Predict: The Next Generation Console Tech

[liolio]

I'll attempt to predict the next XGPU assuming that die space dedicated to the GPU is close to Xenos, and that it is built with 22 fab process, and must draw less than 150 Watts at peak.

Looking back at AMD gpus that had die space between 250 mm2 and 260 mm2, the candidates are:

10/1/2005 X1800 254mm2 90nm 321 million transistors 500mhz 256mb GDDR3 ~140 Watts 83 GFLOPS
11/16/2005 Xenos ~260mm2 90nm 337 million transistors 500mhz 512mb GDDR3 125? Watts 240 GFLOPS
6/25/2008 4850 256mm2 55nm 956 million transistors 575mhz 512mb GDDR5 110 Watts 1000 GFLOPS
10/22/2010 6850 256mm2 40nm 1700 million transistors 775mhz 1024mb GDDR5 127 Watts 1488 GFLOPS

I would put my bet on HD 6850 lvl of perfs a bit lower actually (lower clock and possibly a bit less SIMD array).

A rough extrapolation of the next XGPU based on extremely dubious mathematics, namely the squaring of ratios between fab processes, the next XGPU looks like:

Fall 2013 Next XGPU ~ 260mm2 22nm ~5000 million transistors <1000mhz <=4096mb GDDR5/XDR2 <150Watts <~4000 GFLOPS

It's never going to happen, 22nm won't be available to anyone but Intel by fall 2013.

I am by no means an expert at this stuff, so any help with this crystal balling will be appreciated.

I would not discard Charlie's claims about MS using a SoC and expecting the first wafer to come back Q1 2012. So most likely a chip using the same process as llano, 32nm.
Still iffy, I hope that Arstechnica's mole will shine in soon

 

[jonabbey]

I am by no means an expert at this stuff, so any help with this crystal balling will be appreciated.

First, don't call it "crystal balling".

 

[function]

I think Global Foundries' 28nm is supposed to come online next year isn't it? That may be an option so long as they can get yields up. Things are supposed to be pretty rough for their 32nm process at the moment according to Birght Side of SemiZilla (one of them, can't remember).

Edit: would this "28nm HPP" process be suitable for fast console SOC?

http://semiaccurate.com/2011/08/30/global-foundries-and-samsung-split-28nm-processes/

Edit 2: not so sure based on this:

http://www.globalfoundries.com/technology/28nm.aspx

 

[torbor]

Why future consoles shouldnt use more watts then the previous ones?
They also could build bigger boxer for the next gen.
Just remember:

 

[Heinrich4]

Only for information:

Cell 45nm consumes less than 20 watts
http://forum.beyond3d.com/showthread.php?t=47022


A system with 7900GT at 90nm and less transistor than RSX(278million versus 300+million RSX) reachs at least =~180 watts/TDP:

http://techreport.com/articles.x/10039/9


Perhaps a system in first quarter 2012 with a Radeon HD 7870/28nm/1536 SIMD/120watts(for SoC? Enough for a specs for system coming in 2013) customised with some tweaks for console and cpu 4/8 core PPC A2 2+GHz are not wattage expensive as it was in an RSX/cell or a Xenon/Xenos under 90nm.

 

[msia2k75]

Only for information:

Cell 45nm consumes less than 20 watts
http://forum.beyond3d.com/showthread.php?t=47022


A system with 7900GT at 90nm and less transistor than RSX(278million versus 300+million RSX) reachs at least =~180 watts/TDP:

http://techreport.com/articles.x/10039/9


Perhaps a system in first quarter 2012 with a Radeon HD 7870/28nm/1536 SIMD/120watts(enough for a specs for system coming in 2013) and cpu 4/8 core PPC A2 2+GHz is not wattage expensive as it was in an RSX/cell or a Xenon/Xenos under 90nm.

About CELL at 20w you should read the rest of the thread.

 

[Proelite]

Is 28nm the best that console manufacturers can use it 2013? Maybe both Sony and MS should wait even longer because these fab delays have been affecting AMD and Nvidia.

 

[tunafish]

Is 28nm the best that console manufacturers can use it 2013?

Yes. And if they want eDRAM on the same die as either the GPU or CPU, the best process available will probably be 32nm. T-RAM on GF 28nm is a wildcard -- but if anyone in either company decides to bet their futures in GlobalFoundries capability to deliver, they deserve to be shot.

 

[liolio]

Is 28nm the best that console manufacturers can use it 2013? Maybe both Sony and MS should wait even longer because these fab delays have been affecting AMD and Nvidia.

Hell no, no more delay! it will be at the very least 7 years since we saw something new. 32nm should be enough really even within a Soc it should be easy to hit +1TFLOPS from the GPU, 4/5 times the raw throughput of Xenos, texturing power should go through a greater jump and all this while being more efficient. If we consider RSX ROPs and the bandwidth to feed the there will also be a beefy jump, in the mean thime we will see only only a x2 in resolution.
It's an awful generation imho, usual consoles costumers are stuck to outdated hardware thanks to to manufacturers mistakes, enough of that 10 year life cycle it is bullshit. Damned the whole internet is just melting as the first tiny rumors about something new, clearly it's time already and that should show for the first willing to launch something new (worse it and WiiU may not qualify).

 

[Tahir2]

Is 28nm the best that console manufacturers can use it 2013? Maybe both Sony and MS should wait even longer because these fab delays have been affecting AMD and Nvidia.

