Baseless Next Generation Rumors with no Technical Merits [pre E3 2019] *spawn*

Status
Not open for further replies.
[...]

It wouldn't surprise me if the blade boards have high speed interconnects so that 2 or maybe all 4 boards can be stacked.

[...]

Would make sense, Microsoft is a member of all next generation interconnect consortiums after all (Gen-Z, CCIX, CXL and OpenCAPI) and with the exception of CXL AMD is part of those consortiums as well.

I think for ps5 monolith maybe the choice, even though it will also be used for game streaming.

Yeah, I'm thinking the same and I'm starting to believe that they really wanted to release in 2019 since a user mentioned the bink changelog: http://www.radgametools.com/bnkhist.htm

Back in 2012 they added two secret platforms which was ~16 months before the PS4 and Xbox One launched. Around the same time last year they "Added a new secret NDA-ed platform.", which could be the PS5 and would fit a ~november 2019 launch. They recently mentioned added support for Stadia which could be this NDA-ed platform. On the other hand they never mentioned that they added PS4 or Xbox One support in the past, I could only find "Added PS4 and Xbox One docs." with ctrl+f.

They also never mentioned "Added Nintendo Switch support" but at the same time they added a secret platform in 2016 which could fit with the Switch launch in 2017. The first thing they mention about the Switch is "Updated to Nintendo Switch SDK 3.0 [...]", so it would be weird if Stadia was the NDA-ed platform and then get's mentioned again with "Bink now has a Stadia target" but consoles get never mentioned.

Also, add to that the rumors about Navi problems, the rumors about AMD RTG dedicating 2/3 engineers to Sony, the AMD slide where Navi was targeted at 2017 (but will only launch later this year) and the rumor about AMDs 70% yield for a tiny ~74mm² chiplet on 7nm, which could translate to ~30% yield for a die with 300mm² or more like a monolithic PS5 APU would probably be.

Unlike AMD who can sell their datacenter GPUs for multiple thousands and use the worse ones as a high priced gaming and creator card thanks to Nvidia, Sony would have a huge problem with such a bad yield.

Maybe they just said "forget it, we pay the 20 million (?) dollar per mask and go for 7nm EUV at the end of 2020 to increase the yield", the PS4 is still selling really well anyway.

The biggest problem is in the pc space. Server and console not so much.
Servers probably already have to handle mgpu type set ups.
DirectX already has mgpu functionality.
It may not be invisible, but it may not be as big a deal in a static box. Patch unity, unreal, other engines to handle it in a basic way which may leave some performance on the table, but be nice fallback.
For MS the loss in performance may be worth it for the overall benefits.
Just waiting to be told that I'm 100% wrong.

Yeah, you could be right (I have no clue and am waiting to be told that's wrong as well, heh). Another question would be how they deliver the memory bandwidth to the chiplets? Would the chiplets have their own memory (HBM?) or would they connect to an IO die? In the latter case they would need an interconnect faster than current Infinity Fabric (100 GB/s or 200 GB/s with two links). Gen-Z claims it can deliver such bandwidth.

Is the GPU's actually very different or is it just disabled?
I'm wondering how much actual die saving is there if you don't have double precision?
Edit :
It seems that AI and ML tend to prioritize lower precision not higher, which is one of the use cases Phil gave. So maybe not having FP64 may not be a big deal for the intended work loads.

As far as I know they are different and don't have the silicon for it. For example Vega 10 (64, Instinct Mi 25 etc.) has not a single card which can deliver good double precision performance, all Vega 10 cards only have 1:16 double precision. But they can also decide to actually limit the double precision rate even though the silicon is there, as can be seen on Vega 20 (Radeon 7).


To be honest I don't get it how this rumor is discussed so much. This person is a "third party small developer from EU" but for some reason he knows the roadmap/plans of big shot studios like Take 2 (GTA 6), Guerilla Games (Horizon), EA/Dice (Battlefiled), UbiSoft (Assassins Creed) and Sonys remaster plans.

Then there is the unrealistic price for PSVR2 which would technically compete with StarVR that sells for 3200 dollars. How would a "third party small developer from EU" acquire such information? Just doesn't compute for me.

