Predict: The Next Generation Console Tech

Status
Not open for further replies.
Or... more likely the SoC of 360. His August article, which he even references here, discusses that.

That would easily fit in with the 360 set top box rumour along with an announcement at CES (where a set top box for movies/cable services is a "CE Device" and would actually make sense at CES). 360S is already SOP with modest power consumption (90W).

IBM has their own edram tech, and they were already involved with the 360S. The last part would be moving the edram to a common node for the final SoC design/integration, which is possibly still 45nm (since the daughter die was still on 65nm with unchanged dimensions from prior revisions).

The xb360 CPU and GPU have already been integrated ... is there really that much more to be saved power-wise that such an "achievement" would be news worthy? ... or is there perhaps some slight chance that he's maybe got wind of the nextgen xb720 chip which just taped-out on perhaps 28nm...

Possible?
 
The xb360 CPU and GPU have already been integrated ... is there really that much more to be saved power-wise that such an "achievement" would be news worthy? ... or is there perhaps some slight chance that he's maybe got wind of the nextgen xb720 chip which just taped-out on perhaps 28nm...

Possible?

Considering Southern Islands taped out nearly 10 month ago, 28nm tape out for Xbox Next is definitely possible.
 
Last edited by a moderator:
The xb360 CPU and GPU have already been integrated ... is there really that much more to be saved power-wise that such an "achievement" would be news worthy?

Any talk of next gen rumours will generate site hits even if it isn't actually the next gen we want to hear about. Apart from speculated commentary, the "mole's" information just sounds like another revision of 360.

What the mole said: IBM/CPU/PPC + ATI/GPU + edram SoC. Oban. Tapeout now.

The rest is speculation and commentary.

I see no good reason to believe next gen would be an SoC unless they were planning on a really shitty leap.

Blame journalism if they hear "the next xbox chip" then report it as "the xbox next chip" - world of difference in meaning, but the latter gets people excited for the implied meaning of next gen. Not saying that's what happened, but there's not much to the article aside from the above list of facts.

-----------

Power-wise, there's probably not much worthwhile to brag about. However, I wouldn't underestimate the advantage of having a single chip over two. It's just simplification of the manufacturing. You have only one chip to test, only one manufacturing process. You don't have to wait for the second chip to arrive and be put on the package before testing it further.

There would be some changes on the mobo to reflect the new chip's characteristics.

Googling Oban, it seems the main result is some small town in Scotland. If it is on 28nm, a 360 evolution would be tiny indeed.
 
The xb360 CPU and GPU have already been integrated ... is there really that much more to be saved power-wise that such an "achievement" would be news worthy? ... or is there perhaps some slight chance that he's maybe got wind of the nextgen xb720 chip which just taped-out on perhaps 28nm...

Possible?

But the rumor does sound like a 360+. It could be a chip designed to run whatever 360 can run, but also enable more juice to run some apps specially design for it...

It would be really... Interesting, if true...
 
But the rumor does sound like a 360+. It could be a chip designed to run whatever 360 can run, but also enable more juice to run some apps specially design for it...

It would be really... Interesting, if true...

The 360 already has plenty of power, so realistically what kind of apps would need more than what it already has? A hardware video decoder would be useful though.
 
The xb360 CPU and GPU have already been integrated ... is there really that much more to be saved power-wise that such an "achievement" would be news worthy? ... or is there perhaps some slight chance that he's maybe got wind of the nextgen xb720 chip which just taped-out on perhaps 28nm...

Possible?

Certainly possible. But Al's theory is much more likely.

BTW there is still a lot of room left on the table for the 360 even at 40/45nm. Integrating everything onto one chip and optimizing the shit out of it could bring some really nice efficiency improvements.

If they're going to do it all on 28nm, the thing could be absolutely tiny and work in a power envelope not too far from the cable box that sits under your TV...
 
Any talk of next gen rumours will generate site hits even if it isn't actually the next gen we want to hear about. Apart from speculated commentary, the "mole's" information just sounds like another revision of 360.

What the mole said: IBM/CPU/PPC + ATI/GPU + edram SoC. Oban. Tapeout now.

The rest is speculation and commentary.

I see no good reason to believe next gen would be an SoC unless they were planning on a really shitty leap.

