PlayStation 5 System Architecture Deep Dive |OT| Secret Agent Cerny [SEE 4/1 STAFF POST]

How much do you think the PS5 will cost?

  • $399

    Votes: 661 31.3%
  • $449

    Votes: 615 29.1%
  • $499

    Votes: 677 32.1%
  • 25 Bags of Walnuts

    Votes: 159 7.5%

  • Total voters
    2,112

Pheonix

Member
Dec 14, 2018
2,586
St Kitts
lmao at your avatar.

For Sony's sake, i hope that Cerny's secret sauce in the gpu overcomes that tflops advantage because i am pretty sure its going to cost them sales. i used to visit gamespot forums back in the day and almost everyone switched from sony to ms that gen. they did come back later. but the initial power delta really did have an effect.
All this becomes moot the second the PS5 is priced $100 lower than the XSX. At that point, the conversation completely changes. Then the burden of proof falls on MS to really show why its console cot a $100 more.

As I used to say in the speculation thread, My prediction was that sony ends up with a 10TF machine, but I also said, whoever is weaker must but be cheaper.

If sony is within 20%o the XSX performance and can be $100 cheaper then the XSX, then MS has a serious problem.

If sony somehow makes the PS5 the same price with the XSX, then they really have fucked up. But as long as its cheaper, then being weaker doesn't just make sense, its even expected.
 

GING-SAMA

Banned
Jul 10, 2019
1,426
All this becomes moot the second the PS5 is priced $100 lower than the XSX. At that point, the conversation completely changes. Then the burden of proof falls on MS to really show why its console cot a $100 more.

As I used to say in the speculation thread, My prediction was that sony ends up with a 10TF machine, but I also said, whoever is weaker must but be cheaper.

If sony is within 20%o the XSX performance and can be $100 cheaper then the XSX, then MS has a serious problem.

If sony somehow makes the PS5 the same price with the XSX, then they really have fucked up. But as long as its cheaper, then being weaker doesn't just make sense, its even expected.

50$ cheaper I think for PS5.

That why a guy at sony say they wait for XSX price.

Both console cost almost the same price. A little more expensive for MS because of the GPU and the ram but I think that Microsoft made savings on the SSD with their different way of streaming games assets than sony.


499 - XSX
449 - PS5
299/349 - XSS
 

AudiophileRS

Member
Apr 14, 2018
203
What I'd like to know is if that SPU-like Audio CU is in addition to the 36 or one of them.

Could we see:

  1. 40CU die with 3 Disabled, 1 Audio & 36 Compute
  2. 40CU die with 4 Disabled, 1 Audio & 35 Compute
  3. 40CU die with 4 Disabled, 36 Compute and the 1 Audio "tacked on" in the APU design.

Hopefully it's 1 or 3.
 
Last edited:

Sankt Ra

Member
Oct 27, 2017
2,023
Cerny said inner SSD has 6 priority levels and NVME SSD spec has only 2 priority levels.

This means games that make use of more than 2 priority levels (most 1st party) HAVE to run from the built in SSD, not directly from the NVME.

When you want to game, the game will be copied from the external NVME to the internal drive.

Fortunately, at 5GB/s, this would take mere seconds to accomplish, and can be hidden from the player.

Thus the external NVMEs should have 0 overheating problems.
I understood it as, 7GB/s NVME cards will be certified, the overhead (2GB/s) is needed so the internal I/O chip can assign 6 levels of priority even with off the shelf NVMEs. So you can game off of them and simply expand storage. Its written in on of the DF guys articles I think.
 

Simuly

Member
Jul 8, 2019
114
All this becomes moot the second the PS5 is priced $100 lower than the XSX. At that point, the conversation completely changes. Then the burden of proof falls on MS to really show why its console cot a $100 more.

As I used to say in the speculation thread, My prediction was that sony ends up with a 10TF machine, but I also said, whoever is weaker must but be cheaper.

If sony is within 20%o the XSX performance and can be $100 cheaper then the XSX, then MS has a serious problem.

If sony somehow makes the PS5 the same price with the XSX, then they really have fucked up. But as long as its cheaper, then being weaker doesn't just make sense, its even expected.
Actual performance delta will be ~5-10% as I believe PS5 has heavily customised their GPU for efficiency and getting each CU working instead of sitting their idling like on a 52 CU GPU, plus other customizations with primitive shaders and culling engines etc.

