• 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.

Why did Sony stay with 36CUs for the PS5?

JonnyMP3

Member
Well you can't say that. MSFT and Sony have unique HW feature sets. (Sony with the geometry engine.) This could definitely impact the size of each CU.
No, size of the CU doesn't change. Transistor layout for the specialised job within the CU changes.
 

Azurro

Banned
I think Series X going with the higher-end big die, took Mark by surprise. Hence the PS5 clocks went from 2Ghz target to 2.23Ghz.

You don't just change in input field in the firmware because "MS got you by surprise". Sony and Cerny were targeting the highest clock they could get in order to obtain the most amount of performance out of a CU. They designed a very unorthodox power delivery design and a cooling solution to go along with it, just to be able to run it as fast as possible within a reasonable power budget and not make the machine melt. You take your design, you test it, then push it as far as you can and then those are your final clocks. There was no "overclocking".

Seriously, from where do you get this narrative that you can simply make these big architectural decisions on a whim? Is it just fanboysim?
 

Azurro

Banned
If you want the most powerful console you get the xbox series x pure and simple. Lots of people care about this sort of thing ( powerful cars, motorbikes, phones etc). Power is everything imo. Combine that with consumer friendly practices and the xbox game pass and it becomes impossible to justify a ps5.

Until you want to play Horizon: Forbidden West, or Final Fantasy XVI, or Final Fantasy VII remake part 2, or the next Sucker Punch game, or the next God of War, Gran Turismo, etc. etc. As long as the consoles are close enough, the only thing that matters are the games.

Unless you want to get your Series X and stare at Craig all day. :)
 
Before the consoles were revealed I did think they were going to be almost the same but with one having more GPU power.

But there are actually some fundamental differences between the two. Going to be interesting to see how it works out.

I am most interested to see the results of Xbox's massive CU count and PS5's SSD/IO solution.

I was reading about PS5's Tempest engine, sounds amazing. Can someone point me to information about Xbox's solution?
 
I think sony has fought smart. Reason being is the SSD despite the CU despite the speed. Its the SSD theres going to be a point where the game engines are overtime going to be enhanced. More better effects better gfx. Everyone knows how fast the ps5 can do geometry they know how fast it can do the assets. Unreal 5 for instance despite at 1440p 8k only used half a gig of ram due to the speed. It will eventually hit the gpu harder on the series x compared to ps5
 
Until you want to play Horizon: Forbidden West, or Final Fantasy XVI, or Final Fantasy VII remake part 2, or the next Sucker Punch game, or the next God of War, Gran Turismo, etc. etc. As long as the consoles are close enough, the only thing that matters are the games.

Unless you want to get your Series X and stare at Craig all day. :)
Only god of war appeals to me from the games you listed. However, after trying a couple of the sony exclusives on my PC I'm starting to doubt all the hype and reviews. Something fishy going on... Horizon feels like a budget assassins creed game and death stranding is like marmite. The thing is pc gaming has tons of games that similarly to death stranding appeal to certain audiences but they never enjoy such recognition. Something stinks about the whole thing regarding Sony games easily winning gotys, high review scores etc.
 

Azurro

Banned
Only god of war appeals to me from the games you listed. However, after trying a couple of the sony exclusives on my PC I'm starting to doubt all the hype and reviews. Something fishy going on... Horizon feels like a budget assassins creed game and death stranding is like marmite. The thing is pc gaming has tons of games that similarly to death stranding appeal to certain audiences but they never enjoy such recognition. Something stinks about the whole thing regarding Sony games easily winning gotys, high review scores etc.

So, all of those copies sold are a conspiracy? Tell me more, let's get a crack team of Phil Spencer, Craig and you to take down the evil Sony empire. We can get you a theme song, a 60s VW van, a talking brown dog would also be a fine addition. Do you want to wear the dress and glasses? Or should I give them to Phil?
 