Blame journalism if they hear "the next xbox chip" then report it as "the xbox next chip" - world of difference in meaning, but the latter gets people excited for the implied meaning of next gen. Not saying that's what happened, but there's not much to the article aside from the above list of facts.

-----------

Power-wise, there's probably not much worthwhile to brag about. However, I wouldn't underestimate the advantage of having a single chip over two. It's just simplification of the manufacturing. You have only one chip to test, only one manufacturing process. You don't have to wait for the second chip to arrive and be put on the package before testing it further.

There would be some changes on the mobo to reflect the new chip's characteristics.

Googling Oban, it seems the main result is some small town in Scotland. If it is on 28nm, a 360 evolution would be tiny indeed.

Put another way:

What would MS benefit from redesigning the APU (GPU+CPU) again to integrate a few more parts on the same node as it currently sits in xb360s?

I'm thinking if they spent the time to do a full SOC, it will be on 28nm or 32nm. Otherwise it isn't worth the investment for such minimal gains.
 
I'm sorry but that example is little bit too much cherry picking. 560Ti 448-edition is a limited time special edition card, that is 560 in name only. Much proper name for it would be 570 LE or something like that. It uses the same 520mm2 die that is in GTX 580 and GTX 570, other 560 cards do not. It's more expensive than the regular 560s as well. It's just a handy way for nVidia to sell chips that didn't quite make it to be in the GTX 570. I'm sure no-one meant that model or Evga's GTX 560 2win card either, when they were talking about the GTX 560.

Imo the regular GTX 560 is an upper mid range card that has pretty good value proposition.
Or entry level high end...

Even looking at the normal GTX 560 the mainstream technical press refers to this product as Performance -- not mainstream. e.g. A recent example where Tom's was looking at a leak for the upcoming NV Kepler architecture and how it fits into the current framework and their use of Mainstream, Performance, High End, and Flagship (they didn't mention Entry level but that is the level below Mainstream; I used the current gen and how they slide across, some straddle, I think they overshoot at the top omitting the Kepler dual-GPU solution, aka Enthusiast, and the 590 itself was a later product, but that is picking at nits and gnats):

* (Entry Level: On-die/IGP GPUs; 520M 64bit; not mentioned in that article)
* Mainstream: 48 core, 128bit
* Performance: 336 core, 256bit
* High End: 480 core, 320bit
* Flag Ship: 512 core, 382bit
* (Enthusiast: SLI; inferred)

As you can see from their own graphic there is a HUGE drop from the Performance to the Mainstream and the 550TI GPU (192 cores, 192bit) disappears.

,6-A-317746-3.jpg


Looking at the market segments by die:

http://en.wikipedia.org/wiki/GeForce_500_Series

520: 79mm2, 29W, 155GFLOPs, 64/128bit, 48 core
530: 116mm2, 50W, 268GFLOPs, 128bit, 96 core
550: 238mm2, 116W, 691GFLOPs, 192bit, 192 core - Also 545
560: 360mm2, 170W, 1263GFLOPs, 256bit, 384 core
580: 520mm2, 244W, 1581GFLOPs, 384bit, 512 core - Also 570, 560 v.2 (320bit for both)
590: 520mm2 x 2, 365W, 2488GFLOPs, 2x 384bit, 1024 core

* Entry Level: On-die/IGP GPUs; 64bit bus models
* Mainstream: 520
* Performance: 530
* ? 550
* High End: 560
* Flag Ship: 580
* Enthusiast: SLI

As you can see there is even a distinct manufacturing/marketing tier (550) that doesn't fit into the chart which only pushes the 560 further away from mid range--which makes sense because logically it seems odd to have a 520mm2 chip as "High end" and a 79mm2 chip as "Low End/Mainstream" or even "Entry Level" and then have a 360mm2 chip "in the middle" especially when you consider the 360mm2 chip is clocked faster (882MHz vs 772MHz) and the functional units (384:64:32 vs 512:64:48) it clearly a jump well over the middle point.

And then there are the pricing brackets that get complicated due to slotting and binning (e.g. you may have a surplus of 580 chips with disabled units that become 560 models).