But I think price difference will be PS5 $50 cheaper as they have room to wriggle with BOM being slightly below XsX.

In this scenario XsX will struggle just as much as XB1 did this gen outside NA.
 

mueske

Member
Apr 11, 2019
133
What I'd like to know is if that SPU-like Audio CU is in addition to the 36 or one of them.

Could we see:

  1. 40CU die with 3 Disabled, 1 Audio & 36 Compute
  2. 40CU die with 4 Disabled, 1 Audio & 35 Compute
  3. 40CU die with 4 Disabled, 36 Compute and the 1 Audio "tacked on" in the APU design.
Didnt they rework a CPU for that, and not GPU units?
 

Lady Gaia

Member
Oct 27, 2017
1,131
Seattle
What I'd like to know is if that SPU-like Audio CU is in addition to the 36 or one of them.
The Road to PS5 description made it quite clear that it's not a CU, it just borrows from their design. Everything is inference until we actually see the die layout and some analysis, but as described it's a distinct processing unit that runs in parallel with the 36 CUs. Whether there's any redundancy to prevent a local "audio CU" defect from reducing yields is something we really don't have any basis for speculation regarding. Chances are they've come up with either some form of redundancy, or they have enough evidence that it's a small enough portion of the die that it's not going to have much of an impact (there are always areas in any die that can't take a defect without rendering a given chip unusable, you just need to make sure the odds are low enough.)
 

Pheonix

Member
Dec 14, 2018
2,586
St Kitts
I hate to get in on the system wars, especially in a dedicated thread so I'll only make one post on the matter..

We seem to be going round in circles going from one extreme to another. The answer will no doubt lie somewhere in between

Will the XSX have a compute advantage over PS5? Yes..

Will the 16% to ~20% difference between the system's TF counts represent the real world performance differential? No..

When the entire GPU is clocking much higher everything outside the CUs is running faster; in addition you can make slightly more efficient use of those CUs.

The CPU difference is ~3x smaller than the difference between the OG X1 and PS4.

The SSD won't magically give you more frames but it does remove the most pressing bottleneck games have been facing for years.

I expect the real-world performance differential to be about half the GPU TF differential. So perhaps 8-10% in resolution as a rough example. Think 2052p vs 2160p.

The question you need to ask yourself as a developer or consumer is, is that tiny real-world difference worth the gains in I/O and the possibilities that brings in changing the fundamentals of the games we get to play or develop? I without a doubt think it is.
Well said... and to just add to this, its even not as complicated as the simple example you just gave. Its not even going to be a case of 2052p vs 2160p. It will be a case of [email protected] for 100% vs [email protected] for 90% of the time with framerates locked. That's it. That's what accounting for this new age of dynamic resolutions would amount to.

We are not at a place where one console would be able to hit one rez and the other just would never hit it. They both will hit whatever the target rez is set by the developer. The PS5 would just drop from that target rez 10%-15% of the time to maybe something like 2052p or 1900p.

Thats what that compute difference will amount to.
 

AegonSnake

Banned
Oct 25, 2017
7,534
All this becomes moot the second the PS5 is priced $100 lower than the XSX. At that point, the conversation completely changes. Then the burden of proof falls on MS to really show why its console cot a $100 more.

As I used to say in the speculation thread, My prediction was that sony ends up with a 10TF machine, but I also said, whoever is weaker must but be cheaper.

If sony is within 20%o the XSX performance and can be $100 cheaper then the XSX, then MS has a serious problem.

If sony somehow makes the PS5 the same price with the XSX, then they really have fucked up. But as long as its cheaper, then being weaker doesn't just make sense, its even expected.
i think sony did fuck up and this is the best they could do at $499. cerny got too cute with the design and got himself in a jam. but thats all speculation and we will see come the price reveals soon enough. i just have a bad feeling about this, and my instincts are usually right. those titanic sinking references leading up to the reveal were me just having that bad feeling about this whole thing even though i had full faith in klee.

as for sony being $100 cheaper, i agree conversation changes, but i dont see the platform ambiguous hardcore sticking with sony. i said this before, everyone i talked to yesterday was reluctantly staying with sony but they are all into sony exclusives. i think they will split the hardcore 50-50 which will be great for MS.

