• 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.
T

Three Jackdaws

Unconfirmed Member
That'll be 250 million dollars please!
giphy.gif
 

JonnyMP3

Member
Hold up, I may be reading it wrong, but you can use Rapid Packed Math to compress geometry Nanite style? Please correct me if I'm wrong, I'd love for the clarification.
In some implementations, the vertex parameter values include parameter values of all three vertices of a triangle primitive

In some implementations, compressing the vertex parameter values includes storing two floating point numbers together as one value having twice as many bits as each of the two floating point numbers.

In some implementations, compressing the vertex parameter values includes storing two 16-bit floating point numbers together as one 32-bit value.

In some implementations, the method further includes performing pixel shader computations on the interpolated parameter values.

In some implementations, the method may include, before compressing, modifying the parameter values with the vertex shader.


After trying to decipher the wall of text.....
I can say the vertex can include all 3 vertices for a primitive triangle which can be calculated at fp16 with RPM but also can be modified in the vertex shader, and the pixel shader can actually start shading the calculation of the vertex even as the vertex shader is being modified.

This is some rapid on the fly geometry! And I can definitely see it needing a robust geometry engine to continuously keep a hold of the continuous changing sequencing.
This is why the cache scrubbers are needed in the GPU because if the geometry map of vertices and shaders are saved on cache, it needs to be constantly updated of the constantly changing render parameters, deleting all unwanted data but keep necessary stuff and saving wastage in the render pipeline as the GPU rasterizes...

My God... This is definitely customisation heavy if it's true. Because it's definitely CPU intensive normally. And I dropped out of MSCE class like 20 years ago.
 
T

Three Jackdaws

Unconfirmed Member
In some implementations, the vertex parameter values include parameter values of all three vertices of a triangle primitive

In some implementations, compressing the vertex parameter values includes storing two floating point numbers together as one value having twice as many bits as each of the two floating point numbers.

In some implementations, compressing the vertex parameter values includes storing two 16-bit floating point numbers together as one 32-bit value.

In some implementations, the method further includes performing pixel shader computations on the interpolated parameter values.

In some implementations, the method may include, before compressing, modifying the parameter values with the vertex shader.


After trying to decipher the wall of text.....
I can say the vertex can include all 3 vertices for a primitive triangle which can be calculated at fp16 with RPM but also can be modified in the vertex shader, and the pixel shader can actually start shading the calculation of the vertex even as the vertex shader is being modified.

This is some rapid on the fly geometry! And I can definitely see it needing a robust geometry engine to continuously keep a hold of the continuous changing sequencing.
This is why the cache scrubbers are needed in the GPU because if the geometry map of vertices and shaders are saved on cache, it needs to be constantly updated of the constantly changing render parameters, deleting all unwanted data but keep necessary stuff and saving wastage in the render pipeline as the GPU rasterizes...

My God... This is definitely customisation heavy if it's true. Because it's definitely CPU intensive normally. And I dropped out of MSCE class like 20 years ago.
I'll just pretend to know what on Earth you have just said lol

Although I do remember a remark by Mark Cerny about how the "Primitive Shaders" are great for synthesising geometry on the fly!
 

JonnyMP3

Member
I'll just pretend to know what on Earth you have just said lol

Although I do remember a remark by Mark Cerny about how the "Primitive Shaders" are great for synthesising geometry on the fly!
Even my own brain can't fully comprehend.
If I had to use a Marvel character to explain my situation... It'd be Hawkeye because this is above my pay grade.

But the ELI5 version is the GPU can render stuff a lot more effectively and efficiently.

Because normally when the GPU gets instructions, that's it. It has to render the whole frame and then in the next one can things be adjusted according to data input of the game.
What PS5 GPU can do is render stuff on the fly WITH the new data.

So instead of waiting for the GPU to finish drawing the frame from the current blueprint given by the CPU and each frame is in this process.
Game code is effectively telling the GPU what it wants to draw without having to use a blueprint on the fly, because it had a fuck ton of tricks to help manage it.

I don't even feel that's an ELI5...

