By: anon (a.delete@this.b.c), August 6, 2014 2:48 pm
Room: Moderated Discussions
Aaron Spink (aaronspink.delete@this.notearthlink.net) on August 5, 2014 2:32 am wrote:
> Exophase (exophase.delete@this.gmail.com) on August 4, 2014 8:45 am wrote:
> > The quoted article from 2011 cites the top of the line i7 used at the time cost $378..
> > based directly on the Ark price. Then says that $300+ is the typical price for even the
> > lowest end MBPs and that they'd save hundreds moving to something else. The point is
> > that Apple probably doesn't play anywhere close to the prices Intel lists on Ark.
> >
> > The big jump in list price for high end MBP CPUs between then and now is mostly due to
> > high end Iris Pro with the big L4 cache. Maybe Apple really is paying a big price, but
> > it's not going to be so easy to replace with their own solution for very cheap either.
> >
> I find it funny how all these people think they will save hundreds. Certainly Intel get a favorable margin
> on their parts, but a decent portion of that margin is created by shear volume. There isn't another CPU design
> out there that is even in the same ballpark volume wise. Intel is able to amortize R&D over roughly 300+
> million chips per design. There really aren't many chips out there that do even 100+ million.
>
> And pretty much all the iOS volume does nothing for amortizing a design aimed
> at laptop and desktops. Esp since apples volume in the laptop/desktop market
> is a small fraction of the market size that Intel is amortizing over.
>
True.
> In order to take the current or future Ax design and take it up to desktop levels, they would have to completely
> re-pipeline the processor, and that would likely require a pretty heavy re-design of the micro-architecture.
A8 is rumored to run at 2GHz. A12, if not A10 will probably have a CPU core running @2.5/3GHz or close by. iOS volume will help the CPU core.
> Then they would likely have to design a brand new uncore from scratch targeted at a significantly higher
> performance than the iOS device uncore and include significant PCIe connectivity.
>
> For integrated graphics, they would likely need a completely new boutique design based
> off their current graphics design. They would then also need to create new uncore
> infrastructure to an on package memory to get the same levels of bandwidth.
GPU's are scalable designs anyway. IMG has demonstrated probably the world's most scalable GPU architecture with their hybrid sort-first and sort-middle tiled architecture.
Given all the on-package mem initiatives, (HBM/HMC/WIO) ongoing, Apple will have a high-perf uncore and on package memory soon any way. iOS volume will help here.
>
> They then would need a lot more people developing software infrastructure for all of the above. etc.
They probably have more people on iOS than Mac already.
>
> Also the people just saying throw more cores at it are way off in left field. There is a reason
> that apple has gone against the general trend in smartphones/tablets of just adding more cores,
> and that's because it doesn't work. Most applications just can't make use of more cores.
>
> And if performance wasn't an issue, like some people say, then Apple wouldn't
> be pushing on Intel to develop and deliver solutions like IRIS PRO, nor would
> Apple generally use the top of the line parts in their machines.
>
Absolutely.
> And as far as Apples cost, apple is unlikely to be paying anything resembling list price, same for all
> the other OEMs with significant volume. When you actually look at all the costs, it is unlikely that Apple
> would move to ARM based laptops or desktops for any cost based reasons because the cost savings are most
> likely not there to begin with and if they are, they are marginal at best with huge massive risk.
>
> And then there are those who argue that they would move do to having better visibility for planing reasons,
> which is pretty much BS as well. Intel actually has a much better track record of hitting their targets
> than most and the reality is that having things in house is no guarantee of schedule.
>
> I won't even get into the whole bin distribution issues that would result in overall
> lower performance. Suffice to say, Apple largely benefits from being able to get good
> pricing on better binned chips than they would likely be able to self produce.
>
It doesn't make sense in 2014. But in 2016? 2018? There, it looks to be a toss up to me. If it's a toss up, I think Apple will bring mac chips inhouse.
> Exophase (exophase.delete@this.gmail.com) on August 4, 2014 8:45 am wrote:
> > The quoted article from 2011 cites the top of the line i7 used at the time cost $378..
> > based directly on the Ark price. Then says that $300+ is the typical price for even the
> > lowest end MBPs and that they'd save hundreds moving to something else. The point is
> > that Apple probably doesn't play anywhere close to the prices Intel lists on Ark.
> >
> > The big jump in list price for high end MBP CPUs between then and now is mostly due to
> > high end Iris Pro with the big L4 cache. Maybe Apple really is paying a big price, but
> > it's not going to be so easy to replace with their own solution for very cheap either.
> >
> I find it funny how all these people think they will save hundreds. Certainly Intel get a favorable margin
> on their parts, but a decent portion of that margin is created by shear volume. There isn't another CPU design
> out there that is even in the same ballpark volume wise. Intel is able to amortize R&D over roughly 300+
> million chips per design. There really aren't many chips out there that do even 100+ million.
>
> And pretty much all the iOS volume does nothing for amortizing a design aimed
> at laptop and desktops. Esp since apples volume in the laptop/desktop market
> is a small fraction of the market size that Intel is amortizing over.
>
True.
> In order to take the current or future Ax design and take it up to desktop levels, they would have to completely
> re-pipeline the processor, and that would likely require a pretty heavy re-design of the micro-architecture.
A8 is rumored to run at 2GHz. A12, if not A10 will probably have a CPU core running @2.5/3GHz or close by. iOS volume will help the CPU core.
> Then they would likely have to design a brand new uncore from scratch targeted at a significantly higher
> performance than the iOS device uncore and include significant PCIe connectivity.
>
> For integrated graphics, they would likely need a completely new boutique design based
> off their current graphics design. They would then also need to create new uncore
> infrastructure to an on package memory to get the same levels of bandwidth.
GPU's are scalable designs anyway. IMG has demonstrated probably the world's most scalable GPU architecture with their hybrid sort-first and sort-middle tiled architecture.
Given all the on-package mem initiatives, (HBM/HMC/WIO) ongoing, Apple will have a high-perf uncore and on package memory soon any way. iOS volume will help here.
>
> They then would need a lot more people developing software infrastructure for all of the above. etc.
They probably have more people on iOS than Mac already.
>
> Also the people just saying throw more cores at it are way off in left field. There is a reason
> that apple has gone against the general trend in smartphones/tablets of just adding more cores,
> and that's because it doesn't work. Most applications just can't make use of more cores.
>
> And if performance wasn't an issue, like some people say, then Apple wouldn't
> be pushing on Intel to develop and deliver solutions like IRIS PRO, nor would
> Apple generally use the top of the line parts in their machines.
>
Absolutely.
> And as far as Apples cost, apple is unlikely to be paying anything resembling list price, same for all
> the other OEMs with significant volume. When you actually look at all the costs, it is unlikely that Apple
> would move to ARM based laptops or desktops for any cost based reasons because the cost savings are most
> likely not there to begin with and if they are, they are marginal at best with huge massive risk.
>
> And then there are those who argue that they would move do to having better visibility for planing reasons,
> which is pretty much BS as well. Intel actually has a much better track record of hitting their targets
> than most and the reality is that having things in house is no guarantee of schedule.
>
> I won't even get into the whole bin distribution issues that would result in overall
> lower performance. Suffice to say, Apple largely benefits from being able to get good
> pricing on better binned chips than they would likely be able to self produce.
>
It doesn't make sense in 2014. But in 2016? 2018? There, it looks to be a toss up to me. If it's a toss up, I think Apple will bring mac chips inhouse.
