Now show the tweet or quote where him or cerny claim geometry engine changes do more than adding primitive shader capabilities.
If they exist I've missed them.

 

[Elog]

Yes, you may come to that conclusion.
But it's hard to tell without further explanation.

When I paraphrase his statement a bit, he is actually saying that VRS is a nice tool to reduce shading cost and to increase performance but it doesn't come close to the potential performance benefits of the Geometry Engine.

Let's construct a hypothetical performance case.
We are going to use VRS for our game and we get 5-10% better performance, depending on the scene.
Now on the other console we don't have VRS but the Geometry Engine is handling geometry processing more efficiently, less geometry data ends up being processed in further stages.
Performance may be over 15% better.

The question is now, has Sony really a customized Geometry Engine which is better than on the Xbox Series and AMD's vanilla RDNA1/2 Hardware?
If so what is required for the performance benefits?
Does it work automatically or do developers have to use it explicitly?
Are those only fixed-function unit improvements for the legacy geometry pipeline or do they only(or also) apply to the new and more programmable geometry pipeline?

WIthout further transparency it's hard to judge.

Agree and thank you! If one goes into a conjecture it seems to me (please note - seems) that the way Cerny discussed handling of geometry in the Road to PS5 and how Epic discussed the handling of geometry in UE5 was fairly similar and a different way of handling the pipeline than what is done currently - both in hardware and in engines.

I might be reading too much into it but it seems to me as if the GE performs culling before additional rendering is performed (Epic referred to this as lossless, i.e. there must be a small rasterisation step of the geometry only to know what to cull to be that efficient), furthermore by knowing the rough outline of the rasterised image you are able to prioritise the geometry by basically putting your primitives in order of importance (which is very similar to the tier 2 VRS per-primitive rate that is the the Tier 2 VRS function that gives the most control to the programmer).

Maybe I am reading too much into this but this is how I have understood the PS5's handling of geometry - what of this that is a better software interface and what is hardware is something I have no clue about though

 

[Red_Driver85]

What in the world happened...

Some people are talking nonsense, simply because they can't do otherwise. This often happens from ignorance of the material/technical part, but in this case something more happened to this particular gentleman.

 

[SenjutsuSage]

Machine Learning
"More generally, we're seeing the GPU be able to power Machine Learning for all sorts of really interesting advancements in the gameplay and other tools."
Laura Miele, Chief Studio Officer for EA.
Source: https://www.wired.com/story/exclusive-playstation-5/

Two sources:
One from a random tweet on twitter that could have been photoshop by that @blueisviolet clown.
The other is an official statement from official people.
Sorry, I tend to believe official places before I believe anyone on twitter even if they claim to work at Sony.
Sony would have to tell me they work with them.

Here's the problem: see how far back you need to go just to pluck that out? Literally 2019. What confirmation has there been since then from anyone at Sony? Any of its architects? It's pretty weak evidence, whereas evidence on the Xbox side is pretty overwhelming. Why didn't Mark Cerny ever speak on this, not even at the deep dive? I find that awfully strange. All signs suggest that EA person misspoke or may have been thinking about something else entirely when they made those remarks and applied it to Playstation 5, but no evidence since supports hardware accelerated machine learning on Playstation 5.

You have former Playstation only developers who have already confirmed that Xbox Series X's Compute Units are more suited to Machine Learning. Why is that? It can't be because there are more of them. It must be because there is capability in a Series X shader core that isn't in a Playstation 5 shader core.

Xbox Series X's Advantage Could Lie in Its Machine Learning-Powered Shader Cores, Says Quantic Dream

Looking at the hardware of Xbox Series X and PS5, Quantic Dream believes Microsoft's advantage could lie in the console's ML-powered shaders.

wccftech.com

"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)."

As for blueisviolet on twitter, since you mentioned him, I see his posts from time to time and I've never known him to fake anything. He will literally take information from either official or highly public sources, and then from there you would be right in saying he will make all kinds of wild guesses and assumptions that are either impossible to prove or likely to not be as he thinks, and then sometimes he appears like he might actually be onto something due to how things line up and the way he presents it (who really knows though till confirmed) plus there's a language barrier that's obvious. English is not his primary language. He has been much more right than people like RGT and Moore's Law have been though, that's for certain. So depending on what you think about him, that should tell you something about people trusting Moores Law and RGT when it comes to either console .

But accusing him of literally faking a tweet from a sony employee? That's a bridge too far. Never known him to make up a tweet. I recall actually seeing that tweet back when it was made and making the rounds. If it's impossible to find now, it's because it's been deleted. Even the poster he was responding to has largely deleted most of their tweets, possibly a request by the individual. I've been on the receiving end of requests by employees (always respectfully, never hostile) to delete specific tweets as it can create issues for them if left up, and I've done so. You think there aren't statements I have screenshotted, I'd love to use in these discussions, but sometimes people slip and say a little too much, and then reach out to have it pulled for sake of their job.

 

[Fafalada]

..if you really want to defend that, or show how it's possible for that statement to be true then do.

I'm not engaging with the 'beats the hw implementation' and 'does it in software' parts because there's assumptions in there that I'm not interested in trying to unpack.

As for 'Geometry engine does VRS' bit - here's an example (if you can excuse the patent language obfuscation).

Rather, in this implementation, it is possible to first apply the different rendering parameters later, during primitive assembly.
As indicated at 938, a primitive assembler may assemble primitives to screen space for rasterization from the batches of primitives received. In this example, the primitive assembly may run iteratively over each zone of the screen that the object covers, as shown at 939. Accordingly, it may re-assemble each primitive in a batch of primitives of an object according to the number of zones, with unique primitive assembly rendering parameters for each zone the object of the primitive overlaps. As indicated at 940, each of the resulting primitives may then be scan converted to its target zone's screen space for further pixel processing.

 

[Elog]

I'm not engaging with the 'beats the hw implementation' and 'does it in software' parts because there's assumptions in there that I'm not interested in trying to unpack.

As for 'Geometry engine does VRS' bit - here's an example (if you can excuse the patent language obfuscation).

He is really mixing up two different ways to conduct VRS - it can be based on screen space or geometry. Tier 2 added primarily ways to do it on the geometry instead of the screen space.

 

[x@3f*oo_e!]

I'm not engaging with the 'beats the hw implementation' and 'does it in software' parts because there's assumptions in there that I'm not interested in trying to unpack.

As for 'Geometry engine does VRS' bit - here's an example (if you can excuse the patent language obfuscation).

I don't see where or how that is anything like VRS - do you have the patent link.

The common definition of VRS is variable resolution rasterisation (specifically the shading step), not variable LOD, tesselation or anything else, which could easily be done at the geometry stage

 

[martino]

I'm not engaging with the 'beats the hw implementation' and 'does it in software' parts because there's assumptions in there that I'm not interested in trying to unpack.

As for 'Geometry engine does VRS' bit - here's an example (if you can excuse the patent language obfuscation).

this is a valid example