I don't think we are necessarily disagreeing, but what homer and I were taking issue with was pushing a 560 style card down into the mid range or as some were calling it mainstream market segment. It just isn't that sort of chip. I can see where hardcore PC gamers would see it as such, but that just isn't how these products are slotted.

That is all we were saying--I am not sure when roughly 80% of the performance (give or take) of your top single GPU solution becomes "mainstream" or "mid range" I think a fresh look of what the GPU landscape really looks like is in order. But I agree, a 560 is not in the same class as a 580, but homer was right to insist that there were more segments and categories in the market than others were allowing.
 
The 360 already has plenty of power, so realistically what kind of apps would need more than what it already has? A hardware video decoder would be useful though.

Good point. The only good use i could think of was to increase the resources allocated to background OSes tasks and for the in game/app guide, because i believe running windows would require more than whatever the current OS does. But it could really use more memory. (Yeah, that is not related to the chip design, so i guess i'll just shut up XD)
 
Assumed die costs:


Code:
$6,000	/Wafer
	
300mm	Wafer size
150mm^2	Die size
400	Dies per wafer
75.00%	Yield
300	Usable dies
[B]$20	Cost per CPU die[/B]
	
	
$6,000	/Wafer
	
300mm	Wafer size
300mm^2	Die size
200	Dies per wafer
50.00%	Yield
100	Usable dies
[B]$60	Cost per GPU die[/B]
 
What would MS benefit from redesigning the APU (GPU+CPU) again to integrate a few more parts on the same node as it currently sits in xb360s?

Aside from what I already wrote? 40/45nm is still going to be cheaper for mass production for a while longer.

I'm thinking if they spent the time to do a full SOC, it will be on 28nm or 32nm.
Sure. But what's not so clear at all is whether anyone has functioning edram on 28nm for this time frame. TSMC's and Renesas' last known plans only mentioned 40nm (G) (2010-2011) availability with 28nm eDRAM development currently under way.

eDRAM does not use the same manufacturing processes as regular CMOS logic, but you already knew that. eDram lags regular processes by quite a bit longer.

So what does that leave us? A common 40nm node that can already be produced or perhaps IBM's own 45nm experience with eDRAM (since Power 7). So... 40/45nm would be the logical choice. You can wish for jumping onto the next node ASAP, but doing that so soon for a new node makes less sense in the context of cost savings for an old chip design than it does for a whole new processor.

It may well be done on 28/32nm in the end, but again, there's no information about anyone having eDRAM ready for that node.
 
TSMC Production Capabilities:

20k wafers per month by the end of the year 2011.

Up to 100k wafers per month in 2012.

Let's assume just 60k wafers per month in 2012.

That puts the die capacity at 12million 300mm^2 dies per month.

Assuming 50% yield, that's 6million chips/month.

You think maybe MS might be able to manage 5% of that capacity for a couple months?

*Edited numbers*
 
Last edited by a moderator:
Aside from what I already wrote? 40/45nm is still going to be cheaper for mass production for a while longer.

Sure. But what's not so clear at all is whether anyone has functioning edram on 28nm for this time frame. TSMC's and Renesas' last known plans only mentioned 40nm (G) (2010-2011) availability with 28nm eDRAM development currently under way.

eDRAM does not use the same manufacturing processes as regular CMOS logic, but you already knew that. eDram lags regular processes by quite a bit longer.

So then it isn't feasible to assume it is a SOC with EDRAM as the cost savings wouldn't justify the design costs on the same 40/45nm.

Maybe I'm wrong and someone could shed some light on this.
 
Last edited by a moderator:
That puts the die capacity at 24million 300mm^2 dies per month.

For a 300mm^2 die, that's about 98 dice per 300mm wafer with your assumption of a 50% yield. So... <6 million chips per month.

So then it isn't feasible to assume it is a SOC with EDRAM as the cost savings wouldn't justify the design costs.

The rumour says it's an SoC with edram, and I'm working with what we know publicly about edram production. There are already cost savings by only having to manufacture one chip as opposed to two and then there's all the associated costs involved with QA testing.

If by some magic they have 28nm edram that is ready, then by all means. If not, then the rumours would simply imply that the design cost was worth merging the two even on 45nm.
 
Last edited by a moderator:
Status
Not open for further replies.
Back
Top