Edit: But just to be clear, if all this is really correct and it works.
First Party Games will definitely look as good or maybe better than the Unreal Engine 5 demo as they get better at getting grips with the tech.

Addendum: We say it can render on the fly quickly because the GPU is 2.23GHz. That clock is refreshing rapidly so it's also rendering rapidly.
The bottle neck used to be either dataspeed bottle neck... I.E. I/O throughput or not enough RAM to store assets.
 
Last edited:

Dabaus

Banned
According to Xbox fandom that I was reading halo infinite delay will have no impact on the Xbox’s launch this year and Ps5 will also be 600 dollars while the series x will be 500. Like, imagine being that guy who types something like that out and hits send or post or whatever.
 

qiqiqiiq

Member
According to Xbox fandom that I was reading halo infinite delay will have no impact on the Xbox’s launch this year and Ps5 will also be 600 dollars while the series x will be 500. Like, imagine being that guy who types something like that out and hits send or post or whatever.
I imagine that they want their E3 2013 moment when Sony announced a more powerful console for 100 dollars less.
 
I understand. Should've elaborated further. Assumption I was making is that they are building Halo Inf. on the current gen engine.
Trying to brute force fps where engine is not optimised for Zen2 CPU. Just speculating of course.

Regards
Agree but even now exists titles like Doom which looks much more better in current console with hardware of 2013, so yeah its tools
are big part of the problem but also the chiefs in Xbox should say when they saw the version they were going to show that day and
say "we cannot share that".
 
Last edited:

HAL-01

Member
Agree but even now exists titles like Doom which looks much more better in current console with hardware of 2013, so yeah its tools
are big part of the problem but also the chiefs in Xbox should say when they saw the version they were going to show that day and
say "we cannot share that".
60fps is not the problem by itself but it’s part of a series of bad executive decisions each of which make the game harder to develop for: not only does it need to be optimized for native 4K, both 60fps and 120fps, with split screen co-op, but also optimized for 4 different consoles hitting a vast range of performance levels, all the while being an open-world with dynamic time of day. Add to that having a revolving door for a studio and you get the halo infinite gameplay demo
 
Last edited:

FeiRR

Banned
I understand. Should've elaborated further. Assumption I was making is that they are building Halo Inf. on the current gen engine.
Trying to brute force fps where engine is not optimised for Zen2 CPU. Just speculating of course.

Regards
There are many problems with Halo presentation and a huge part of them is Microsoft targeting a wide variety of hardware. First of all, they say that Xbox Series X has many advanced features (let's put them together under RDNA2 name). However, for some reason they don't work with their advanced hardware, instead they show games off PCs. This means, those advanced features have to be software emulated because RDNA2 cards aren't out there (there's a chance Microsoft has access to them but I don't know if they do).

You can notice in the Halo demo that, for example, VRS makes a lot of distant textures look bad, people pointed out 720p textures. It's a result of software emulation of that particular feature on a PC which doesn't have any other way to do it. But VRS is quite simple, can be software emulated.

Now let's think about Geometry Engine. It's a hardware solution to drastically increase rendering power. Can you emulate it somehow? Well, of course, you can. But to do that, you'll actually decrease the speed of rendering many times, instead of improving it. A game which uses that feature, ported directly from PS5 to PC, would crawl on even the fastest GPUs provided they don't have those hardware components. Current HZD PC port is a good example of that.

Now imagine you're Microsoft and Game Pass is your song. You need to make a game for Xbox last gen, Xbox Series X/S and actually TWO kinds of Windows platforms: those with Nvidia GPUs and those with AMD GPUs. There are also Intel discrete GPUs but you can just set specs higher and say it won't work (even though Intel is trying to up their GPUs now). Can you use specific RDNA2 features in your games? No, either Xbox original will suffer or it's possible that Nvidia cards in PCs won't be able to handle those. They may have similar features but done in a different way because of, for example, patents. But then it means a lot more work for your devs.