Last edited:

JonnyMP3

Member
Only god of war appeals to me from the games you listed. However, after trying a couple of the sony exclusives on my PC I'm starting to doubt all the hype and reviews. Something fishy going on... Horizon feels like a budget assassins creed game and death stranding is like marmite. The thing is pc gaming has tons of games that similarly to death stranding appeal to certain audiences but they never enjoy such recognition. Something stinks about the whole thing regarding Sony games easily winning gotys, high review scores etc.
That mate, is what is known as 'Quality'
 
Last edited:
Horizon zero dawn used GPGPU procedural generation of foilage and environment using slim APIs and to the point. How does HZD do that performance on a taoster CPU and low TF ?

Runs great on big fat abstract DX12 I hear.
Does Death Stranding use the GPGPU procedural generation? Because it supposedly runs really well on Direct X 12.
 

geordiemp

Member
Does Death Stranding use the GPGPU procedural generation? Because it supposedly runs really well on Direct X 12.

No Death stranding does not. HZD is much more complex and doing allot more dedicated stuff.

I can imagine PC CPU then GPU trying to do the same thing between 2 RAM pools ,,,,it is not hard to speculate why some things are better suited to a dedicated design.
 
Last edited:

Silver Wattle

Gold Member
My guess would be, they went conservative with the GPU because they were taking a big gamble with such a cutting edge SSD design and needed to leave enough money for cost over run.

Alternatively, they are targeting 449 and not the 499 like everyone expects.
 

llien

Member
OK, I have given a few examples, your turn on why you think its only a few %.
Because had it been not a couple of %, we would see major performance jump from going from one API to another.
20% is more than a typical difference between GPU tiers.

How fast was Windows IO api when Linus attached at 30 gbs SSD ? erm,....it almost stopped :messenger_beaming:
I need context on what on earth you are talking about.
 
Crytek

You talked about the CUs. The PlayStation 5 now has 36 CUs, and the Xbox Series X has 52 CUs are available to the developer. What is the difference?


The main difference is that the working frequency of the PlayStation 5 is much higher and they work at a higher frequency. That's why, despite the differences in CU count, the two consoles’ performance is almost the same. An interesting analogy from an IGN reporter was that the Xbox Series X GPU is like an 8-cylinder engine, and the PlayStation 5 is like turbocharged 6- cylinder engine. Raising the clock speed on the PlayStation 5 seems to me to have a number of benefits, such as the memory management, rasterization, and other elements of the GPU whose performance is related to the frequency not CU count. So in some scenarios PlayStation 5's GPU works faster than the Series X. That's what makes the console GPU to work even more frequently on the announced peak 10.28 Teraflops. But for the Series X, because the rest of the elements are slower, it will not probably reach its 12 Teraflops most of the time, and only reach 12 Teraflops in highly ideal conditions.

Sony says the smaller the number of CUs, the more you can integrate the tasks. What does Sony's claim mean?

It costs resources to use all the CUs at the same time. Because CUs need resources that are allocated to the GPU when they want to run code. If the GPU fails to distribute all the resources on all the CUs to execute a code, it will be forced to drop a number of CUs in use. For example, instead of 52, use 20 of them because GPU doesn't have enough resources for all CUs at all times.

Aware of this, Sony has used a faster GPU instead of a larger GPU to reduce allocation costs. A more striking example of this was in the CPUs. AMD has had high-core CPUs for a long time. Intels on the other hand has used less core but faster ones. Intel CPUs with less cores but faster ones perform better in Gaming. Clearly, a 16- or 32-core CPU has a higher number of Teraflops, but a CPU with a faster core will definitely do a better job. Because it's hard for gamers and programmers to use all the cores all the time, they prefer to have fewer cores but faster.


This.

Also lmao at the post "the PS5 is an inferior console", what? It's I/O architecture + SSD, geometry engine, etc, are inferior? 😂

Real world performance wise I see it being the superior machine in many scenarios due to the low level access and efficient custom design choices. Theoretical flops from CUs that will not all be saturated won't translate to better performance.
 

longdi

Banned
You don't just change in input field in the firmware because "MS got you by surprise". Sony and Cerny were targeting the highest clock they could get in order to obtain the most amount of performance out of a CU. They designed a very unorthodox power delivery design and a cooling solution to go along with it, just to be able to run it as fast as possible within a reasonable power budget and not make the machine melt. You take your design, you test it, then push it as far as you can and then those are your final clocks. There was no "overclocking".