What do you think about the lockhart? do you think they steal casuals from Sony with that? i see lots of parents getting it for their kids who just play f2p games like fortnite and cod all day. i see both sony and ms selling 85 million units next gen each. it wont be a 110-45 million result like this gen. not sure what sony would do when losing 30% revenue, but i think the CEO and lead engineer would go if that was the case.
 

Lady Gaia

Member
Oct 27, 2017
1,131
Seattle
My wife has taken pictures of my ears, I’m ready.
I'm waiting for my wife's new iPad to arrive so we can do a Lidar-assisted 3D scan. 😉 Seriously, though, that same technology is almost certainly going to be in a ton of new iPhones by the end of the year and could be a fast path to the kind of HRTF customization that would be required. The potential is really interesting but there are a lot of other factors at play as well. The width of your skull plays a role in determining timing differences between a sound source arriving at your ears, so we've trained our brain not just to adapt to the shape of our individual ears, but other factors as well. I wouldn't be entirely surprised if a complete head scan turns out to be most effective.

What I really want to know is how divergent my personal HRTF is from the conventional "average" that has been in use for decades now. How much more impressive should 3D audio be than what I've heard to date?
 

Sankt Ra

Member
Oct 27, 2017
2,023
So this HRTF stuff is already available as a 199$ plug in , with the whole smartphone/ear/analysis stuff Cerny talked about.
This makes me think Sony is def. going to make this available for free within the (highly likely) PS5 app or part of PS+.
This has me really exited, as a (somewhat) audiophile who bought a 1000$ HP amp to use two EQs per channel/ear precisely to optimize (center) imaging.
SSD>cHRTF>TF. I'm pumped again.
 
Last edited:

revben

Member
Nov 21, 2017
43
MS have a very fast SSD and a from everything we've been told a great I/O stack and set of optimisations; in itself it deserves praise (particularly their texture solution), I don't question whether their solution will change game design, it will. But the systems have clear differences in priorities and Sony's solution has simply gone further. When Cerny refers to "the dream" from a developer perspective, it's referring to lifting the bottleneck and reaching a greater threshold where you can just load what you want on the fly; they seem to have surpassed that threshold.
That MSFT goal too, hence having API that give XSX 6/gb compressed optimized...yes PS5 is far better, but its not like Ubisoft will make the world less dense on XSX...it will take three seconds more to stream in data, not a minute...plus you cannot stream data faster than the GPU can draw and CPU calculate.
 

dbcyber

Member
Oct 25, 2017
1,000
I started buying more games over the last year because PS4 BC was confirmed, now it appears most of my games may not be playable which sucks. I really hope Sony/Cerny clears things up, maybe they wanna drop a bombshell in their games related event and confirm BC works for 99% of games but so far they only have 100 working with boost specs or something. I'll give them the benefit of the doubt until the next PS4 event.
 

VinFTW

Member
Oct 25, 2017
2,948
I guess I'm confused on how 3p SSD drives are going to work on this thing.

Do we have any inkling on how a far inferior/slower piece of SSD tech will be able to work in tandem with Sonys fucking mind boggling SSD/speeds? Like, how in the world is that not gonna work.
 

AudiophileRS

Member
Apr 14, 2018
203
Didnt they rework a CPU for that, and not GPU units?
According to Cerny they took the IP of an AMD GPU CU and modified it so that it is similar in nature to an SPU from PS3's Cell CPU. For audio this is the perfect middleground, highly parallel and near 100% utilisation.

He equated the performance of this modified CU to that of an entire 8-Core Jaguar CPU.

