By: Daniel Bizo (fejenagy.delete@this.gmail.com), February 5, 2013 6:31 am
Room: Moderated Discussions
Richard Cownie (tich.delete@this.pobox.com) on February 4, 2013 5:01 pm wrote:
> Daniel Bizo (fejenagy.delete@this.gmail.com) on February 4, 2013 3:45 pm wrote:
>
> I would agree the numbers are questionable, and if anyone cares to present evidence
> for different numbers, I would welcome better information.
>
Fair point, I don't care enough to dig up examples. I've seen Apple iPhone 5 BOM analysises that varied greatly for the SoC.
>
> I think that trend may reverse, because a) with A15/Krait-class cores in future,
> together with better-optimized Android software, it seems that dual-core phones
> will be sufficient for the vast majority of users; b) now that screen resolution
> has reached 1080p, the pixel counts aren't likely to increase much further so the
> GPU's don't need to get much faster (again, for the vast majority)
>
Good point, I cannot argue with any hard evidence to the contrary, but I disagree.
My projection is that even if what you just described was true, that stagnation will only be temporary and smarphones will be become pretty much primary personal devices due to their intimacy with users. As such, they should be in centre for personal computing with other devices (tablets, notebooks, TVs) playing a support, auxiliary role. As a result, smartphone logic and storage will get ever more complex and sophisticated (app specific circuits, agressive OoOE, DRAM stacking, Flash stacking, Storage Class Memory stacking) powerful but at the same time, extremely limited in power usage. This should really open up the market for differentiation and an ever tougher SoC development race.
>
>
> I remain skeptical about whether Intel really has a strength in microarchitecture
> design for low-power devices - it seems to me that's a very different problem than
> high-end core design, and also one where both ARM cores, and AMD's Bobcat, seem
> arguably as good as Intel's efforts so far.
>
Bobcat is a much higher power design, can't touch smartphones. Atom can scale-up, but it seems Haswell may scale down for tablets after all. I think that's pretty darn impressive compared to where Intel was only a couple of years ago. Admittedly, it took many good years for them to produce these, and will take another 1-2 to really get there where they are headed.
>
> The process disadvantage only comes into play if Intel is willing to use bleeding-edge
> fab capacity for sub-$20 ASP smartphone chips instead of over-$60 laptop chips,
> which seems very questionable (even given the likely ratios of die area, the revenue
> per mm2 would seem unlikely to favor the smartphone chips).
>
Assuming you are right that SoC ASPs will erode and that revenue per sqmm will also be much lower. On the other hand, Intel claims Atom will be on process parity with other products by 2014, on the 14nm node. So even it won't be the very first to gain access to a new process, they won't lag behind by more than 6-9 months. Which is still an huge advantage over TSMC, GloFo, what have you. Especially when you consider their history for paper launching new processes without any volumes coming out for another 3-4 quarters with products that push the evelope.
> It seems that part of the reason why they sold XScale in 2006 was that it couldn't
> compete effectively against TI and others on old fabs, and they weren't willing to use the
> new fabs for a low-margin product. Possibly that changes if the decline of x86
> continues, we'll see.
>
Exactly, times are very different, and they may not really have a choice but give Atoms priority. After all, they won the PC and server wars against everybody else, why bother? Delaying a PC or server chip by 6 months won't really make a difference, AMD is irrelevant and incapable beyond belief.
>
> It's the same thing, isn't it ? They can make good chips, but can they sell them
> at a price they'll be happy with ? There seem to be quite a lot of non-technical
> obstacles to achieving anything Intel would count as "success" - viz, enough revenue
> *and* gross margin to keep building big fabs even as x86 declines.
>
Sadly for Intel, not really the same. The smartphone market is not that fluid at all on the supply side. You can't just gradually penetrate the market with small steps, you have to win over OEMs in chuncks.
Intel has to break the ARM ecosystem. Winning some designs with Motorola here and with ZTE there just won't cut it. You have to build huge commitments that run into buying millions and tens of millions of chip a year.
Simply having a better chip is necessary but not enough. Intel doesn't need to be simply better, it has to maniacally race ahead and leave Qualcomm and the likes so far behind that will make at least one big OEM to switch big time.
