I think BeyondImgTec has a front page, where as B3D doesn't...
Beyond3D => InsideImgTec!
Predict: The Next Generation Console Tech
- Thread starter Acert93
- Start date
- Status
Beyond3D => InsideImgTec!
french toast
Veteran
Yea alright guys...bloody show offs!!
With the x86 Durango rumor and some speculation about Atom processors I did some mental scenario day dreaming. How could we get Intel technology into a console?
The important question is what would get Intel to make a bid? Some reasons, some better than others:
* Intel sees emerging markets in tablets and mobile, of which Atom plays a key role in terms of Intel's offering. Getting a many-core Atom architecture into a console sets the table for software to go "multiplatform." You could have your 2.xGHz 8 core Atom "Xbox 3" game sku aimed at 1080p and then the 4 core 1.4GHz tablet 720p mobile/tablet game sku. In the end it helps Intel's case the more gaming is tied to x86. If x86 rules high end console gaming that helps PC gaming which is one of the avenues Intel's high margin chips are sold.
* Intel could use this as a defensive move against Jaguar (in the PS4?) and prevent MS from going with IBM or ARM.
* Intel would not have to swallow deep margin cuts as they would with their Core/IVB/SB/etc architecture. Atom processors are already lower margin iirc so the trade off may be 60m units in 5 years of guaranteed volume.
* MS would be aiding the new model of Atom: Out of Order with HT, with AVX, more scalable core architecture, on 22nm with Tri-Gate. While Intel doesn't need MS's money for this having a ready client helping invest in the development of the "next gen" Atom doesn't hurt--especially if that means MS in turn is going to evangelize hundreds of game developers to the platform.
Now there are a lot of benefits to MS:
* Intel has a large library of very robust x86 tools, compilers, etc.
* MS could start working with Atom cores right now.
* Intel offers the ONLY reliable and predictable process plan. MS would be one process node ahead and would likely be the only company with 3 process node moves in a 7 year window as TSMC is going very slow these days.
* Intel's foundry tech is advanced. Look, what MS would "save" in having smaller chips is going to be eaten by Intel in terms of price premium. So if MS's chip on 22nm is half the size of Sony's on TSMC's 28nm they will cost about the same. Where MS comes ahead is Intel already has volume 22nm flowing, it is a more mature process, and I will take a wild stab here: MS could get higher clocks and better performance with a LOWER TDP per core with a next gen Atom with AVX on Intels 22nm with Tri-gate than AMD will get with 28nm/32nm Jaguar with AVX on GF or TSMC.
* Atom cores are small; Atom cores with AVX strike that balance of "small cores with large SIMD". Not a lot would be "wasted" on efficiency advantages the other Intel cores have; what they would do is make up for it in (1) having more real cores and (2) having more SIMD units. This in theory gives the platform long legs for developers to squeeze out more performance.
While I am sure Atom processors make people cringe OOOe Atom cores with HT and AVX are probably faster, per core, than the PPEs in Xenon/Cell. When you are talking about 8 real cores (16 threads) with SIMD there some more room than, say, 4 Jaguar cores.
If TDP, followed by footprint, are the walls consoles will hit first Atom makes sense from this angle. Each Atom core has a very low TDP, so what you lose in robustness you gain in thermals. It won't be the world beater that a next-gen Cell would have been or an 8 core SandyBridge -- but the consoles don't have the fiscal or thermal budgets for such advanced chips. An 8 core x86 OOOe processor with HyperThreading and AVX that runs well under 50W on 22nm at least has to sound appealing.
That said... I don't think Intel would have the incentive to shift fab space away from their high margin products for a low margin console part. I just don't see it.
But the prospect of seeing an ARM processor or a meandering 4 core evolution of BobCat would have me drooling for a next gen 8 core Atom with AVX -- kind of scary how far we have fallen from the OP!
The important question is what would get Intel to make a bid? Some reasons, some better than others:
* Intel sees emerging markets in tablets and mobile, of which Atom plays a key role in terms of Intel's offering. Getting a many-core Atom architecture into a console sets the table for software to go "multiplatform." You could have your 2.xGHz 8 core Atom "Xbox 3" game sku aimed at 1080p and then the 4 core 1.4GHz tablet 720p mobile/tablet game sku. In the end it helps Intel's case the more gaming is tied to x86. If x86 rules high end console gaming that helps PC gaming which is one of the avenues Intel's high margin chips are sold.
* Intel could use this as a defensive move against Jaguar (in the PS4?) and prevent MS from going with IBM or ARM.
* Intel would not have to swallow deep margin cuts as they would with their Core/IVB/SB/etc architecture. Atom processors are already lower margin iirc so the trade off may be 60m units in 5 years of guaranteed volume.
* MS would be aiding the new model of Atom: Out of Order with HT, with AVX, more scalable core architecture, on 22nm with Tri-Gate. While Intel doesn't need MS's money for this having a ready client helping invest in the development of the "next gen" Atom doesn't hurt--especially if that means MS in turn is going to evangelize hundreds of game developers to the platform.
Now there are a lot of benefits to MS:
* Intel has a large library of very robust x86 tools, compilers, etc.
* MS could start working with Atom cores right now.
