Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]

[N2O]

There are other ways to sell a console other than specs alone. New generations are about features; not just the increase in power. Right now everyone is declaring a winner based on power; you've not even heard what they actually have planned for their ecosystems.

But doesn't mean they have to give PS5 the slight edge when it's possible make it same power.
And just because they boost the clock a bit doesn't mean will make them have to cancel a random new feature, it's not CRPG game.

Also, Phil Spencer clearly want the power crown according to the interview.

 

[Deleted member 13524]

New Seems like AMD has booked the shit out of 7nm waffers at TSMC. Next year Apple is moving to 5nm and with console designs, AMD will become biggest TSMC customer.

If I put that site on google translate, the title says Supermicro is going #1 on 7nm, not AMD.
Then on the picture legend, it says AMD is the one going #1.

Regardless, this is pretty much a confirmation that apple is moving to 5nm for A14 and a sign that Snapdragon 865 + 765 might not be in as much demand as Qualcomm had hoped (also another sign that Samsung might not be using Snapdragon for the US versions of their 2019 S and A series).

The answer is that if they manage to hit 12TF for release silicon, it will hit _their_ business objectives. It's not about besting Sony. Once people understand that, you'll understand why this whole discussion about who will have more power is nothing more than a projection of fans hoping to come out on top. Once people understand that, you'll understand why this whole discussion about who will have more power is nothing more than a projection of fans hoping to come out on top.

Well then Phil Spencer sure is trying hard to fool us all by constantly claiming they'll have the faster console of the next generation. And the rethoric of comparison isn't even something he started with Scarlett:

https://twitter.com/x/status/867050680149606400

There's nothing wrong in claiming you have the best performing hardware if it's true. That factor alone moves a number of console sales higher than 0, that's for sure.

Right now any EUV is expemsive, not high enough volume, and whoever wants it will pay a lot more for it. So it might be much less expensive to spend a bit more on cooling than trying to compete against apple for the available EUV volume.

We might see 7nm+ being skipped for a 5nm or 3nm slim revision, by then the EUV mask issues might be solved, and they'll have much more powerful frickin lazer beams.

Huawei sold millions of Mate 30 5G phones using a TSMC 7nm EUV (N7+) SoC for the last 3 months. It's been on high volume production for a whole quarter now.
Xiaomi will start selling the K30 5G with the Snapdragon 765G which uses Samsung's 7nm EUV and that's a mid-range phone, let alone all the Samsung Note 10 units sold using the 7nm EUV Exynos 9825.
That said: EUV is fine.

Sony or Microsoft probably don't have to "pay a lot more" for N7+ since they're making multi-year, tens-of-millions of chip orders deals.


Regardless, I'd say it's more possible that both SeX and PS5 dev kits and first production models are using relatively high-binned N7P SoCs that will quickly transition to lower-binned chips on N6 during the second half of 2020.
So despite seeing chips that are probably pushing the boundaries for N7P (especially that Oberon clocking at 2GHz), we should be looking at chips that will reallistically be made on N6 EUV during most of their lifetime.



Fun fact BTW:
- N6 offers 18-20% higher transistor density than N7/N7P. This means that a 355mm^2 chip made on N7P chip will be 301mm^2 on N6.
Now where have I seen those ~300 and ~350 mm^2 die size numbers... hmm...

 

[Esrever]

It is AMD's first RT implementation so wouldn't be surprised.

I really don't see how that is an excuse when it's being compared to Nvidia's first implementation, especially since Nvidia was doing their thing before Microsoft fully implemented DirectX RT. If anything, barring any significant problems, I would expect AMD to do better than Nvidia's first gen and probably be eclipsed by nvidia's second gen.

 

[AbsoluteBeginner]

On the PS5 leaks, have you see that RT performances is announced between 2060 and 2070 ? Not very good if true.

Any link on this?

 

[PSman1700]

I would expect AMD to do better than Nvidia's first gen and probably be eclipsed by nvidia's second gen.

Maybe, magic does happen afterall.

 

[AbsoluteBeginner]

Pretty sure someone who knows chip size going into TSMC fabs now would also know which node its using. Besides this, if we are to believe report about Oberon 300 vs Arden 350mm2 from our Taiwanese friend, we could also assume, with relatively high certainty, that AMD leaked doc specifying Oberon has to be true as well.

Which would also match with difference between 36CU and 56CU. So, Occam's razor and all...