We're going to witness a vicious circle: Xbox Series X/S launches with no exclusive games -> few people buy the console -> Microsoft first-party face the inevitable: most of their customers are PC users with two different GPU platforms -> they don't use advanced features of their own 9th gen consoles -> Xbox games look far worse than PS5 games -> Xbox still doesn't sell... and so on, at nauseam.

And this is their own studios, if you think about third-party: so many platforms to code for, small install base and the competition of GP games being more attractive than your brand new $60 offer. I think most of them will just pass and not release the Xbox version. Only the biggest publishers with sure sales of millions (EA, Activision, Take 2) will bite that bitter sandwich.

All this is simple industry logic. Microsoft not seeing this means they are pretty stupid, ignorant, call it as you like. Showing Halo in its current state to gamers was an offence. They thought people would just accept how it looks. That's a whole new level of arrogance.
 
Last edited:

Elog

Member
I think people are not fully taking in what the Boon twitter thread means.

Let's start with resolution. We know both the XSX and the PS5 have sufficient Tflops to push both 4K30 and 4K60. The XSX can squeeze out more FPS than PS5 if need be due to the TFlops advantage but both can do 4K60 which is the key point here.

What Boon is saying is that at both 30 and 60 FPS the real bottle neck is post-processing or in other word how much eye candy you want. And this budget is mainly measured in ms i.e. time.

Post-processing will in turn be determined by the amount of silicon dedicated for this task (it needs to be sufficient to run those 30 or 60 FPS per second at 4K) but the real constraint is time. Time will in turn to a major degree be determined by cache management and frequency...

You know where I am going with this. In addition, the PS5 has a more advanced geometry engine and can push more and higher-fidelity textures.

Cerny is no fool.
 
Last edited:

h00jraq

Neo Member
So why AestheticGamer is twitting about PS4 huge problems with 60fps @ 1080p while XSX can reach 60fps @ 4K without big issues?
He has so many good leaks that I believe he is saying the truth.
 

geordiemp

Member
The fp16 stuff was right from the start I told people PS4 Pro was 8.4TF fp16 before Cerny did but people are childish & attack what they don't understand.

I re,ember your persecution well, FP16 on ps4 was ahead of its time but was never really used for that power rating.

Ps5 its difficult as most of teh patent suggests its bandwidth / cache efficiency and compression and bottlenecks, if there is MUCH more performance GAF would melt.
 

Grinchy

Banned
Do you guys think destruction all stars will be a launch game for PS5? it looks really fun, it reminds me of Vigilante 8/Twisted Metal.
I think it looks awesome too, but I don't think it'll be a launch game. If they felt confident in a date, I think they would have said "Holiday 2020" or something. They didn't even feel confident saying early 2021 :messenger_crying:
 

mejin

Member
So why AestheticGamer is twitting about PS4 huge problems with 60fps @ 1080p while XSX can reach 60fps @ 4K without big issues?
He has so many good leaks that I believe he is saying the truth.

learn to differentiate what is just information and what is opinion.

There is malice when he is projecting his opinions.

He and Grubb started to side MS a few days ago. They were just leakers before.
 

Shmunter

Member
So why AestheticGamer is twitting about PS4 huge problems with 60fps @ 1080p while XSX can reach 60fps @ 4K without big issues?
He has so many good leaks that I believe he is saying the truth.
Because his source would be someone that observed it without any direct understanding of the context of what they were looking at. The ps5 version may have been compiled for the first time that day vs XsX version having months of work. Etc, that’s just a guess as good as any, but an educated guess.

Why? because 1080p 60 issues on one vs 4K 60 no problems on the other, simply do not stand up to any scrutiny of logic.
 

FunkMiller

Member
There are many problems with Halo presentation and a huge part of them is Microsoft targeting a wide variety of hardware. First of all, they say that Xbox Series X has many advanced features (let's put them together under RDNA2 name). However, for some reason they don't work with their advanced hardware, instead they show games off PCs. This means, those advanced features have to be software emulated because RDNA2 cards aren't out there (there's a chance Microsoft has access to them but I don't know if they do).

