Predict: The Next Generation Console Tech

[Megadrive1988]

Well it's likely to be beyond that in feature but I think that Kietech was speaking about "raw" throughput (in GFLOPS).

I disagree., I believe the raw GFLOPS / TFLOPs in next-gen console GPUs will be higher than any 2010 PC GPU, no counting Nintendo's next console. In their case, something along the lines of a 2010 GPU would be more than enough, however the next Xbox and PS4 should be ahead in both features and raw floating point performance

 

[Rangers]

Even if TSMC or Global Foundries are "ready" for 22nm production, their capacity to supply MS/Sony/Nint will likely be limited.

So 28nm it is then?

That will leave the console makers with probably two sure shrinks left, 22nm and 15nm. Which should be enough to get them through next gen.

If they really do aim reasonably high next time, our minds are really going to be blown. Since incremental high end PC development basically stopped. I cant even imagine all these teraflops in next gen GPU's actually being put to use. IMO it's gonna be ridiculous.

 

[Squilliam]

Even if TSMC or Global Foundries are "ready" for 22nm production, their capacity to supply MS/Sony/Nint will likely be limited.

Isn't the big thing coming not the 22nm shrink but the adoption of 450mm wafers and how that changes cost / yield profiles for chips of various sizes and types.

 

[Blazkowicz]

Back to internet infrastructure, I think the future for rural America will be wireless based using Wireless phone services to send data to modem boxes or cards like what is already available. Rural residents will just have to live with the costs.

I'd hope fiber would be considered for rural areas, it does have afterall a huge range while keeping full bandwith, you may be able to hang it on poles, and the switching stations would be small scale and cheap.

Such kind of deployment would not be unheard of
http://en.wikipedia.org/wiki/Rural_Electrification_Administration

sad state of affairs when infrastructure projects that could be carried in the past are deemed impossible nowadays (while WW2 level military spendings are conveniently allowed)
here's a quick account of how things could get done ages ago (from the wiki article) :

RUS traces its roots to the Rural Electrification Administration (REA), one of the New Deal agencies created under President Franklin Delano Roosevelt. The REA was created on May 11, 1935 with the primary goal of promoting rural electrification.[1] In the 1930s, the U.S. lagged significantly behind Europe in providing electricity to rural areas due to the unwillingness of power companies to serve farmsteads.

Private electric utilities argued that the government had no right to compete with or regulate private enterprise, despite many of these utilities having refused to extend their lines to rural areas, claiming lack of profitability. Private power companies set rural rates four times as high as city rates.[2] Under the REA there was no direct government competition to private enterprise. Instead, REA made loans available to local electrification cooperatives, which operated lines and distributed electricity.

 

[Megadrive1988]

So 28nm it is then?

That will leave the console makers with probably two sure shrinks left, 22nm and 15nm. Which should be enough to get them through next gen.

If they really do aim reasonably high next time, our minds are really going to be blown. Since incremental high end PC development basically stopped. I cant even imagine all these teraflops in next gen GPU's actually being put to use. IMO it's gonna be ridiculous.

I agree, especially if we get an Nvidia Maxwell based GPU in PS4, even if it's lower-midrange by PC standards, it should be leaps and bounds beyond RSX. Nvidia doubles the transistors every GPU-generation.Lets only go back as far as NV47/G70/GTX 7800/RSX, that was 300M transistors. G80 was 686M (double) the GT200 was 1.4B (double) Fermi is 3B (double) Kepler should be 6B and Maxwell 12B. Lets say that PS4 GPU is half of a highend Maxwell, that's still 6B transistors for the GPU! That's 20 times the transistors of RSX (300M). It boggles the mind, the kind of flops performance PS4 could have with the GPU alone, nevermind the next-gen Cell.

Even if worst comes to the worst and PS4 only gets a midrange Kepler GPU with half the trannies as the 6B highend, that's still 3B transistors and still 10 times what RSX has. That could be integrated with a 32 SPE Cell on a single chip, that's say another 1 to 1.2 billion transistors for a total of just 4B+ transistors. That's my worst-case senerio, keeping the TDP to less than 100w.