 

[PSman1700]

DF came to the conclusion it is PS5, so 9TF PS5, GDDR6 (amount?), most likely NVME as MS has it, AMD first gen RT and maybe MS has advantage regarding VRS also.

 

[iroboto]

Well then Phil Spencer sure is trying hard to fool us all by constantly claiming they'll have the faster console of the next generation. And the rethoric of comparison isn't even something he started with Scarlett:

There's nothing wrong in claiming you have the best performing hardware if it's true. That factor alone moves a number of console sales higher than 0, that's for sure.

Having the power crown is great sure. Its about all they want to talk about for now. there will be additional factors for purchasing decisions later down the line. And what I'm getting at is that I don't believe MS or Sony will be willing to compromise everything just to earn the power crown. There are and will be other factors in play.

 

[Globalisateur]

If I put that site on google translate, the title says Supermicro is going #1 on 7nm, not AMD.
Then on the picture legend, it says AMD is the one going #1.

Regardless, this is pretty much a confirmation that apple is moving to 5nm for A14 and a sign that Snapdragon 865 + 765 might not be in as much demand as Qualcomm had hoped (also another sign that Samsung might not be using Snapdragon for the US versions of their 2019 S and A series).



Well then Phil Spencer sure is trying hard to fool us all by constantly claiming they'll have the faster console of the next generation. And the rethoric of comparison isn't even something he started with Scarlett:

https://twitter.com/x/status/867050680149606400

There's nothing wrong in claiming you have the best performing hardware if it's true. That factor alone moves a number of console sales higher than 0, that's for sure.



Huawei sold millions of Mate 30 5G phones using a TSMC 7nm EUV (N7+) SoC for the last 3 months. It's been on high volume production for a whole quarter now.
Xiaomi will start selling the K30 5G with the Snapdragon 765G which uses Samsung's 7nm EUV and that's a mid-range phone, let alone all the Samsung Note 10 units sold using the 7nm EUV Exynos 9825.
That said: EUV is fine.

Sony or Microsoft probably don't have to "pay a lot more" for N7+ since they're making multi-year, tens-of-millions of chip orders deals.


Regardless, I'd say it's more possible that both SeX and PS5 dev kits and first production models are using relatively high-binned N7P SoCs that will quickly transition to lower-binned chips on N6 during the second half of 2020.
So despite seeing chips that are probably pushing the boundaries for N7P (especially that Oberon clocking at 2GHz), we should be looking at chips that will reallistically be made on N6 EUV during most of their lifetime.



Fun fact BTW:
- N6 offers 18-20% higher transistor density than N7/N7P. This means that a 355mm^2 chip made on N7P chip will be 301mm^2 on N6.
Now where have I seen those ~300 and ~350 mm^2 die size numbers... hmm...

N6 apparently doesn't improve power consumption. The only advantage being the higher transistor density. So it won't really help with high clocks. 7nm EUV will give about the same transistor density gain than N6 + 10% power reduction and maybe some performance improvement (depending of the source).

Besides, N6 won't be ready for mass production in 2020. While 7nm EUV is already produced in high volume. Finally we know AMD is going to use 7nm EUV for all its future products planned allegedly for....this year: Zen 3 and RDNA2 (I am not saying PS5 will use RDNA2).

 

[PSman1700]

.this year: Zen 3 and RDNA2 (I am not saying PS5 will use RDNA2).

Zen 3, RDNA2, would be nice Somewhere Zen2 is kinda old already.

 

[Deleted member 13524]

Pretty sure someone who knows chip size going into TSMC fabs now would also know which node its using. Besides this, if we are to believe report about Oberon 300 vs Arden 350mm2 from our Taiwanese friend, we could also assume, with relatively high certainty, that AMD leaked doc specifying Oberon has to be true as well.

Which would also match with difference between 36CU and 56CU. So, Occam's razor and all...

Occam's razor says this github log dump leak without context provided is 100% accurate and says 100% everything there is to know about the PS5's hardware that can't possibly change between the leak's date of mid-2019 and the console releasing 18 months later?

Where exactly is this miraculous log dump? I see a lot of articles and forum users commenting on it but not the document itself anywhere.

And what I'm getting at is that I don't believe MS or Sony will be willing to compromise everything just to earn the power crown. There are and will be other factors in play.

Who said anything about compromising everything?

N6 apparently doesn't improve power consumption. The only advantage being the higher transistor density. So it won't really help with high clocks. 7nm EUV will give about the same transistor density gain than N6 + 10% power reduction and maybe some performance improvement (depending of the source).

