Predict: The Next Generation Console Tech

Status
Not open for further replies.
I think your estimates are too high for what would be needed for dram stacks:


Even with ddr4 you wouldn't need much more than 100mm2 of interposer area for 2 stacks. That then is an interposer in the 425-450mm2 range and a mobo around the size of an index card (10cm x 15cm).
You are right. If one keeps it down to two stacks, one doesn't need a lot of space (can be arranged one one side of the SoC die). I was more thinking of the maximal configuration with 4 stacks on two sides (and still overestimated the needed area, when I look at the pictures with that prototype APU/GPU on an interposers with two DRAM stacks again, one can pack it quite tightly).
From what I understood lately, this sums up the whole dram problem. The major part of the die which is the dram arrays is pretty much the same things for GDDR5, DDR3, DDR4, WideIO, etc... it's the IO, buffer, interface, and signaling which are different, power hungry, inefficient, and needs to be a standard in order to have millions of identical chips produced, competition kicks in, and the cost goes way down. If you change or remove the IO part of GDDR5 it's basically an expensive custom chip.

HBM is similar in performance to what you describe, but it's a better choice because it's a standard. A single layer will be 1024bit wide, and 256GB/s per chip (1066 DDR), or lower power at 128GB/s per chip (would be either 1066 SDR or 533 DDR). There's no reason to make a custom memory chip, the best solution is right there in a JEDEC standard.
Actually, I described the rough concept behind HBM/WideIO2. WideIO uses afaik four 128bit channels per stack/device running at 266 MHz SDR signaling to save on area and power for a high speed serial interface (one needs this twice, on the memory die or the logic layer of a HMC and on the SoC again, WideIO/HBM tries to get around that by "directly" transfering the wider internal accesses over wide and slower SDR interfaces). WideIO2/HBM is supposed to double the width (to 1024 bit, 4x256Bit) and upping the speed considerably. It lines up quite nicely with the 1024bit (4x256Bit) at 750MHz SDR signaling I made up from thin air (taking the width and speed of the DRAM arrays of typical GDDR5 as base, HBM would be made by "removing" the serial IO parts). The DRAM is capable of that and a preliminary or minimal implementation of the upcoming WideIO2 standard could probably be used already.
 
1.8 TFLOPs vs 1.2 on Durango, or as bg said, almost 2 TF for Orbis and 1+ for Durango.

Actually, I was talking to bgassassin on here (after we figured out what GPUs were in the leaked devkit) and he got those two pieces of info from different people and the one who reported 1+ for Durango could well have been trying not to be too specific.

We know for a fact that the Durango alpha kits have either a 2TF or 2.25TF GPU in them.

lherre has said Durango is 'more beast' than Orbis and the results of IGNs developer survey also show most devs prefer working on Durango than Orbis and think it will win next gen.

And Sony is really not in a position to break the bank on hardware this time around, it isn't Kutaragi's Sony anymore - neither financially or culturally (as can be seen by the Vita).
 
Actually, I was talking to bgassassin on here (after we figured out what GPUs were in the leaked devkit) and he got those two pieces of info from different people and the one who reported 1+ for Durango could well have been trying not to be too specific.

We know for a fact that the Durango alpha kits have either a 2TF or 2.25TF GPU in them.

lherre has said Durango is 'more beast' than Orbis and the results of IGNs developer survey also show most devs prefer working on Durango than Orbis and think it will win next gen.

And Sony is really not in a position to break the bank on hardware this time around, it isn't Kutaragi's Sony anymore - neither financially or culturally (as can be seen by the Vita).

if so then will 3rd party studios who make games for both platforms favour xbox3 with better visuals ,etc ? in my opinion they will not make two radically different versions of their software visually and it seems that the extra power of durango compared to orbis will only be used by 1st party studios which are far less compared to sony .
 
lherre has said Durango is 'more beast' than Orbis and the results of IGNs developer survey also show most devs prefer working on Durango than Orbis and think it will win next gen.


Dont necessarily have a dog in this fight, but lherre also told me via pm that more RAM, lesser GPU was a good summation of Durango vs Orbis (when I asked). That was about six months ago too.

People love to qoute lherre saying Durango more beast, it may well be, but I got that conflicting info from him.
 
if so then will 3rd party studios who make games for both platforms favour xbox3 with better visuals ,etc ?
They'll run the same visuals at better framerates and IQ. Cross-platform engines will be pretty scalable, so a significant difference in performance will allow Durango to run with better lighting too. It'll be exactly the same as PC, with the better GPU giving pretty much effortless advantages to the developers, and if both are on the same tech level, none will see software features dropped due to minimum standard. That is, both being DX11 or 12, software will be written to that standard instead of a lower standard failing to utilise features of the other console.