Since PS4 should be coming in late 2013, that means that Kepler will be 2 years old because Kepler is coming in late 2011. Would PS4 be using a refresh / souped up Kepler, or a cut-down Maxwell? Maxwell is due in 2013, I think in time for PS4. Kepler will probably still be DX11 or DX11.1 or whatever but Maxwell should be DX12 and much more efficient per watt and per mm2. The only reason I could think of why PS4 won't use Maxwell is because PS3 did not use NV50/G80/GTX 8800, but instead used a 1+ year old NV47/G70 derivative. Hopefully though, Sony will go higher-end or at least more modern with the GPU this time. Like everyone is saying, they don't need such a bleeding-edge monster CPU. A 32 SPE Cell was meant to be due in late 2010 or early 2011. By 2013 on 28nm or 22nm, that won't be such a monster, but still a big increase from PS3 Cell configuration.

Give 32 SPE Cell + Kelper or Maxwell GPU (one combined superchip) at least 2 GB of Rambus next-gen Terabyte Bandwidth Inititive RAM and a huge pool of shared L3 cache or eDRAM, at least 24 MB. If there's going to be a seperate CPU and GPU, then 2 GB Rambus Terabyte stuff and 2 GB of the fastest flavor GDDX.

 

[AlNom]

Isn't the big thing coming not the 22nm shrink but the adoption of 450mm wafers and how that changes cost / yield profiles for chips of various sizes and types.

That's nice for cost per chip, but they still wouldn't want a big one... But in terms of supply capacity, you're also assuming that 450mm wafers are going to also be in quantity as well.

 

[Squilliam]

That's nice for cost per chip, but they still wouldn't want a big one... But in terms of supply capacity, you're also assuming that 450mm wafers are going to also be in quantity as well.

Well according to this Intel, Samsung and TSMC are looking to transition to 450nm wafers from 2012. If the other players follow accordingly soon after then perhaps it might swing the cost balance from fewer larger chips to many more smaller chips on account of the possibility that yields will follow the route for the last transition and fall for a couple of years and rise again back to previous levels.

By the time consoles are starting to really ramp up production in the next generation between 2013 and 2014 im pretty sure that 450nm wafers would be starting to become a lot more common. So it could be said that the next generation consoles for the most part may be fabbed on 450mm wafers.

 

[kyetech]

I disagree., I believe the raw GFLOPS / TFLOPs in next-gen console GPUs will be higher than any 2010 PC GPU, no counting Nintendo's next console. In their case, something along the lines of a 2010 GPU would be more than enough, however the next Xbox and PS4 should be ahead in both features and raw floating point performance

Your dreaming again think about this properly.

OK, lets assume they begin producion on 28nm process and move quickly over to say 20nm a year or two down the line.....

40nm to 28nm represents one full process shrink.... so....

Consider this carefully.....

projected cayman specs....

TDP: 220w underload.
Size: 400 / 450mm2
Transistor count: 3 billion. Approx.​

Console requirements. (correct me if im wrong)

TDP: 70 80 w underload.
Size: 150 / 200mm2
Transistor count: ??​

Here is what you are suggesting should be possible....

Increase raw and feature performance.
Reduce the TDP by a factor of three.
Reduce its physical size by more than half.​

In one process shrink?

Dream on....

 

[DuckThor Evil]

Your dreaming again think about this properly.

OK, lets assume they begin producion on 28nm process and move quickly over to say 20nm a year or two down the line.....

40nm to 28nm represents one full process shrink.... so....

Consider this carefully.....

projected cayman specs....

TDP: 220w underload.
Size: 400 / 450mm2
Transistor count: 3 billion. Approx.​

Console requirements. (correct me if im wrong)

TDP: 70 80 w underload.
Size: 150 / 200mm2
Transistor count: ??​

Here is what you are suggesting should be possible....

Increase raw and feature performance.
Reduce the TDP by a factor of three.
Reduce its physical size by more than half.​

In one process shrink?

Dream on....

I think that while your analysis is more inline with reality, it's still slightly pessimistic on every count and that adds up in the end. For launch I think the console chip could be 250mm2 and about 100w. Depending on the launch timeframe, it could launch smaller than 28nm, which already is only half of 40nm, if 20nm is just around the corner, it gives more breathing room. It won't be a shrink of Cayman either, but a newly designed chip with expected efficiency gains.

 

[3dilettante]

Well according to this Intel, Samsung and TSMC are looking to transition to 450nm wafers from 2012. If the other players follow accordingly soon after then perhaps it might swing the cost balance from fewer larger chips to many more smaller chips on account of the possibility that yields will follow the route for the last transition and fall for a couple of years and rise again back to previous levels.

That's what the chip-makers want to do, but has there been any significant statement on the part of the equipment makers?
The companies responsible for the equipment needed to make and process 450mm wafers took a bath on the 200->300mm transition. Obviously, chip manufacturers like Intel made plenty of money from the jump, but the people responsible for their equipment did not.