Deep down I'm chearing for this rumor though, since PSVR2 would be just like StarVR and eye tracking would allow for foveated rendering to bring down the hardware requirements. In a console environment where developers can take full advantage of everything and know exactly what hardware the users have this would be just too cool. Especially at this unrealistic price point of 250 dollars this would win Sony the entire consumer VR market, and I'm sure many businesses would choose the PS5 as well. Heck, Sony could probably sell a much more expensive business edition of the PS5 which includes support contracts etc.

Right? Why would these systems need 2 to 3 times as much memory? It doesn't sound necessary to me. I think ~16GB with cery high bandwidth would make more sense... These leaks state some obvious things and then some things which sound reasonable until you dig a little deeper.

Unless they use HBM or the consoles don't have 10+ TFLOPs then bandwidth for the GPU and CPU. If the 50 GB/s bandwidth per TFLOP "rule" from current consoles still stands you would need 500 GB/s for a 10 TFLOPs console. Maybe even more since ZEN is much faster compared to Jaguar.

8 or 16 GB of currently available GDDR6 can deliver ~384-512 GB/s on a 256 bit bus which would be enough for 10 TFLOPs. 12 or 24 GB with a 384bit bus would be 672-768 GB/s which would be enough for 13-15 TFLOPs. So if the "rule" still stands and they want to go over 10 TFLOPs then they need more bandwidth.

16 GB of HBM2 on the other hand could be anything between 205 GB/s and 1640 Gb/s but I doubt we see HBM2 in this generation of consoles.

Phil Spencer said the hardware would be put to use for "machine learning and other non-entertainment uses".
AFAIK machine learning has little use for DP, and all other purposes could exclude scientific workloads that need DP.

Just think of all the professional uses that Vega 10 got, such as ML on Alibaba's servers, video rendering, offline 3d rendering, etc.

Just because they'll be renting the servers for other uses, it doesn't mean they need their servers to be capable of all uses.

Yeah, you're right, I assumed all or nothing and ignored that there could be something in between. :O
 
Last edited by a moderator:
Well shit, scratch that then. The way I read it it looked like the E3 no show was part of the rumour.

Yeah without that I wouldn't put much credence to that rumour. Weren't we discussing a potential flash hybrid type system so that the consoles wouldn't need much ram here on beyond around that time?

I think Shifty was a big proponent of it.

Well there was that reddit leak the day before the news broke which is probably what you are thinking of.

And going back and reading that reddit thread the leaker also mentions PS5 being 8 core Ryzen, that there is a camera in the DS5, no VR breaker box, small reveal in mid 2019 and some more stuff that a couple of weeks later ends up in the pastebin leak...

I dunno but right now that pastebin leak seems to me like a compilation of that information combined with some other reported rumors, some educated guesses and some imagination.
 
There's no hard evidence or reliable source behind it. At the moment it's just another unsubstantiated rumour.
 
I have been wondering more about how Microsoft might try to design their family of devices. I feel the idea of base<premium<server strange in the fact the server will probably be wanting to stream at 1080 for bandwidth reasons and the slight apex boosts don't seem compelling for a server.

Microsoft are using the S in the cloud ATM which is the smallest, I would assets this is because bandwidth and compression ruin the X potential and they want higher density. I also think console memory quantity might be low and in general it would not be too compelling a product, unless they can furful its use internally or there are as hoc tasks that folks pay for that are just slotted into spare capacity.

I have been wondering if we should consider Microsoft have started with a more data centre design and scaled down.

Watching the latest adored TV video Jim talks about the new super computer AMD are creating. It seemed to me if we look at that 4 GPU one CPU idea that could be 4 streams from one instance. The CPU as a single chiplet for the console could easily scale up to 4 or more chiplets to be more data centre ready.

Memory as in that video be hbm2 as well as ddr4, we have seen from S to X that Microsoft can deal with differing memory setups providing the bandwidth is provided, an IO die and VM can present things in an expected fashion even if the reality is different, and I don't feel expediting data centre hardware to be able to meet the base level console as being too hard. SDK can also help as this is being devised form the launch and not retro fitted like the S in the current cloud.

Anyhow I am not technical but as we discuss 4chan stuff I don't feel to bad putting this up there.