EUV itself doesn't improve power consumption. It improves yields, which is why I mentioned the difference between higher-binned and lower-binned chips.

Besides, N6 won't be ready for mass production in 2020.

Which is why I suggested that the devkits and first batches of production models will be using higher-binned N7P chips, with the transition to lower-binned N6 being made between 2020 and 2021.
BTW, mass production N6 is set for last quarter of 2020.

 

[PSman1700]

date of mid-2019 and the console releasing 18 months later?

It actually fits perfectly in the timelines, august or what it was, production starts soon. Dev kits better be near final specs half a year before production starts etc.

 

[MrFox]

Having the power crown is great sure. Its about all they want to talk about for now. there will be additional factors for purchasing decisions later down the line. And what I'm getting at is that I don't believe MS or Sony will be willing to compromise everything just to earn the power crown. There are and will be other factors in play.

There's a range of options, if they go too high power they compromise power consumption, form factor, and specially retail price. If they go too low price they obviously compromise on power.

When Sony and MS ask AMD to make a zen/rdna custom SoC for a 2020 release, they can both chose anything in the power envelope available. Sony could make a 13 or 14TF console whenever they chose to, and the cost would be very high. They chose not to. MS could make a more balanced cost/performance, they chose the power crown.

Same thing happened with XB1X versus PS4P, delay a year and raise the price $100 for the power crown. It doesn't comes free. But this time it doesn't have a year delay for process improvement, so the difference is smaller for what we're assuming will be probably be a $100 higher cost.

That nvme is going to cost, so I have no idea how the expectation of 399 ps5 and 499 xbsx is possible. I'm expecting more like 499 and 599 respectively.

 

[AbsoluteBeginner]

Occam's razor says this github log dump leak without context provided is 100% accurate and says 100% everything there is to know about the PS5's hardware that can't possibly change between the leak's date of mid-2019 and the console releasing 18 months later?

Where exactly is this miraculous log dump? I see a lot of articles and forum users commenting on it but not the document itself anywhere.

So I think I need to clarify Github leak before I explain Occam's razor part

• Github officially came from AMD employee (contained alot of employee data, CV etc. now deleted from twitter and Github)
• Docs posted referenced future AMD GPUs and SOCs (among them Renoir, MI100, Arden, Oberon and Navi10)
• Besides Navi 10, not one of them has been released
• Github document describes, to very high level, Oberon chip
• Oberon chip features 36CUs and 2.0GHz in its native configuration
• Oberon chip also features GEN1 and GEN0 mode, with 36CU/911MHz/216GB/s bandwidth for GEN1 and 18CU/800MHz/176GB/s for GEN0
• Github document specifies another chip - Arden, featuring Xbox One and XBX BC modes (4RBs)
• Arden document also contains reference to 3584SPs - meaning 56CUs and RT/VRS

So, having said all that we can agree on following :

1. Github leak, while 6+ month old, features chips once designated for next gen Playstation and Xbox
2. Github leak specifies these chips as Arden and Oberon
3. Arden has 56CUs 320bit bus, Oberron 36CUs 256bit bus
4. Well known leakier from Taiwanese forum - AquariusZi, reported Arden chip being "one size larger" (~50mm2) then Oberon
5. AquariusZi also confirmed Arden is Xbox, while Oberon chip is PS5
6. His opinion was that 50mm2 difference was duo to RT (300 v 350 mm2)

With all this, Occam's razor would be this :

Confirmed AMD Github leak from June specifies Oberon as PS5 chip > This chip is, from data we have, smaller then Arden (Xbox) chip > Taiwanese leaker close to TSMC says Oberon is PS5 and Arden is Xbox > Says difference in die size is around 50mm2.

What would be conclusion based on all this? AMD's leak was legit and on point, as was AquariusZi info about size difference. Size difference can be explained by 36CU vs 56CU configuration. Thats it.

Your hypothesis might include Oberon being on 6nm instead of 7nm. But this would mean that not only AMD data is outdated and not representative, and the chip was redesigned, it would also mean leaker's assumption on what is causing 50mm2 difference is completely wrong and he didn't know one console was on completely different node. It would also mean that AMD Navi announcement, which specified Next Gen PS as "7nm chip based on RDNA arch" is wrong.

So, pretty much, everyone would be wrong when all points to opposite. Oberon is smaller chip duo to less compute units.

*Leak is now deleted, I doubt anyone will share it publicly now.