The Road to PS5 description made it quite clear that it's not a CU, it just borrows from their design. Everything is inference until we actually see the die layout and some analysis, but as described it's a distinct processing unit that runs in parallel with the 36 CUs. Whether there's any redundancy to prevent a local "audio CU" defect from reducing yields is something we really don't have any basis for speculation regarding. Chances are they've come up with either some form of redundancy, or they have enough evidence that it's a small enough portion of the die that it's not going to have much of an impact (there are always areas in any die that can't take a defect without rendering a given chip unusable, you just need to make sure the odds are low enough.)
Great to know, thanks! 👍
 

Pheonix

Member
Dec 14, 2018
2,586
St Kitts
i think sony did fuck up and this is the best they could do at $499. cerny got too cute with the design and got himself in a jam. but thats all speculation and we will see come the price reveals soon enough. i just have a bad feeling about this, and my instincts are usually right. those titanic sinking references leading up to the reveal were me just having that bad feeling about this whole thing even though i had full faith in klee.

as for sony being $100 cheaper, i agree conversation changes, but i dont see the platform ambiguous hardcore sticking with sony. i said this before, everyone i talked to yesterday was reluctantly staying with sony but they are all into sony exclusives. i think they will split the hardcore 50-50 which will be great for MS.

What do you think about the lockhart? do you think they steal casuals from Sony with that? i see lots of parents getting it for their kids who just play f2p games like fortnite and cod all day. i see both sony and ms selling 85 million units next gen each. it wont be a 110-45 million result like this gen. not sure what sony would do when losing 30% revenue, but i think the CEO and lead engineer would go if that was the case.
What of if it was sony's plan all along to make a $399 console? If this is them making a $399 console, then isn't it safe to say they made one helluva $399 box?

I wouldn't put too much stock into said "platform ambiguous hardcore". PS has enough hardcore fans regardless, and at the end of the day... een with the hardcore you are talking about...games are what it boils down to. It won't be the first time the PS was the weaker console and yet still sold the best.

Yes Lockhart would end up being the XS bet selling sku.. simply by it being cheaper. But only in NA. A $400 PlayStation would still dominate everywhere else. Another thing to consider is what would Lockhart specs be? If that 4TF rumors true? And it cost $300, then it makes a really, really, really good case for a $400 PS5 that has 10TF. I personally don't think MSreleases Lockhart in 2020. Maybe in 2021 when they have had time to gauge sales and or certain component prices likely drop.

Tome, this whole thing rides on how much sony prices the PS5.
 
Oct 25, 2017
5,837
I'm waiting for my wife's new iPad to arrive so we can do a Lidar-assisted 3D scan. 😉 Seriously, though, that same technology is almost certainly going to be in a ton of new iPhones by the end of the year and could be a fast path to the kind of HRTF customization that would be required. The potential is really interesting but there are a lot of other factors at play as well. The width of your skull plays a role in determining timing differences between a sound source arriving at your ears, so we've trained our brain not just to adapt to the shape of our individual ears, but other factors as well. I wouldn't be entirely surprised if a complete head scan turns out to be most effective.

What I really want to know is how divergent my personal HRTF is from the conventional "average" that has been in use for decades now. How much more impressive should 3D audio be than what I've heard to date?
I think it is very cool and will continue to be underrated.
 

Calo☆

Member
Jan 1, 2020
409
Has this been posted before? This is interesting. An aggressively overclocked 5700 (36 CUs) performs (very, very) slightly worse than stock 5700 XT. When you calculate the Teraflops rating the overclocked 5700’s is 10% higher than that of the 5700 XT, yet the real world performance is only on par.

If we assume PS5’s GPU indeed spends most of its time at boost frequency, this suggests that the real world performance may be 10% lower than a 10.28TF RDNA 2 GPU of different configuration. (it’ll still be faster than a 5700XT because we know RDNA 2 has some IPC improvements)

I guess the reason for this discrepancy is memory bandwidth, although the L2 on the overclocked card is supposedly faster, when it comes to bandwidth limited work, it just can’t be fed fast enough.
 

MrWilson

Member
Jun 11, 2019
8
I started buying more games over the last year because PS4 BC was confirmed, now it appears most of my games may not be playable which sucks. I really hope Sony/Cerny clears things up, maybe they wanna drop a bombshell in their games related event and confirm BC works for 99% of games but so far they only have 100 working with boost specs or something. I'll give them the benefit of the doubt until the next PS4 event.
Thats not what they meant, What they mean is they have tested the top 100 games and so far they all work with perhaps a very few with very minor bugs. They anticipate that to be the case with the rest of the ps4 catalog.
 

