Predict: Next gen console tech (9th iteration and 10th iteration edition) [2014 - 2017]

Status
Not open for further replies.
MrFox said:
http://wccftech.com/tsmc-promises-10nm-production-2016-7nm-2017/

So TSMC said 10nm will begin in 2016, 7nm in 2017.
Could 7nm become a viable possibility for 2019 then? Is two years enough for the process to ramp up and mature?
Click to expand...
TSMC began producing 28nm in 2011, consoles were released 2 years later (so yes 2 years seem to be enough for consoles)

So if they really produce 7nm in 2017 or are even delayed into 2018 then we could have next gen consoles at 7nm in 2020. Which would fit perfectly with my prediction. :LOL:
 
I read on a recent node you dont get any cost savings moving to new node. Just some 30% power use reduction for same performance.

Which I guess is still reason for new consoles to use it, but I wouldn't expect consoles to be at all aggressive on process going forward.
 
Rangers said:
I read on a recent node you dont get any cost savings moving to new node. Just some 30% power use reduction for same performance.

Which I guess is still reason for new consoles to use it, but I wouldn't expect consoles to be at all aggressive on process going forward.
Click to expand...

Eh? If the chips take up less physical die space on the wafer then how can you not get a cost saving? Will the yields on the new node really be that bad?
 
It's not just yield, but the number of steps required to produce the chip.

But you do get performance improvement in both power and performance. If you need those, you have no choice but to use a new node.
 
Prophecy2k said:
Eh? If the chips take up less physical die space on the wafer then how can you not get a cost saving?
Click to expand...
The cost to produce per unit die space is increasing with smaller nodes, where for a long time you could regard it as constant. If with a halving of die space for a given chip you double the cost per mm^2, you end with a smaller, more power efficient chip that costs the same as the larger one.
 
What about the move to 450mm wafers to lower the cost?
Any chance of this happening with 10nm and 7nm?
 
Wasn´t AMD moving away from TSMC and embracing Samsung at 14nm and beyond?
At least they were licensing the process for Glofo.

What about multi die stacking??
They are starting with HBM, but apparently they want to use it as well for its chips and apus in the Zen core Time-frame.

Imagine a multi module Zen, a gpu (or many) DRAM, all stacked without resorting to a huge die.
 
Xilinx have been doing this to make really large dies. They use a 65nm interposer, and the main chip is split in separate slices from 28nm. The problem is that it doesn't change the cost per mm2, nor the power consumption. These are FPGA that cost many thousands of dollar each. We're far from a $100 or so console SoC.
 
Michellstar said:
Imagine a multi module Zen, a gpu (or many) DRAM, all stacked without resorting to a huge die.
Click to expand...
It's still the same amount of silicon though. There's probably some savings in handling defects (throw out part of the chip instead of the whole thing) but the basic costs are the same.
 
Shifty Geezer said:
It's still the same amount of silicon though. There's probably some savings in handling defects (throw out part of the chip instead of the whole thing) but the basic costs are the same.
Click to expand...

Yes i know, but smaller dies with future nodes are easier to handle, so they say :)
And could give them a more modular design, without resorting to larger socs, more complex masks, etc etc

I came across this paper yesterday, because i remembered that amd was behind stacking for zen cpus and apus

http://www.microarch.org/micro46/files/keynote1.pdf
 
I guess someone has already said this, but 18 pages? not a chance I'm reading all them to see.

I think the next generation both consoles (if indeed MS and Sony proceed in the market) will be virtually identical. Same GPU (+- 1 CU), same CPU (+- 100mhz).

I'm gonna suggest that the GPU will be on par with the next AMD gpu's "main stream" version (so the 380x or whatever they call it)

CPU, whatever AMD has in the pipeline for the mobile market again, potentially running at 2 ghz.

100% BC for both if MS include a shape 'emulator' of sorts.

I guess both will use interposers and stacking so bandwidth will be extremely high I guess.
 
Last edited:
MrFox said:
Xilinx have been doing this to make really large dies. They use a 65nm interposer, and the main chip is split in separate slices from 28nm. The problem is that it doesn't change the cost per mm2, nor the power consumption. These are FPGA that cost many thousands of dollar each. We're far from a $100 or so console SoC.
Click to expand...

Well not now, but it could be in the next-gen time-frame (2019/2020)

I read somewhere that (apart from HMB which is happening next month) AMD plans to use an interposer for the top of the line ZEN, opteron range, 2 "4 cores" module, this will trickle-down to mainstream cpus and apus.

Will see (if AMD is around lol)
 
DJ12 said:
I guess someone has already said this, but 18 pages? not a chance I'm reading all them to see.

I think the next generation both consoles (if indeed MS and Sony proceed in the market) will be virtually identical. Same GPU (+- 1 CU), same CPU (+- 100mhz).
Click to expand...

I somehow think that there is a good chance that Nvidia will try to jump again in the console space. They will make a deal with MS, better [ok, more fair :D] deal than the one they made for first Xbox.
 
DieH@rd said:
I somehow think that there is a good chance that Nvidia will try to jump again in the console space. They will make a deal with MS, better [ok, more fair :D] deal than the one they made for first Xbox.
Click to expand...

DJ12 said:
Which will surely mean Intel will supply the cpu as and will want all or nothing. Seems unlikely.
Click to expand...

I think there is better chance for that next time around. Console manufacterers will probaly target SOC designs again. Last time , amd jaguar was only cost effective/realistic cpu solution for apu ( for production at TSMC). Denver was not avaialable, 64bits arm was not avialable etc. Next time there will be much more options with cpu. From gpu side, when negotiations were underway, I bet Jen Sen only put power hungry fermi on table:d (remember rsx vs g80, and xbox1 stuff... ).

Thats said, a while ago nvidia seemed like they are more open with IP licensing. "everything is possible license " Have anyone heard if anything came to fruit from that?. If i were Jen Sen, i would give margins for few quarters and give them fair chip for domination of console space, which is de facto motors for high end , mass market , real time graphic develpoment. Down the road ; Easier for driver team,scalability -visibility, gameworks proliferation , nvidia tentacles over everything and much more:d. The could achieve unreachable domination from that:D.

Anyway , whatever they will choose must be cost effective, and must resonate with mass market immediately ( 400$ or so). Strong first year or... they will see swarm of those "next gen " 100mln$ console selling, titles from 3rd party... 4-5 years later... Time/work/budget requiments for software have killed sense for developng "exotic " and expensive hardware and software is next. I still feel, the only option for developing strong sustainable console ecosytem are faster forward/backward compatible hardware cycles, as per this thread. "smaller jumps , faster iterations is way to go " as the say lately... and its much easier with todays gamers obsessed mainly with pixels and fps .
 
Status
Not open for further replies.

Similar threads

A
Full transcript of John Carmack's QuakeCon 2005 Keynote
3 4 5
Replies
98
Views
32K
nemisis
N
Back
Top