 

[Deleted member 13524]

It actually fits perfectly in the timelines, august or what it was, production starts soon. Dev kits better be near final specs half a year before production starts etc.

What timelines?
Production needs to start a whopping 11 months before the console goes to the shelves? Devkits need to contain final hardware 18 months before release? Who said that?
People seem to be getting super convinced of these ultra long periods of time needed for production before the consoles go on sale, as if we're living before the 4th industrial revolution ever happened and mass automated production hasn't been implemented.

Are we also assuming the iphone 12 has already started production in December 2019 for a release 11 months later in October 2020? With a N5 SoC that has barely any test chips available yet?


There's this new wave of super conservationism in hardware development and production that I honestly don't get.

*Leak is now deleted, I doubt anyone will share it publicly now.

Are you fucking serious?

No one. On the entire Internet. Thought of hitting copy+paste / download / screenshot of the github leak. And the leak - which is completely unavailable to the public and can't be fact-checked or studied - is now taken as gospel.

This is hilarious.
I'm sorry for the language people, but this is hilarious.

 

[PSman1700]

*Leak is now deleted, I doubt anyone will share it publicly now.

Which means it was something.

 

[iroboto]

There's a range of options, if they go too high power they compromise power consumption, form factor, and specially retail price. If they go too low price they obviously compromise on power.

When Sony and MS ask AMD to make a zen/rdna custom SoC for a 2020 release, they can both chose anything in the power envelope available. Sony could make a 13 or 14TF console whenever they chose to, and the cost would be very high. They chose not to. MS could make a more balanced cost/performance, they chose the power crown.

Same thing happened with XB1X versus PS4P, delay a year and raise the price $100 for the power crown. It doesn't comes free. But this time it doesn't have a year delay for process improvement, so the difference is smaller for what we're assuming will be probably be a $100 higher cost.

That nvme is going to cost, so I have no idea how the expectation of 399 ps5 and 499 xbsx is possible. I'm expecting more like 499 and 599 respectively.

absolutely. You take all your goals and you aim for a design goal. But if for some reason your competitor comes out on top by like 5%, you're not going to find a way to upclock your chip to best them is all I'm saying. You're going to take the L and move on. That 5% at the very last minute is not easy to compensate for if your whole design and costing is set for a specific clock speed.

 

[PSman1700]

What timelines?

Half a year before release, production starts, somewhere. So between this leak and production is about half a year, somewhat more or less. You would think specs are about final half a year before production, right? Testing would take many months i assume.

 

[AbsoluteBeginner]

What timelines?
Production needs to start a whopping 11 months before the console goes to the shelves? Devkits need to contain final hardware 18 months before release? Who said that?
People seem to be getting super convinced of these ultra long periods of time needed for production before the consoles go on sale, as if we're living before the 4th industrial revolution ever happened and mass automated production hasn't been implemented.

Are we also assuming the iphone 12 has already started production in December 2019 for a release 11 months later in October 2020? With a N5 SoC that has barely any test chips available yet?


There's this new wave of super conservationism in hardware development and production that I honestly don't get.

Dev kits, such as "V" ones, almost certainly contain actual early APUs (instead of PC GPU/CPU combo described as "big silver tower").

 

[iroboto]

What timelines?

https://www.resetera.com/threads/ne...ark-tower-see-staff-post.159131/post-27749164
ex Product Manager for the Xbox spells out approximate timelines here.

quote"""
I would define "tapeout" as the ready-to-manufacture silicon. This would be final production ready silicon that is ready to mass-produce in volumes.

That generally happens very shortly before launch - typically less than a year. In fact, Scorpio tapeout IIRC happened in March 2017. Bringup (or the first piece of manufactured silicon that can be put in prototype boxes) came in December 2016 (so about a little less than a year before launch).

Phil has tweeted that they had takehome units shortly before Christmas, which by my estimation would mean they are past bringup, but prior to tapeout, but actually ahead of schedule relative to Scorpio which was a pretty smooth process. So I would say things seem to be going well.

Highly unlikely that any changes would be able to be done with either console at this point, unless two things are true.

1. The exterior case and cooling design can handle whatever increases they would like to implement (could be CU's, could be clock speed but either way it's more heat)
2. They are seeing much better yields than anticipated and can meet their launch volume targets and long-term cost reduction goals.

Basically, at this point the consoles that are shipping are the ones that both companies have been planning to ship for at least a year, unless they delay the launch (and eat a ton of sunk costs) or had build in the ability to react to any bumps in performance in advance.
"""