Seriously, from where do you get this narrative that you can simply make these big architectural decisions on a whim? Is it just fanboysim?

but the thing is, increasing clock speeds is the easiest way to do just that.

was PS5 targetting 2.23ghz from the start? we wont know. but let see where Amd 7nm+ rdna2 ends up in their clocks. :messenger_smiling_with_eyes:
 
No Death stranding does not. HZD is much more complex and doing allot more dedicated stuff.

I can imagine PC CPU then GPU trying to do the same thing between 2 RAM pools ,,,,it is not hard to speculate why some things are better suited to a dedicated design.
Wasn't sure. Thought I'd read that they run off the same engine. I get what you are about HZD being really complex. But other complex open world games like RDR2 were ported over to PC while still allowing DirectX 12 and had better performance (more consistent). HZD is a bit of a cheery picked example.
 

DForce

NaughtyDog Defense Force
Imo Mark chose to stick with what brought them success.
cheaper lower power midrange die (36CU), to hit that $399 price tag, and work on cost reducing PS5 SSD storage.
From Epic china team, Sony using 12 channels SSD controller helps with keeping costs lower.

I think Series X going with the higher-end big die, took Mark by surprise. Hence the PS5 clocks went from 2Ghz target to 2.23Ghz.

Their plan to use smart shift tech was there from the beginning. They always had plans to sue high clocks.
 

Thirty7ven

Banned
Imagine if CU count was all you needed. Why exactly would it be hard for Sony to put the same CU count on PS5 as XSX?

Do people think it takes some genius engineering to up the CU count?
 

geordiemp

Member
Because had it been not a couple of %, we would see major performance jump from going from one API to another.
20% is more than a typical difference between GPU tiers.


I need context on what on earth you are talking about.

When have you compared Ps4 console vs pc with same power (CPU + GPU)) to compare apis ? I dont recall any.

No idea what you mean.

Linus bought a 10 grand SSD and tried to run it as fast as Ps5 on W10 lol...TIMESTAMPED

 
Last edited:

Stuart360

Member
There is a reason why the higher end Nvidia gpus, in the same family, often have lower clocks than the less powerful cards, yet produce vastly better results and performence.
Simply put, there is no substitute for more cu's.
 

Azurro

Banned
but the thing is, increasing clock speeds is the easiest way to do just that.

was PS5 targetting 2.23ghz from the start? we wont know. but let see where Amd 7nm+ rdna2 ends up in their clocks. :messenger_smiling_with_eyes:

No you are wrong, you can't just increase the clocks if you haven't designed your system to reach those clocks in the first place, and 2.23 GHz is really high for a GPU.

Cerny even mentioned that they had to cap the GPU at that speed as any higher broke the logic. How would he know that if they weren't trying to achieve the highest possible clocks within the thermal envelope and power consumption constraints?
 

JonnyMP3

Member
but the thing is, increasing clock speeds is the easiest way to do just that.

was PS5 targetting 2.23ghz from the start? we wont know. but let see where Amd 7nm+ rdna2 ends up in their clocks. :messenger_smiling_with_eyes:
Nope. The target was as high as possible within the new cooling solution. As Cerny stated in the Road to PS5 GDC talk, Sony have had to limit the GPU to 2.23Ghz as the most stable frequency. Meaning that the design could have gone a lot higher, if it didn't mean the GPU melting itself.

Edit: Again, I write the same thing??? What is this madness?
 
Last edited:

geordiemp

Member
There is a reason why the higher end Nvidia gpus, in the same family, often have lower clocks than the less powerful cards, yet produce vastly better results and performence.
Simply put, there is no substitute for more cu's.

Ps5 and XSX are not in the same family so your post comparison is irrelevant.