Calo☆

Member
Jan 1, 2020
409
What I'd like to know is if that SPU-like Audio CU is in addition to the 36 or one of them.

Could we see:

  1. 40CU die with 3 Disabled, 1 Audio & 36 Compute
  2. 40CU die with 4 Disabled, 1 Audio & 35 Compute
  3. 40CU die with 4 Disabled, 36 Compute and the 1 Audio "tacked on" in the APU design.

Hopefully it's 1 or 3.
The audio engine is confirmed to be independent.
 
Oct 25, 2017
5,837
I started buying more games over the last year because PS4 BC was confirmed, now it appears most of my games may not be playable which sucks. I really hope Sony/Cerny clears things up, maybe they wanna drop a bombshell in their games related event and confirm BC works for 99% of games but so far they only have 100 working with boost specs or something. I'll give them the benefit of the doubt until the next PS4 event.
 

Calo☆

Member
Jan 1, 2020
409
We’re currently evaluating games on a title-by-title basis to spot any issues that need adjustment from the original software developers.
That’s some lousy “backward compatibility” there. Still as long as we don’t need to pay for the same game twice I think we’re good.

For games whose developers never bothered to even enhance for PS4 Pro (e.g. Bloodborne), if they do run into issues it’ll be very unfortunate.
 

anexanhume

Member
Oct 25, 2017
7,367
Has this been posted before? This is interesting. An aggressively overclocked 5700 (36 CUs) performs (very, very) slightly worse than stock 5700 XT. When you calculate the Teraflops rating the overclocked 5700’s is 10% higher than that of the 5700 XT, yet the real world performance is only on par.

If we assume PS5’s GPU indeed spends most of its time at boost frequency, this suggests that the real world performance may be 10% lower than a 10.28TF RDNA 2 GPU of different configuration. (it’ll still be faster than a 5700XT because we know RDNA 2 has some IPC improvements)

I guess the reason for this discrepancy is memory bandwidth, although the L2 on the overclocked card is supposedly faster, when it comes to bandwidth limited work, it just can’t be fed fast enough.
Maybe this is why Flute had higher RAM bandwidth.
That’s some lousy “backward compatibility” there. Still as long as we don’t need to pay for the same game twice I think we’re good.

For games whose developers never bothered to even enhance for PS4 Pro (e.g. Bloodborne), if they do run into issues it’ll be very unfortunate.
This is the best case scenario based on the statements previously made. Most games will be compatible, and many games can be boosted.
 

III-V

Member
Oct 25, 2017
8,049
You mean all the games will be playable in legacy mode but possibly not in boost mode, just as described by Cerny?

 
Oct 27, 2017
4,642
Somewhere South
BC talk reminds me of the Boosted Mode for the PS4 Pro. We were told not all games would work. Vast majority of the games ran just fine, with just a few running into very minor problems and a handful that didn't work properly. BC situation is probably similar, just them being overzealous.
 

Calo☆

Member
Jan 1, 2020
409

III-V

Member
Oct 25, 2017
8,049
Is Flute PS5’s APU? Because we know PS5’s memory bandwidth already, it’s 448GBps, the same as 5700 and 5700 XT.
This was curious. In this sense there was actually a downgrade in RAM. But of course, we are not really sure. Many tests showed 448 GB/s, like we ended up with.
It looked like possibly testing with de-tuned 18 GB/s modules. Perhaps had to scale down for economic reasons.
 

AegonSnake

Banned
Oct 25, 2017
7,534
thank god. only took them 2 days. progress!

this gives me hope. this proves cerny isnt clueless lol. this also bodes extremely well for the other things he talked about. hype rising.

also, while team github might have won the round, turns out the native tests at 2.0 ghz were indeed running BC tests like the anti-githubbers were saying all along.
 

doncabesa

Banned
Jun 27, 2018
1,122
User banned (1 week): Concern trolling
You mean all the games will be playable in legacy mode but possibly not in boost mode, just as described by Cerny?

Expected to be, eventually. They still didn't clearly state when they would all work and that 100 or so are expected to be ready by launch. This is the same thing that we already knew. Their BC isn't everything day 1, it's upwards of 100 games at launch and then we don't know how long before the rest.