You can notice in the Halo demo that, for example, VRS makes a lot of distant textures look bad, people pointed out 720p textures. It's a result of software emulation of that particular feature on a PC which doesn't have any other way to do it. But VRS is quite simple, can be software emulated.

Now let's think about Geometry Engine. It's a hardware solution to drastically increase rendering power. Can you emulate it somehow? Well, of course, you can. But to do that, you'll actually decrease the speed of rendering many times, instead of improving it. A game which uses that feature, ported directly from PS5 to PC, would crawl on even the fastest GPUs provided they don't have those hardware components. Current HZD PC port is a good example of that.

Now imagine you're Microsoft and Game Pass is your song. You need to make a game for Xbox last gen, Xbox Series X/S and actually TWO kinds of Windows platforms: those with Nvidia GPUs and those with AMD GPUs. There are also Intel discrete GPUs but you can just set specs higher and say it won't work (even though Intel is trying to up their GPUs now). Can you use specific RDNA2 features in your games? No, either Xbox original will suffer or it's possible that Nvidia cards in PCs won't be able to handle those. They may have similar features but done in a different way because of, for example, patents. But then it means a lot more work for your devs.

We're going to witness a vicious circle: Xbox Series X/S launches with no exclusive games -> few people buy the console -> Microsoft first-party face the inevitable: most of their customers are PC users with two different GPU platforms -> they don't use advanced features of their own 9th gen consoles -> Xbox games look far worse than PS5 games -> Xbox still doesn't sell... and so on, at nauseam.

And this is their own studios, if you think about third-party: so many platforms to code for, small install base and the competition of GP games being more attractive than your brand new $60 offer. I think most of them will just pass and not release the Xbox version. Only the biggest publishers with sure sales of millions (EA, Activision, Take 2) will bite that bitter sandwich.

All this is simple industry logic. Microsoft not seeing this means they are pretty stupid, ignorant, call it as you like. Showing Halo in its current state to gamers was an offence. They thought people would just accept how it looks. That's a whole new level of arrogance.

Thought exercise:

What would Xbox Series X, Microsoft and the landscape in general look like if Halo Infinite had been designed as an Xbox Series X exclusive? How different would things be right now?
 
This Dusk golem twitter thread and lots of dumb people believing that shit in there is just baffling to me. I get you may like a plastic box over the other but come on! Do not leave your brain in a dumpster when you are faced with facts and need to decide, the signs are there that none of these "insider" fucks are telling the truth and boasting power once again when the software library is vehemently lacking. Do not believe their lies and base your buying decision on them and spend your money erroneously... PS5 is just another class above XSX, in terms of custom hardware made exclusively for the platform and also in terms of exclusive software library that will only be on that platform. If your friends are on xbox and they can buy only one console at the beginning, I would convince everyone to migrate to Sony this generation (if they haven't already to begin with LOL). Isn't the last gen big enough warning that they have not learned anything and what's more they are exactly replicating their mistakes in other ways....
 

3liteDragon

Member
Holy fucking shit...

Also, that FP16 part goes perfectly along with what Cerny’s been saying since the PS4 Pro’s release.

Cerny: Finally, there's better support of variables such as half-floats. To date, with the AMD architectures, a half-float would take the same internal space as a full 32-bit float. There hasn't been much advantage to using them. With Polaris though, it's possible to place two half-floats side by side in a register, which means if you're willing to mark which variables in a shader program are fine with 16-bits of storage, you can use twice as many. Annotate your shader program, say which variables are 16-bit, then you'll use fewer vector registers.

If you place two half-floats together instead of one full-float on the vector registers, it literally means more wavefronts running on the CUs.

Cerny: Multiple wavefronts running on a CU are a great thing because as one wavefront is going out to load texture or other memory, the other wavefronts can happily do computation. It means your utilisation of vector ALU goes up," Cerny shares.

Anything you can do to put more wavefronts on a CU is good, to get more running on a CU. There are a limited number of vector registers so if you use fewer vector registers, you can have more wavefronts and then your performance increases, so that's what native 16-bit support targets. It allows more wavefronts to run at the same time.