Ps5 has direct apis, XSX has abstract apis.
Ps5 has custom Geometry engine, patents on foveated randering type VRS linked with different type of mesh shaders
Ps5 has cache scrubbers and coherency engine
XSX will use mesh shaders and VRS.
XSX has more CUs clocked 20 % less.

The apis and clock speed alone will close the 18 % gap, dont be so sure lol. NO I did not say ps5 was more powerful, I said gap will be minimal.

And ps5 first party will stand out.

There is more to be revealed about ps5 and XSX, they will be more different in many instances.
 
Last edited:

Thirty7ven

Banned
There is a reason why the higher end Nvidia gpus, in the same family, often have lower clocks than the less powerful cards, yet produce vastly better results and performence.
Simply put, there is no substitute for more cu's.

The 2080ti vs 2080 super has a bigger difference than PS5 vs XSX on paper, and yet the 2080ti doesn’t produced vastly better results and performance.

This is of course if you care about language and how closely it translates the real world into words.

It’s the kind of rhetoric you only see with console fanboys, as it’s completely detached from reality.
 
Cheaper silicon and yields, no more, no less.

Besides, the sony marketing machine is now Apple tier. The playstation nation will lap sonys arsehole with their tongues no matter what they deliver anyway.


Cerny has reached jony ive/steve jobs level of cult messianic opiate.
 

Stuart360

Member
Ps5 and XSX are not in the same family so your post comparison is irrelevant.

Ps5 has direct apis, XSX has abstract apis.
Ps5 has custom Geometry engine, patents on foveated randering type VRS linked with different type of mesh shaders
Ps5 has cache scrubbers and coherency engine
XSX will use mesh shaders and VRS.
XSX has more CUs clocked 20 % less.
Its not irrelevant at all. There is no substitute for cu's, simple as that. I dont care what anyone says, Cerny included, you dont set out making a console with a 2.23ghz gpu clock speed in mind, you dont. I remember when the first rumours sstarted about next gen, and the early rumours were PS5 targeting 2.0ghz, and even that speed sounded ridiculous to me at the time. That speed is ridiculous even by PC standards.
Its painfully obvious that the clocks are that high because they decided to go with a low CU count, for whatever reason.
 

geordiemp

Member
Its not irrelevant at all. There is no substitute for cu's, simple as that. I dont care what anyone says, Cerny included, you dont set out making a console with a 2.23ghz gpu clock speed in mind, you dont. I remember when the first rumours sstarted about next gen, and the early rumours were PS5 targeting 2.0ghz, and even that speed sounded ridiculous to me at the time. That speed is ridiculous even by PC standards.
Its painfully obvious that the clocks are that high because they decided to go with a low CU count, for whatever reason.

Its simple in your mind, simple numbers are easy for some.

You can stick with Github if you like if it makes you happy. Sony also patented a special cooling solution, must be Github as well and whatever dumb rubbish the discord gang can muster - problem is, it would sound better if it was not so dumb.

And PC RDNA2 is not out yet, I bet frequencies are not 1.825 Ghz. The real question is, why is XSX clocked so low, what is the problem ?

You may think its not low, but wait uintil PC parts appear. Bookmarked.
 
Last edited:

BootsLoader

Banned
They have already lost the power war, the real question is whether MS will undercut them and they will also lose the value war at the same time. MS has way more money to be able to take a loss if they want, should be interesting to see if they decide to do so or just continue keeping on.
The Xbox division can’t take any loses “again”. Don’t think that Microsoft’s money will go to their gaming division. Please understand that.
 

LordOfChaos

Member
My thinking for a while has been that GNM is more bespoke to hardware than the DX12 variant in Xbox. DX12 did lower overhead compared to the DX11 variant the XBO launched with, but it's likely still built to be more portable as Microsoft also cares about PC cross platform games, or had planned ahead to the upgraded consoles. GNM seems further lower level, but the trade off is less portability, which is why Sony was so careful about even simply increasing clock speeds in Boost mode.

Getting someone to talk about both (heavily NDA'ed) would be very hard, but this is my strong hunch.