" We’re currently evaluating games on a title-by-title basis to spot any issues that need adjustment from the original software developers." Not just a plug and play solution unfortunately.
 

Calo☆

Member
Jan 1, 2020
409
This was curious. In this sense there was actually a downgrade in RAM. But of course, we are not really sure. Many tests showed 448 GB/s, like we ended up with.
It looked like possibly testing with de-tuned 18 GB/s modules. Perhaps had to scale down for economic reasons.
Flute is to Oberon as Gonzalo was to Ariel. Ariel had PS5 bandwidth, but Flute had ~528GB/s.
Curious indeed. Looks like they ultimately scaled it back.
 
Oct 27, 2017
4,642
Somewhere South
Expected to be, eventually. They still didn't clearly state when they would all work and that 100 or so are expected to be ready by launch. This is the same thing that we already knew. Their BC isn't everything day 1, it's upwards of 100 games at launch and then we don't know how long before the rest.
That's not what they're saying at all. All they're saying is that they've tested and benchmarked the top 100 games and most of of them work perfectly, and that they're going to be testing the rest of the games until the console launches. Testing, not making them work. BC here isn't a gated thing, it's a finding the stuff that needs tweaking thing.
 

III-V

Member
Oct 25, 2017
8,049
Expected to be, eventually. They still didn't clearly state when they would all work and that 100 or so are expected to be ready by launch. This is the same thing that we already knew. Their BC isn't everything day 1, it's upwards of 100 games at launch and then we don't know how long before the rest.

" We’re currently evaluating games on a title-by-title basis to spot any issues that need adjustment from the original software developers." Not just a plug and play solution unfortunately.
Nah. It is like I said. The vast vast majority will be running at launch, not 100. thousands. They are speaking on the ultra-boost mode.
 

AegonSnake

Banned
Oct 25, 2017
7,534
BC talk reminds me of the Boosted Mode for the PS4 Pro. We were told not all games would work. Vast majority of the games ran just fine, with just a few running into very minor problems and a handful that didn't work properly. BC situation is probably similar, just them being overzealous.
yeah, the problem is that the gpu clock was only 111mhz higher. or roughly 15% higher.

we are now going up to 2.0 ghz from 911 mhz. or 800 mhz for games like bloodborne that never got a pro patch. thats a massive almost 150% jump. what happens to games with unlocked framerates like dark souls 3 and god of war's 1080p 60 fps mode. if the devs didnt bother to lock the framerate to 60 fps, we could see the framerate in the 120s and you wonder if console ports can handle that framerate without any dev tinkering.
 

doncabesa

Banned
Jun 27, 2018
1,122
That's not what they're saying at all. All they're saying is that they've tested and benchmarked the top 100 games and most of of them work perfectly, and that they're going to be testing the rest of the games until the console launches. Testing, not making them work. BC here isn't a gated thing, it's a finding the stuff that needs tweaking thing.
Nah. It is like I said. The vast vast majority will be running at launch, not 100. thousands. They are speaking on the ultra-boost mode.
I hope you're right, and that they mean boost mode is the only thing that needs " We’re currently evaluating games on a title-by-title basis to spot any issues that need adjustment from the original software developers."
 

III-V

Member
Oct 25, 2017
8,049
yeah, the problem is that the gpu clock was only 111mhz higher. or roughly 15% higher.

we are now going up to 2.0 ghz from 911 mhz. or 800 mhz for games like bloodborne that never got a pro patch. thats a massive almost 150% jump. what happens to games with unlocked framerates like dark souls 3 and god of war's 1080p 60 fps mode. if the devs didnt bother to lock the framerate to 60 fps, we could see the framerate in the 120s and you wonder if console ports can handle that framerate without any dev tinkering.
Yeah, I have been thinking on that. They may end up throttling CPU/GPU for games running in native mode even.
 

III-V

Member
Oct 25, 2017
8,049
I hope you're right, and that they mean boost mode is the only thing that needs " We’re currently evaluating games on a title-by-title basis to spot any issues that need adjustment from the original software developers."
RDNA can run in GCN legacy mode, and the instruction set is nearly 100% backwards compatible. There could be a few that don't function properly, but essentially this is the entire premise behind having hardware-based BC.