I would have to say yes but it is not clear cut. 32nm was scheduled at TSMC for 2009 until it got cancelled sometime last year. 28nm is now scheduled for the end of 2011 according to the noises NVIDIA and AMD are making with regards to their graphics cards.

40nm has been the leading edge process at TSMC for AMD GPU's which began production at around the end of 2009. It looks like the new process nodes are taking longer to perfect and ramp.

Global Foundries is a newcomer in the field (the AMD fabs were geared just for AMD products, it is now an open fab in competition with TSMC) and it is possible that they may be ahead of TSMC by 2013. Currently they are ahead with regards to 32nm with shipping products as TSMC gave up on that process node.

Perhaps by 2013 22nm may be viable ... but delays are likely.

 

[Tahir2]

Hell no, no more delay! it will be at the very least 7 years since we saw something new. 32nm should be enough really even within a Soc it should be easy to hit +1TFLOPS from the GPU, 4/5 times the raw throughput of Xenos, texturing power should go through a greater jump and all this while being more efficient. If we consider RSX ROPs and the bandwidth to feed the there will also be a beefy jump, in the mean thime we will see only only a x2 in resolution.
It's an awful generation imho, usual consoles costumers are stuck to outdated hardware thanks to to manufacturers mistakes, enough of that 10 year life cycle is bullshit. Damned the whole internet is just melting as the first tiny rumors about something new, clearly it's time already and that should show for the first willing to launch something new (worse it and WiiU may not qualify).

I was going to say something similar but looking at the data more closely, it is possible 22nm is ready for 2013. It may also be a reason for AMD and GloFlo getting in on the action for the consoles. Their product roadmap is very aggressive.

A couple of references:

http://www.xbitlabs.com/news/other/..._Production_Using_20nm_22nm_in_Late_2012.html

Test chip shuttles for customers will begin running in Fab 1 in the second half of 2011. The risk production of chips on the 20nm and 22nm nodes is projected to begin in the second half of 2012 and volume production will commence in 2013.

http://www.anandtech.com/show/2814

 

[Blazkowicz]

in general processes are slower to come out and mature than before, power reductions are lower and/or more difficult and investment costs rise a lot as well.

that must be why there's a lot of sentiment for "low" specs consoles.

 

[bomlat]

Is there enough money any more in this business,to justify the razor and blade business model?
I mean 10 years ago it was easy to cheat the top management of the MS to supply a money eating project like this,but now it is seems like the market saturated,the overall sales of the three systems are roughly on the same level like in 2006 (if we consider that at least 20 million xb/ps died by rrod/ylod)

So,to justify a big upfront cost the market gain has to be enormous,and the cost cutting very step.
It hasn't been like this in the past few years.

I mean,if they has to be on the break even from the day one,then the best that can be expected is an ps360 in wii style -more mem,more core and higher frequency,but enough only to double/triple the performance.
And they have to consider to give additional benefit to the customers in wii style again,beyond/instead the raw performance.

 

[tunafish]

GloFlo ... Their product roadmap is very aggressive.

GloFo roadmap has been very aggressive from since it was spun out. The way they achieve this is simply by failing to deliver at the times they originally promise to.

There is no way in hell I'd take GloFo's word on that their 22nm is ready by 2013.

 

[liolio]

in general processes are slower to come out and mature than before, power reductions are lower and/or more difficult and investment costs rise a lot as well.

that must be why there's a lot of sentiment for "low" specs consoles.

Well that's part of it there is also the "it has to compete with high end PC" mentality that hurst perceptions, PC power consumption have gone crazy high during this console generation time line.

 

[Heinrich4]

About CELL at 20w you should read the rest of the thread.

Thanx for information,but still only a 35/40 watt cpu with 8 cores 3.2GHz the difference does that usually have a system 150/200 watts total is minimal,35/40 watts no more than 20% of total TDP,much less than the expected 50% or more usually estimated.

About the manufacturing process, it is probably something like 28nm AMD is now by setting (Radeon 7870 and 7900*) it to this level if GPU separately, but if SoC is perhaps if like Trinity APU 32m coming in first quarter 2012 and come with much better yields in 2013 ( 22nm maybe coming for "V2" or year later launch console).

 

[function]

Nintendo are launching next year on 45 nm.

AMD banked on having two new chips out (one of them a completely new architecture) on a brand new process and have landed in the shit. Again.

 

[Heinrich4]

Nintendo are launching next year on 45 nm.

AMD banked on having two new chips out (one of them a completely new architecture) on a brand new process and have landed in the shit. Again.

Correct me if I'm wrong, but it seems that IBM would be participating in the process of manufacturing a SoC likely for a other console( new xbox?),if same happen in WiiU that would explain why they are still on 45nm.

It seems the manufacturing process in CPU that dictates the terms in a SoC or and APU.

 

[Rangers]

I mean,if they has to be on the break even from the day one,then the best that can be expected is an ps360 in wii style -more mem,more core and higher frequency,but enough only to double/triple the performance.
And they have to consider to give additional benefit to the customers in wii style again,beyond/instead the raw performance.

Break even from day one at 299/399 they can do a whole lot better than 2-3X.

 

[Kaotik]

Nintendo are launching next year on 45 nm.

AMD banked on having two new chips out (one of them a completely new architecture) on a brand new process and have landed in the shit. Again.

AMD landed in the shit where?