May not be the only reason they went with 36CUs though, maybe it was just the figure they saw fit to hit balancing everything else (tempest engine, SSD speed, controller cost etc), and I don't see why they couldn't toggle the extra off.
 
Last edited:

AGRacing

Gold Member
Hey if those recent yield rumors pan out... we COULD be looking...... at THIRTYEIGHT CU'S!!!!! 3 - 8 baby!!! WOOOOOO!!! *John Cena music playing*
 

llien

Member
geordiemp geordiemp
I can't take "what happened when some clickbaity guy inserted something into something" seriously.

I find it amazing that people don't understand the simple basics of 'Direct' will always be quicker than 'Layered' no matter much the power given.

Nothing is "direct" even in consoles and no overhead of practical "layered" (which is built into CPUs, in a way, for most common stuff), bar bizarre scenarios, is more than just barely slower.

And when there is "layer" there are good reasons for it, such as security.

Linus bought a 10 grand SSD and
Seriously, guys, how naive should one be to take this at face value?
He just randomly decided to go and buy 10 grand SSD, to, well, maybe somethings gotta give?
 
Last edited:

JonnyMP3

Member
geordiemp geordiemp
I can't take "what happened when some clickbaity guy inserted something into something" seriously.



Nothing is "direct" even in consoles and no overhead of practical "layered" (which is built into CPUs, in a way, for most common stuff), bar bizarre scenarios, is more than just barely slower.

And when there is "layer" there are good reasons for it, such as security.
No. It's a simple thing The layer is about translation to correct hardware variation.
And direct is always better.
Here's an example... You and another person speak the same language.
That's direct.
You and another person do not speak the same language. You need a translator to interpret the conversation back and forth. That's the layer!
 

llien

Member
Its painfully obvious that the clocks are that high because they decided to go with a low CU count, for whatever reason.

Most likely PR department demanded two digit TF figure, so they did that "if you push it hard, and if CPU is kinda not that busy".... "but, yeah, we kinda cannot guarantee it will always be at that clockspeed".

Here's an example...
I work in IT, dude.
Your example of a layer is, weird, to put it softly.
Of course it adds overhead.
But almost never it is of "translate it from X to Y format" kind (as that could be a major performance hit... and when one needs it real hard, there is 'built into hardware' support for it, e.g. AMD CPUs adding support for AES encrypted RAM)
 
Last edited:

geordiemp

Member
geordiemp geordiemp
I can't take "what happened when some clickbaity guy inserted something into something" seriously.



Nothing is "direct" even in consoles and no overhead of practical "layered" (which is built into CPUs, in a way, for most common stuff), bar bizarre scenarios, is more than just barely slower.

And when there is "layer" there are good reasons for it, such as security.


Seriously, guys, how naive should one be to take this at face value?
He just randomly decided to go and buy 10 grand SSD, to, well, maybe somethings gotta give?

Nothing is direct, some apis are less direct than others. More direct is the word,.

The Linus thing was funny, but did show he got better out of Linux than Windows and both slowed down the SSD considerably. It probably why MS is rewriting this stuff and calling it velocity architecture.
 

JonnyMP3

Member
Most likely PR department demanded two digit TF figure, so they did that "if you push it hard, and if CPU is kinda not that busy".... "but, yeah, we kinda cannot guarantee it will always be at that clockspeed".


I work in IT, dude.
Your example of a layer is, weird, to put it softly.
The whole context is about how the PS4 and Horizon work better than Horizon does with PCs, because of the API has to translate the game into different hardware configurations, whereas originally the game knows it works on PS4 architecture. What's weird about that?
 

llien

Member
Nothing is direct, some apis are less direct than others. More direct is the word,.

Being able to load data from SSD directly into GPU memory has nothing to do with API, it is something hardware is supposed to support.

And there... are different ways to use existing APIs:


The whole context is about how the PS4 and Horizon work better than Horizon does with PCs, because of the API has to translate the game into different hardware configurations, whereas originally the game knows it works on PS4 architecture. What's weird about that?