 

[eastmen]

Meh assuming a nov 2012 launch how many consoles would one need ? 4-6m ?

For MS they might launch nov 2012 in just NA with 3-4m consoles and wait till early the next year for a japan and EU launch. NA is thier big money market anyway.

I think if 22nm is ready then , they will be able to pull off a launch. Sony and nintendo might have a trickier time of it though.

 

[liolio]

Meh assuming a nov 2012 launch how many consoles would one need ? 4-6m ?

For MS they might launch nov 2012 in just NA with 3-4m consoles and wait till early the next year for a japan and EU launch. NA is thier big money market anyway.

I think if 22nm is ready then , they will be able to pull off a launch. Sony and nintendo might have a trickier time of it though.

I guess that greatly depends on launch price and positioning

Spoiler

Sorry but Pete really got me to laugh with that one.

 

[kyetech]

I think that while your analysis is more inline with reality, it's still slightly pessimistic on every count and that adds up in the end. For launch I think the console chip could be 250mm2 and about 100w. Depending on the launch timeframe, it could launch smaller than 28nm, which already is only half of 40nm, if 20nm is just around the corner, it gives more breathing room. It won't be a shrink of Cayman either, but a newly designed chip with expected efficiency gains.

There was a reason why I was pessimistic and it wasnt to highlight my point...

a) I dont think they will push the power envelope as much next time as they did with this generation of consoles. They are too big, too hot and too noisy = too expensive. They will release smaller / (slightly) cheaper consoles almost guaranteed.

b) Power leakage inside a transistor goes up the smaller it gets. You cant afford to target the same size chip as this gen as power leakage across the whole chip is becoming a real problem.

Thus.

They have to target a smaller chip with a lower power draw. (edit for clarification: in comparison to current gen console GPUs)

 

[homerdog]

There was a reason why I was pessimistic and it wasnt to highlight my point...

a) I dont think they will push the power envelope as much next time as they did with this generation of consoles. They are too big, too hot and too noisy = too expensive. They will release smaller / (slightly) cheaper consoles almost guaranteed.

TDP from one generation to the next almost never goes down except for anomalies from Nintendo. I wouldn't expect the next gen to be much smaller or cooler than the original Xbox 360.

 

[liolio]

TDP from one generation to the next almost never goes down except for anomalies from Nintendo. I wouldn't expect the next gen to be much smaller or cooler than the original Xbox 360.

At the same time systems have grown bigger and bigger,(sleeker, sexier design have hidden this a bit tho) I don't think they can go bigger than the ps360 as bigger would be your average medium size pc tower. The 360 has been there for almost 5years and the slim is still big. I've lost hope in seeing a real "ps2" this gen either from MS or Sony.

 

[kyetech]

as liolio points out consoles have had to get bigger to accommodate bigger power draw.

Consoles are big, expensive, hot and ugly (even the slim consoles are not particularly slim) and I think thats why they have struggled to sell like the consumer friendly PS1 / PS2 / wii counter parts.

In order for consoles to get a bit smaller, cheaper and consumer friendly , they need to draw less heat and power. Both MS and Sony will have learnt the lessons from this gen of consoles.

It seems to be a no brainer to me. Maybe im wrong time will tell.

 

[eastmen]

At the same time systems have grown bigger and bigger,(sleeker, sexier design have hidden this a bit tho) I don't think they can go bigger than the ps360 as bigger would be your average medium size pc tower. The 360 has been there for almost 5years and the slim is still big. I've lost hope in seeing a real "ps2" this gen either from MS or Sony.

My nes and 3do are almost as big as my xbox 360. even my saturn is sizable

 

[homerdog]

Xbox is bigger than Xbox 360. Really though, size should not be a huge concern for most users. And I reckon if you took a poll most users would not list power draw as one of their major concerns either.

 

[Alkohallick]

Xbox is bigger than Xbox 360. Really though, size should not be a huge concern for most users. And I reckon if you took a poll most users would not list power draw as one of their major concerns either.

I think most people still think "bigger is better"

 

[kyetech]

Xbox is bigger than Xbox 360. Really though, size should not be a huge concern for most users. And I reckon if you took a poll most users would not list power draw as one of their major concerns either.

Come on now. If size wasnt a concern then sales would not go up when slims are released. The evidence is there to see.

And size is a funtion of power draw.... you need more size to dissapate more heat and power.

As for other post... its all wdll and good pointing to saturn and xbox as big consoles. But how well did they sell? And that is the point.