This is somewhat guided by the 8 page rumor that made me think about multiple streams from single hardware and how that could better fit the cloud.
 
If MS are going with a CPU chiplet design, it's conceivable that the main chip can handle two (or more) CPU chiplets. That'd give you the option of one 4k stream, or two/three 1080 streams, and give more CPU for whatever azure workloads you're running.

A full 64cu chip with 32 GB of HBM2 would even beat out Mi60, and with total volumes produced far higher than Vega 7 / Mi50 / Mi60 costs for MS could work out lower than for competitors.

It's easy to see why the Azure team would be up for this. They get the cream of the crop processors (cu yeild / frequency / power) with costs subsidised by the purchase of tens of millions of chips.
 
What I'd like to see
PS5 monolithic apu, tailored zen 2, custom navi with ID buffer 2.0, 62 cu, etc

Scarlett mcm chiplet apu, of the shelf zen 2+, custom navi gpu 32 cu (chiplet design) , AI accelerator.

Lockhart targeted 1440p-1080p
Zen 8c
Navi 30 enabled cu, 2 disabled (for yield)

Anaconda targeted 2160p
Zen 8c
Navi 2 full 32 enabled cu, 1 30 enabled

Cloud
2 zen 8c
4 32 enabled cu
Streams 2 Lockhart level (2 32 cu) , during azure access all 4 32 cu. Not sure if would be better just to go with 2 32 enabled cu instead of 4.

Games developed for Anaconda using dx mgpu, just runs on Lockhart but may need slight tweak in performance profile if resolution change not enough.

Very out there, but be nice to see a relatively radical design.
 
I had not considered carving up a GPU also, in that case a decent server CPU could feed many streams. Good density in the data centre. That might also allow blades to cache a game for faster loading and general ease of support.

I doubt the hardware components themselves will change much, but as the Xbox architecture uses a vGPU there could be other server side tweaks, perhaps unlocking the more data centre orientated features that are not present in retail hardware.

This generation could get very interesting tech and business choice wise.
 
It's the day after MS conference, so I imagine MS does a brief announcement of Scarlett family with maybe some simple Tflops figures then AMD goes in for a deep dive right after. Question is what will Sony do?
 
I doubt MS will let anybody else do a deep dive. Maybe an interview with DF or something or appear on their stage for a second or two.
Their at the show, they have their own inside Xbox shows also to talk to Phil and engineers if they want to cover more.

AMD have Sony, stadia, pc to cover at different degrees of details.
Even then they would need to be careful not to make any sound better or worse.
So will definitely be interesting to see, how and what is covered
 
It's the day after MS conference, so I imagine MS does a brief announcement of Scarlett family with maybe some simple Tflops figures then AMD goes in for a deep dive right after. Question is what will Sony do?
You think MS will announce specs this early?
 
https://winfuture.de/news,108881.html

$499 PS5 November 2020

Bis zum Start der neuen Sony-Konsole PlayStation 5 müssen sich Fans noch einige Monate gedulden. Dennoch werden die Gerüchte rund um den Preis und den Release der PS5 immer konkreter. Nun äußerte sich ein japanischer Analyst mit einer weiteren Prognose.

Während sich die letzten Spekulationen vor allem mit der Hardware der geplanten Next-Gen-Konsole beschäftigten, dürften für Spieler neben namhaften Exklusivtiteln vor allem die Kosten und der Erscheinungstermin der PlayStation 5 eine große Rolle spielen. Zu denen hat jetzt der japanische Analyst Hideki Yasuda vom Ace Research Institute genauere Vorstellungen. Er geht davon aus, dass Sony die PS5 im November 2020 zu einem Preis von 499 US-Dollar auf den Markt bringen wird.
 
Anaconda targeted 2160p
Zen 8c
Navi 2 full 32 enabled cu, 1 30 enabled
This would make the Anaconda crazy powerful if I attribute a TF to Lockhart.
And I expect Lockhart to be around the 5.5 - 6 TF. Not for marketing reasons, just so it can play 1X games in BC mode.

So Anaconda would have 2 * 32 cu gpu chiplets not 3. So over twice as powerful not 3. Otherwise talking min 15TF. That would be nice but possibly to far.
 
Status
Not open for further replies.
Back
Top