In short, one platform being more powerful than the other will show readily if they are both the same basic architecture. That doesn't help us solve the question of what's in the boxes, but I do understand a fair few people are incapable of separating the intellectual investigation from their interests in the business of competition and how boxes will or won't be 'better'.
 
In short, one platform being more powerful than the other will show readily if they are both the same basic architecture. That doesn't help us solve the question of what's in the boxes, but I do understand a fair few people are incapable of separating the intellectual investigation from their interests in the business of competition and how boxes will or won't be 'better'.

:D
 
And Sony is really not in a position to break the bank on hardware this time around, it isn't Kutaragi's Sony anymore - neither financially or culturally (as can be seen by the Vita).

I dont know what to say about the Vita, but it was and is still considered a very powerful mobile platform, so I am not so sure if Vita is any indication for a change in sony's vision regarding console hardware.

I will ask a question, where a good answer could give us an indication if sony changed or not their vision : PSP at its time (representing the ps3 era ?) VS Vita at its time (representing the new ps4 era ?), which console was more advanced technologically ?
 
They were both the most powerful handheld devices of their era. The difference with Vita is there's a far bigger mobile market that refreshes hardware constantly, so it's been matched and overtaken in a way PSP wasn't for much of its life. The other major difference is Vita uses a more conventional core versus PSP, which shows less focus on custom hardware - something we already know is happening with PS4.
 
#1 AMD china guy another new hint
http://club.tgfcer.com/viewthread.php?tid=6593085&page=2#pid15931245
"嘿嘿,我家有块APU,可是384BIT总线的哦"
He said Durango APU will be 384bit

Would probably mean 6 GB console. 12 4Gb chips each with x32 IO. BW could be anywhere from 89 GB/sec (1.866 DDR3) to 154 GB/sec (3.2 DDR4).

A bus that big would be problematic IMO for die shrinks (unless APU is really, really big) and reducing memory chip count could be an issue.
 
So the "3M netlist" related about this?
From semiaccurate
http://semiaccurate.com/forums/showthread.php?p=173714#post173714

When I googling IBM 3M TSV then it makes sense :

See the timeline Clearly:

Aug 2011:
IBM - 3M partnered for 3D or 2.5D
http://www.dailymarkets.com/stock/2011/09/09/ibm-partners-3m/
http://www.i-micronews.com/lectureArticle.asp?id=7529

Sep 2011
IBM - Micron , IBM will do HMC for Micron
http://www-03.ibm.com/press/us/en/pressrelease/36125.wss

Dec 2011
Suprise the Fab they choose
http://www.advancedsubstratenews.co...debut-of-the-first-commercial-3d-chip-making/

IBM will produce Micron‘s Hybrid Memory Cube (HMC) in the debut of the first commercial, 3D chip-making. HMC parts will be manufactured at IBM’s advanced semiconductor fab in East Fishkill, N.Y.
Dec 2011 , Fab 8 & IBM in production
http://semiaccurate.com/2012/01/09/global-foundries-fab-8-is-making-chips/

Jan 2012, Oban In production in Fab 8 & GloFo, Large 550-600mm^2 die size
http://semiaccurate.com/2012/01/18/xbox-nextxbox-720-chips-in-production/#.UNzwT_JaeSo

Then MS Joint HMC (Formally at July-Aug 2012)

Ok, that's my thought, it fits
 
wow so Microsoft did finally find a solution from the mess that is AMD nowadays in the processor department ?

I'm not sure I understand where you're going with your comment? N2O was mainly talking about memory technology. BTW, that last semiaccurate link in N2O's post is pure bogus as far as I can tell.
 
Wow. 600mm2 die size, and the main memory have to be produced at East Fishkill... 299!!
Maybe it's a 600mm2 interposer, that would make more sense. But I read there were issues with large interposers because they were too fragile to handle. I don't know if 600 is considered "too large".
 
600mm^2 would be right at the reticle/mask limit for lithography tools.
What about organic interposers, the laminate stuff? It's crude, but if it's only to connect the memory to the SoC, they can do 55mmx55mm, which would allow a 600mm2 die plus it's memory.

I can't help thinking that there no such thing as a good-die anymore, it's always a mostly-good-die, where all chips have defects because we can no longer make large perfect dies. If the whole chip has plenty of Shaders and Cores, and all the fuse to disable non-working ones, maybe a 600mm2 die with lots of disabled stuff would be less expensive to make than a 400mm2 where half the dies are thrown away?
 
Is there any need to go huge SoC right off at the beginning? MS may have some serious yield issues if that comes out to be the case.
 
Status
Not open for further replies.
Back
Top