If you are right and that smartphones are reaching a "good enough" altitude, it would be disastrous for Intel's smartphone efforts -- but it would also mean the PC is protected from further cannibalisation. That's not too bad if they can resume growing the global PC penetration, which is still rather low.
> Daniel Bizo (fejenagy.delete@this.gmail.com) on February 4, 2013 3:45 pm wrote:
>
> I would agree the numbers are questionable, and if anyone cares to present evidence
> for different numbers, I would welcome better information.
>
Fair point, I don't care enough to dig up examples. I've seen Apple iPhone 5 BOM analysises that varied greatly for the SoC.
>
> I think that trend may reverse, because a) with A15/Krait-class cores in future,
> together with better-optimized Android software, it seems that dual-core phones
> will be sufficient for the vast majority of users; b) now that screen resolution
> has reached 1080p, the pixel counts aren't likely to increase much further so the
> GPU's don't need to get much faster (again, for the vast majority)
>
Good point, I cannot argue with any hard evidence to the contrary, but I disagree.
My projection is that even if what you just described was true, that stagnation will only be temporary and smarphones will be become pretty much primary personal devices due to their intimacy with users. As such, they should be in centre for personal computing with other devices (tablets, notebooks, TVs) playing a support, auxiliary role. As a result, smartphone logic and storage will get ever more complex and sophisticated (app specific circuits, agressive OoOE, DRAM stacking, Flash stacking, Storage Class Memory stacking) powerful but at the same time, extremely limited in power usage. This should really open up the market for differentiation and an ever tougher SoC development race.
>
>
> I remain skeptical about whether Intel really has a strength in microarchitecture
> design for low-power devices - it seems to me that's a very different problem than
> high-end core design, and also one where both ARM cores, and AMD's Bobcat, seem
> arguably as good as Intel's efforts so far.
>
Bobcat is a much higher power design, can't touch smartphones. Atom can scale-up, but it seems Haswell may scale down for tablets after all. I think that's pretty darn impressive compared to where Intel was only a couple of years ago. Admittedly, it took many good years for them to produce these, and will take another 1-2 to really get there where they are headed.
>
> The process disadvantage only comes into play if Intel is willing to use bleeding-edge
> fab capacity for sub-$20 ASP smartphone chips instead of over-$60 laptop chips,
> which seems very questionable (even given the likely ratios of die area, the revenue
> per mm2 would seem unlikely to favor the smartphone chips).
>
Assuming you are right that SoC ASPs will erode and that revenue per sqmm will also be much lower. On the other hand, Intel claims Atom will be on process parity with other products by 2014, on the 14nm node. So even it won't be the very first to gain access to a new process, they won't lag behind by more than 6-9 months. Which is still an huge advantage over TSMC, GloFo, what have you. Especially when you consider their history for paper launching new processes without any volumes coming out for another 3-4 quarters with products that push the evelope.
> It seems that part of the reason why they sold XScale in 2006 was that it couldn't
> compete effectively against TI and others on old fabs, and they weren't willing to use the
> new fabs for a low-margin product. Possibly that changes if the decline of x86
> continues, we'll see.
>
Exactly, times are very different, and they may not really have a choice but give Atoms priority. After all, they won the PC and server wars against everybody else, why bother? Delaying a PC or server chip by 6 months won't really make a difference, AMD is irrelevant and incapable beyond belief.
>
> It's the same thing, isn't it ? They can make good chips, but can they sell them
> at a price they'll be happy with ? There seem to be quite a lot of non-technical
> obstacles to achieving anything Intel would count as "success" - viz, enough revenue
> *and* gross margin to keep building big fabs even as x86 declines.
>
Sadly for Intel, not really the same. The smartphone market is not that fluid at all on the supply side. You can't just gradually penetrate the market with small steps, you have to win over OEMs in chuncks.
Intel has to break the ARM ecosystem. Winning some designs with Motorola here and with ZTE there just won't cut it. You have to build huge commitments that run into buying millions and tens of millions of chip a year.
Simply having a better chip is necessary but not enough. Intel doesn't need to be simply better, it has to maniacally race ahead and leave Qualcomm and the likes so far behind that will make at least one big OEM to switch big time.
If you are right and that smartphones are reaching a "good enough" altitude, it would be disastrous for Intel's smartphone efforts -- but it would also mean the PC is protected from further cannibalisation. That's not too bad if they can resume growing the global PC penetration, which is still rather low.