THIS is the reason why he decided to keep the CU count the same for PS5, it’s easier to get close to that peak TFLOPs number (10.28 TFLOPs) with way less CUs clocked at an insane frequency and by also running a lot more wavefronts on those CUs. (Extracting more performance out of them and VALU utilization goes up, AKA the amount of TFLOPs used goes up)

You don’t need 32-bits of storage for storing vertex values every frame, that can be achieved with way less precision like FP16.

This is some truly insane stuff, can you imagine the type of games we’re gonna see once devs master the hardware?

Cerny’s an absolute genius and Sony’s honestly super lucky to have him.
 
Last edited:

h00jraq

Neo Member
learn to differentiate what is just information and what is opinion.

There is malice when he is projecting his opinions.

He and Grubb started to side MS a few days ago. They were just leakers before.

Well, it would be strange if he was "bought" by M$ and he is biased. I can't wait for more to hear about the performance of PS5.
Strange thing is that M$ is focused on performance and they are not giving to much info about system features, gamepad or SSD.
 

DrDamn

Member
Is an indies, yes it is but at least I can see it and play it this year.


I get all excited when I see a flight game based around birds of prey like this. So many interesting possibilities with flight mechanics and control. It usually only takes a minute into a video for the birds to start shooting lasers/guns/funkiness and my excitement deflates. So many great things you could do with this sort of idea, and adding "guns" isn't one of them. :(

(see also Eagle Flight - Ubisoft VR game)
 

zaitsu

Banned
Holy fucking shit...

Also, that FP16 part goes perfectly along with what Cerny’s been saying since the PS4 Pro’s release.



If you place two half-floats together instead of one full-float on the vector registers, it literally means more wavefronts running on the CUs.



THIS is the reason why he decided to keep the CU count the same for PS5, it’s easier to get close to that peak TFLOPs number (10.28 TFLOPs) with way less CUs clocked at an insane frequency and by also running a lot more wavefronts on those CUs. (Extracting more performance out of them and VALU utilization goes up, AKA the amount of TFLOPs used goes up)

You don’t need 32-bits of storage for storing vertex values every frame, that can be achieved with way less precision like FP16.

This is some truly insane stuff, can you imagine the type of games we’re gonna see once devs master the hardware?

Cerny’s a fucking genius and Sony’s honestly super lucky to have him.
I agree.
PS4 was his idea " budget testing device".
With PS5 he had green light to pursuit his dreams. Of course without budget constraints he would go for bigger GPU. But budget isn't unlimited. I belive him in him having wise approach.

I hope MS will fight with last thing that can save them - console price. Because without them going cheap SONY would go high. Because at the moment there is no reason to buy XSX if you have XOX.

And i want to clear one thing :
RE8 was ported to nextgen consoles from currentgen and if PS5 was struggling with RE8 and XSX not out of the gate it only means one thing and thats not that XSX is superior but that XSX is continuation of currentgen consoles. PS5 on the other hand is first of its own kind and needs vastly different approach. XSX isnt easiest console to learn but devs already know it from XOX and PC , PS5 is easiest console to learn but it is still the first time devs meet this kind of hardware.
 
Last edited:

FeiRR

Banned
Thought exercise:

What would Xbox Series X, Microsoft and the landscape in general look like if Halo Infinite had been designed as an Xbox Series X exclusive? How different would things be right now?
They'd have one nice exclusive at launch, happy hardcore fanbase with hopes for the same games formula as always (Forza, Fable) plus some nice prospects for new IPs in the future. Honestly, I had believed that would be the case. Not my thing but a lot of people just like those games and I feel sorry for them.
 
Holy fucking shit...

Also, that FP16 part goes perfectly along with what Cerny’s been saying since the PS4 Pro’s release.



If you place two half-floats together instead of one full-float on the vector registers, it literally means more wavefronts running on the CUs.



