- what if
those 14nm chips were 3 times more expensive per transistor than standard 28nm dice on low-power process, and only 50% more energy-efficient? What if production were bound to most ridiculous contractual constraints?
Even Asus has went full blown ARM for their mobile stuff while tossing the scraps for Atom. For a company that practically lives and dies by Intel that really shows how much love they show for their masters.
You wouldn't be far off.
Here is another one of Intel's big problems. Let's say 14nm is 50% more energy efficient than 28nm at low TDP, in terms of phones that's barely even a Watt difference? The other parts are still going to be consuming the same power as any other phone and that saved Watt of power is never going to be worth the extra cost.
Sooner or later Intel is going to be forced to fight the rest on an even fab footing, simply due to diminishing returns.
You wouldn't be far off.
Here is another one of Intel's big problems. Let's say 14nm is 50% more energy efficient than 28nm at low TDP, in terms of phones that's barely even a Watt difference? The other parts are still going to be consuming the same power as any other phone and that saved Watt of power is never going to be worth the extra cost.
Sooner or later Intel is going to be forced to fight the rest on an even fab footing, simply due to diminishing returns.
- what if
those 14nm chips were 3 times more expensive per transistor than standard 28nm dice on low-power process, and only 50% more energy-efficient? What if production were bound to most ridiculous contractual constraints?
You wouldn't be far off.
Here is another one of Intel's big problems. Let's say 14nm is 50% more energy efficient than 28nm at low TDP, in terms of phones that's barely even a Watt difference? The other parts are still going to be consuming the same power as any other phone and that saved Watt of power is never going to be worth the extra cost.
Sooner or later Intel is going to be forced to fight the rest on an even fab footing, simply due to diminishing returns.
My s4 with s600 use only irrelevant amount of power when idle. Its hardly zipping when calling. What matters by far is oled screen efficiency. Soc eff. Is already plenty fine. Gsmarena battery bencmarks tell very much that.
I simply fail to see just the slightest good argumentation to use the extreme high fab cost here. For that to be a success Intel needs a completely new usage pattern.
Process nodes are developed first and foremost to reduce the production cost per transistor.
Improving the electrical characteristics of the transistors is purely secondary.
You don't use expensive fabs to make chips faster unless you can sell the faster chips for more money (or unless you need faster chips just to remain competitive with what your competitors are bringing out).
You use your expensive fabs to make all the less expensive chips.
If, as a fab owner, you are not arriving at those end results then you are doing it wrong.
When a node is designed, when its spec's are drafted and endorsed during the first year of development, an enormous number of engineering tradeoffs are factored into the decision matrix which ultimately determines the primary goals of the node.
If you are still doing areal shrinks then reducing production cost is going to be a primary objective (Intel 14nm) and performance benefits are going to be secondary (bit not non-existent) throughout the entire 4yr development cycle.
If you are not doing areal shrinks then increasing capability of the xtors is going to be the primary objective (GloFo 14nm-XM and TSMC 16nm) and production costs are only going to increase per transistor.
For the people saying "this is fast enough, I don't need lower power or faster processing speed" then you are actually endorsing Intel's sub-20nm roadmap strategy versus that of everyone else who has announced the shrink details of their sub-20nm strategy roadmap.
The foundries are all going for more expensive (no significant cost reduction from areal shrinking) chips that have lower power (finfet). Intel is going for lower cost because they get the areal shrink with their 14nm.
This might help drive your point further home, IDC
Although Intel will be improving performance over the 22nm FinFET node at 14nm FinFET, as well:
cost per transistor is no where near that simplistic. The real test will be, if TSMC/GF 20nm only provides a shrink not a perf increase. lets see who shrinks what and when then.
I wonder at what point we become I/O bound on the "typical" phone/tablet SOC .
Double/Triple/Multiple Patterning is not about speed boosts, but about getting the actual shrink. Without EUV, the further down the nanometer road, the more $$$ has to be spent for fabs (fixed costs, presumably), and for multiple exposure of wafers (running costs, obviously) with each consecutive node. That's where first mover's advantage is to grow in costs, but to diminish in returns, due to the whole setting taking place in a saturated good-enough-market. As those same technological and economic constraints do apply to Intel in the same way as to any other foundry, there is no way saying they can easily produce smaller and cheaper than their combined competition. They can try to somewhat reduce the amount of layers up for multiple exposure (as it is the case with fdSOI using a significantly reduced amount of layers), but they certainly cannot provide significantly reduced die costs on a volume equal to their competition, absent some serious cross-subsidizing techniques. Those graphs are deluding, not to say euphemistic at best.
And yet Intel seems to be doing alright for themselves ($2B profit) selling their 77W/84W TDP 160mm^2/177mm^2 22nm chips while AMD is struggling selling their 125W/220W TDP 315mm^2 32nm chips.
In the end people can extrapolate doom and gloom scenarios for Intel all they want but the reality to date kinda suggests that maybe, just maybe, the folks at Intel (from managers to accountants to engineers) might just know what they are doing and they might just know what they need to do going forward as well
Qualcomm seems to be doing alright for themselves ($2B profit) selling their chips as well, without needing to pay any capex on R&D on fabs.
You might what to investigate what Qualcomm actually makes money on. Their MPU sales for the entire 2012 was "only" 5.3B$. In short, MPU sales is a niche at Qualcomm.