* Intel offers the ONLY reliable and predictable process plan. MS would be one process node ahead and would likely be the only company with 3 process node moves in a 7 year window as TSMC is going very slow these days.
* Intel's foundry tech is advanced. Look, what MS would "save" in having smaller chips is going to be eaten by Intel in terms of price premium. So if MS's chip on 22nm is half the size of Sony's on TSMC's 28nm they will cost about the same. Where MS comes ahead is Intel already has volume 22nm flowing, it is a more mature process, and I will take a wild stab here: MS could get higher clocks and better performance with a LOWER TDP per core with a next gen Atom with AVX on Intels 22nm with Tri-gate than AMD will get with 28nm/32nm Jaguar with AVX on GF or TSMC.
* Atom cores are small; Atom cores with AVX strike that balance of "small cores with large SIMD". Not a lot would be "wasted" on efficiency advantages the other Intel cores have; what they would do is make up for it in (1) having more real cores and (2) having more SIMD units. This in theory gives the platform long legs for developers to squeeze out more performance.
While I am sure Atom processors make people cringe OOOe Atom cores with HT and AVX are probably faster, per core, than the PPEs in Xenon/Cell. When you are talking about 8 real cores (16 threads) with SIMD there some more room than, say, 4 Jaguar cores.
If TDP, followed by footprint, are the walls consoles will hit first Atom makes sense from this angle. Each Atom core has a very low TDP, so what you lose in robustness you gain in thermals. It won't be the world beater that a next-gen Cell would have been or an 8 core SandyBridge -- but the consoles don't have the fiscal or thermal budgets for such advanced chips. An 8 core x86 OOOe processor with HyperThreading and AVX that runs well under 50W on 22nm at least has to sound appealing.
That said... I don't think Intel would have the incentive to shift fab space away from their high margin products for a low margin console part. I just don't see it.
But the prospect of seeing an ARM processor or a meandering 4 core evolution of BobCat would have me drooling for a next gen 8 core Atom with AVX -- kind of scary how far we have fallen from the OP!
Some goods points there and I have to agree that if MS could strike the right deal there are a lot of plus points for them.
I just don't see the same incentives for Intel to use up there high profit latest process capacity. Or even the design time for a custom chip with no place outside of consoles.
I just don't see the same incentives for Intel to use up there high profit latest process capacity. Or even the design time for a custom chip with no place outside of consoles.
tongue_of_colicab
Veteran
As Acert said, maybe there isn't so much incentive for intel to build a console cpu just for the sake of being in the console business but there would be added advantages that might make it worth it like having a lot of studios maken x86 based gamse which could easily be ported to W8 later on where intel wants to sell it atom cpu's for phones/tablets and it iWhatever cpu's for desktops. So a console cpu might be a low profit product for intel but it would give them a lot of ''free'' content that might make people buy other platforms which have intel hardware onboard.
Actually isn't the reason why IBM licenses cheaply is because it means a significant number of developers are familiar with PPC architectures and it helps amortize their high end server chips with volume? Given the above wouldn't IBM 'plan' to have a good new architecture ready in time for a console generation? Perhaps even something which hasn't been released yet.
I wouldn't mind seeing Intel in the next xbox, but I very much doubt it will happen. Not that there aren't any advantages for intel to try and get into to that space, as Acert has already pointed out, but I doubt they will go so far as to give IP right to MS, similar to what they have now and I am sure they would like to have in the future. Wasn't that one of the major problems with the first xbox? But then again, the times they are a-changing so anything might happen...
What about the idea of Intel selling its fab capacity?
22nm is quite expensive but what if a console manufacturer gave them a cool sum of money to use their fabs? Consoles aren't exactly direct competitors to their products.
22nm is quite expensive but what if a console manufacturer gave them a cool sum of money to use their fabs? Consoles aren't exactly direct competitors to their products.
You think a low (the slowest?) salvage model of a GK104 is attractive for a console?
I think a custom GPU slightly below a Pitcairn would be some kind of a sweet spot in my opinion. Much smaller than a GK104 and lower power. You could even integrate it to a SoC, if you want (and I would want that, at least for a future version after the next shrink).
I think a custom GPU slightly below a Pitcairn would be some kind of a sweet spot in my opinion. Much smaller than a GK104 and lower power. You could even integrate it to a SoC, if you want (and I would want that, at least for a future version after the next shrink).
I'm not sure. It is 292mm, since it's gk104 based. That's a pretty big jump from 212mm of pitcairn for probably not much more performance.
I've been wondering the same though.
As some forumers have already quoted my questions I don´t have to do it. You see he has answered everything. But I´d have prefered him to answer me if the 8 cores are low speed ones. I´ll wait a few days until the next question.
But he has clearly answer me the Next Box will be the msot powerful system.
But he has clearly answer me the Next Box will be the msot powerful system.
Not really. He said it'll be the "best nextgen on the market". Open to interpretation. Besides, I wouldnt expect him to know much about Orbis, in fact he already pretty much says he doesnt. When somebody tweeted those old extremely fake PS4 specs to him, all he could muster was a weak "they look illegitimate". If he knew PS4 specs he could say with certainty.
I hope more people continue to pester him with questions, he's been pretty responsive.
But he has already gave at least his opinion. It would be nice to see another insider with Orbis dev kit.
- Status
Similar threads
- Locked