THIS is the reason why he decided to keep the CU count the same for PS5, it’s easier to get close to that peak TFLOPs number (10.28 TFLOPs) with way less CUs clocked at an insane frequency and by also running a lot more wavefronts on those CUs. (Extracting more performance out of them and VALU utilization goes up, AKA the amount of TFLOPs used goes up)

You don’t need 32-bits of storage for storing vertex values every frame, that can be achieved with way less precision like FP16.

This is some truly insane stuff, can you imagine the type of games we’re gonna see once devs master the hardware?

Cerny’s a fucking genius and Sony’s honestly super lucky to have him.
Yeah also FP16 is much more suited to warped GPGPU atomics. Basically lots of filtering and eye candy that previously cost lots of frame time before can be done in much quicker times/more frequently, now running in more-parallel, half precision wavefronts. This is like using GPU actually more like what it really does for graphics but now for non-graphics general purpose computing more so than ever before. And this way you magically double some of the bandwidth available from the RAM, as you double the operation per clock, you also halve the memory print. The trouble in the past was the coherency across RAM and (order of magnitude faster) cache memory inside the GPU itself. But PS5 solves that with cache scrubbers inside the GPU and actually use FP16 in all software.
 

zaitsu

Banned
What i found :
(06-26-2018)
  • Added a new secret NDA-ed platform.
(10-17-2019)
  • Added initial support for a new secret platform. Contact Mitch for details.
(02-17-2020)
  • Added a new secret platform. ( i think its the same platform as 10-17-2019 )
(06-03-2020)
  • Updated to PS5 SDK 1.0. ( i think its 6-26-2018 NDA-ed platform)
  • Updated to secret platform 2020-05. ( i think its the same platform as 10-17-2019 and 02-17-2020)

My conclusion :
- Cerny said they switch to Kraken back in 2017 , Kraken is another RAD tool beside BINK and it was LAST MOMENT decision
- Bink have supported PS5 from 06-26-2018
- they never mention XSX so a new secret platform is just XSX
- Bink support for XSX started in 02-17-2020
- PS5 game development had almost 2 year headstart, and we know for sure that devkits of PS5 could have such a headstart
And thats the reason why we didnt see anything on XSX.
 

chilichote

Member
In some implementations, the vertex parameter values include parameter values of all three vertices of a triangle primitive

In some implementations, compressing the vertex parameter values includes storing two floating point numbers together as one value having twice as many bits as each of the two floating point numbers.

In some implementations, compressing the vertex parameter values includes storing two 16-bit floating point numbers together as one 32-bit value.

In some implementations, the method further includes performing pixel shader computations on the interpolated parameter values.

In some implementations, the method may include, before compressing, modifying the parameter values with the vertex shader.


After trying to decipher the wall of text.....
I can say the vertex can include all 3 vertices for a primitive triangle which can be calculated at fp16 with RPM but also can be modified in the vertex shader, and the pixel shader can actually start shading the calculation of the vertex even as the vertex shader is being modified.

This is some rapid on the fly geometry! And I can definitely see it needing a robust geometry engine to continuously keep a hold of the continuous changing sequencing.
This is why the cache scrubbers are needed in the GPU because if the geometry map of vertices and shaders are saved on cache, it needs to be constantly updated of the constantly changing render parameters, deleting all unwanted data but keep necessary stuff and saving wastage in the render pipeline as the GPU rasterizes...

My God... This is definitely customisation heavy if it's true. Because it's definitely CPU intensive normally. And I dropped out of MSCE class like 20 years ago.

Maybe that's a stupid question, but does that mean the GPU really only has to calculate what is necessary, because what is not needed is sorted out beforehand?

Would that mean the PS5 GPU needs fewer TFlops to compute the same as a GPU with more TFlops?
 

zaitsu

Banned
Maybe that's a stupid question, but does that mean the GPU really only has to calculate what is necessary, because what is not needed is sorted out beforehand?

Would that mean the PS5 GPU needs fewer TFlops to compute the same as a GPU with more TFlops?
It COULD, we will need to wait for ND next game to show everybody why SONY is the only truly nextgen on the market.
 
Last edited:
Status
Not open for further replies.
Top Bottom