Horizon is using API which is drastically different from what is on PC. That is the "translation" in that context, not the actual DirectX overhead.
 
Last edited:

JonnyMP3

Member
Horizon is using API which is drastically different from what is on PC. That is the "translation" in that context, not the actual DirectX overhead.
But literally that's the same thing but in reverse.
The Game's problem is because the PC's API isn't what it knows.
And the PC is going the Game API isn't what it knows.
 

llien

Member
But literally that's the same thing but in reverse.
It's "the same thing" with your analogy, but it has nothing to do with the API overhead.
E.g. if XSeX API is mostly the same as DirectX, XSeX to PC port won't have this "overhead".
 

JonnyMP3

Member
It's "the same thing" with your analogy, but it has nothing to do with the API overhead.
E.g. if XSeX API is mostly the same as DirectX, XSeX to PC port won't have this "overhead".
It's not me making that claim mate...

 

squarealex

Member
Because of this

200809074720944084.png


Now, we know PS5 can't reach XSX on GPU performance raw expect some engine prefer frenquency over CU. (depending how dev make the game).
The only interesting thing on PS5 over XSX is the AMD SmartShift (CPU can help GPU and vice versa)
 

Moonjt9

Member
BC is the answer. It basically gives hardware BC which is different from MS’s software based BC.

also, higher clock speeds will boost other areas as well. As I’ve seen before the ssd and IO structure are so fast that GPUs struggle to keep up, so a faster clock speed here will help much more in a balanced system.
 

Ar¢tos

Member
My thinking for a while has been that GNM is more bespoke to hardware than the DX12 variant in Xbox. DX12 did lower overhead compared to the DX11 variant the XBO launched with, but it's likely still built to be more portable as Microsoft also cares about PC cross platform games, or had planned ahead to the upgraded consoles. GNM seems further lower level, but the trade off is less portability, which is why Sony was so careful about even simply increasing clock speeds in Boost mode.

Getting someone to talk about both (heavily NDA'ed) would be very hard, but this is my strong hunch.

May not be the only reason they went with 36CUs though, maybe it was just the figure they saw fit to hit balancing everything else (tempest engine, SSD speed, controller cost etc), and I don't see why they couldn't toggle the extra off.
The 1st rule of GNM is: you don't talk about GNM!
 

PaintTinJr

Member
I think people are coming at the decision of 36CUs from the wrong angle, and completely missing the importance of these APUs being SoC (System on a chip) that includes most motherboard features internal to the chip.

The heritage of the PS3 Cell was that it was the main atomic system (compute) unit in the world's most powerful super computer cluster for many years (based on set benchmarks, not FLOPs), and the atomic scalability that they provided never went really out of vogue (IMO) - even if developer backlash for the internal complexity demanded an evolved solution that wasn't fully possible for PS4 in the Jaguar APU era.

It seems that the PS4 Pro's 36CUs informed the design of the PS5 that this was an optimal compute unit count of a optimally atomic system(compute) unit completely in balance with unified memory, an IO complex and serviced by zen2 in an AMD Eypc design. Or atleast that is my theory for why AMD have for the first time ever won a major super computing contract with the US government with its up and coming Eypc solution, interfacing the atomic units with infinity fabric.

To put the Sony choice in another context, would you rather build a supercomputer with 36 XsX's or 52 x PS5's?

The architecture design strengths of the PS5 (IMHO) seem to be almost a generation ahead of the XsX - even ignoring the gains of the IO Complex over the Velocity Architecture - and with the developments Sony pioneered in stacked chips - like the Super slow motion cmos camera sensor in my previos XZ premium and current Xperia 1 - I have a hunch that the PS5's SoC is designed for stacking in the next iteration. Being able to constantly boost and save power being wasted as heat by the crazy high clock would just seem like icying on the cake for using them as a System(compute) unit.

Sony as a company have a need for a lot of compute power - cgi for Sony pictures being the obvious one - and the choices they have made seem to compliment that IMHO.
 
Top Bottom