And yet Intel seems to be doing alright for themselves ($2B profit) selling their 77W/84W TDP 160mm^2/177mm^2 22nm chips while AMD is struggling selling their 125W/220W TDP 315mm^2 32nm chips.
In the end people can extrapolate doom and gloom scenarios for Intel all they want but the reality to date kinda suggests that maybe, just maybe, the folks at Intel (from managers to accountants to engineers) might just know what they are doing and they might just know what they need to do going forward as well
$5.3B, the vast majority in phones which is what we're talking about. It'll be a bit higher this year and next with their domination of tablets as well. Meanwhile Intel's other architecture group loses $600m a quarter.
$5.3B, the vast majority in phones which is what we're talking about. It'll be a bit higher this year and next with their domination of tablets as well. Meanwhile Intel's other architecture group loses $600m a quarter.
Huh? They can't even beat Tegra 4 how have they the best "tablet chip"?
And there are no S800 tablets announced. Or are we talking here about 5,5 and 6,5" smartphones?!
Let's start with the 800 and that claim of 1.75x performance. Assuming it turns out to be true in practice, it'll derive from the following improvements relative to the S4 Pro:
a shift to a 28nm HPm ("High Performance for mobile") fabrication process
an upgraded Hexagon V5 digital signal processor
the adoption of 800MHz LPDDR3 memory to match up to the PC-like 12.8GB/s bandwidth of Samsung's Exynos 5
an all-new Krait 400 architecture running at higher clock speeds of up to 2.3GHz
an equally new Adreno 330 GPU that'll offer a double helping of compute power
This SoC will be able to capture and play 4K video at 30fps, handle 2K video (or more precisely 2560 x 2048) at 60fps, and deliver the latest 7.1 channel DTS-HD and DD Plus audio standards.
The 800 should arrive by the middle of the year, but another upgrade option will get here even sooner, by Q2. This is the Snapdragon 600, which can be thought of as an S4 Pro with some significant architectural tweaks, a higher 1.9GHz max clock speed, a "speed enhanced" Adreno 320 GPU, and LPDDR3 memory. Qualcomm is specifically angling this chip against Clover Trail and Tegra 4, which suggests we could see it in a range of form factors very soon.
In his statement, which coincided with Intel's statement of mobile market targeting intent to its investors a few hours ago, Qualcomm's Raj Talluri said "Youll see a whole bunch of tablets based on Snapdragon 800 in the market this year". He spoke directly about the Intel competitive threat, adding "Theres a lot of talk about Intel and tablets. Clearly we see them still being far behind in mobile".
Krait 400 seems to do very well against ARM's Cortex A15, trading positions in terms of performance depending on the test. As these are browser based benchmarks there's a big software component to variability that prevents big conclusions from being made here, but it's clear that Snapdragon 800 is in a similar performance class to current Cortex A15 based designs.
The Java and Native client AndEBench tests echo what we've seen elsewhere: Snapdragon 800 can definitely be quicker than ARM's Cortex A15, and at least is in a similar class.
Qualcomm's Snapdragon 800 is quite possibly its most ambitious SoC to date. The goal? To drive absolute performance while maintaining power efficiency. While Snapdragon 600 was clearly about delivering evolutionary gains in performance, Snapdragon 800 intends to compete with ARM's Cortex A15 and Intel's Bay Trail platform.
On the CPU performance front, Snapdragon 800's 2.3GHz Krait 400 cores do appear to hold their own quite well against ARM's Cortex A15. In some cases ARM holds the advantage, while in others the higher clocked Krait 400 takes the lead. We still have the question of power to answer, but Qualcomm bets it can deliver A15-like performance without A15-like power thanks to the 28nm HPM process at its foundry partners.
Qualcomm didn't have any power demos setup, so power analysis and battery life performance will have to come at a later date, but the claim is better performance at equivalent platform power as Snapdragon 600.
On the GPU side, we have a new king. Adreno 330 delivers huge performance improvements over Adreno 320 and everything else we've tested thus far. Snapdragon 800 is the new benchmark to beat. It's very clear to me why many tablet designs scheduled for later this year are based on Snapdragon 800 silicon.
That's the scariest part for Intel, I think. Qualcomm makes an overwhelming majority of their profits on smartphones, and to this date they've been relatively absent in the tablet space (minus a couple of design wins for the S4 Pro). This year Qualcomm won't just have the best smartphone SoCs but also the best tablet SoCs with the Snapdragon 600 and 800, so their revenue should look even better. Bay Trail will provide some competition, but the current Clover Trail chips are absolute dogs.
The foundries are all going for more expensive (no significant cost reduction from areal shrinking) chips that have lower power (finfet). Intel is going for lower cost because they get the areal shrink with their 14nm.
The 20-LPM process claims a 40% reduction in power from the 28nm generation, and the 14XM claims 40%-60% increased battery life over 20-LPM. The 20nm generation is scheduled for next year, and as noted earlier 14XM is due out in 2014, a year later, breaking the two-year cadence that we've all got used to. Apparently 20nm wafers are running the full process in the Malta, NY fab right now.
They're accelerating the process launch by using the 20-LPM middle/back end-of-line metal stack with the finFET front end. In the 20nm process the 1x metal pitch is 64nm and the single-patterned metal is 80nm -- coincidentally, the latter is the same as Intel's tightest pitch in their 22nm product.
The economic challenge in going to 14nm is almost as huge as the technical challenge, and keeping the cost/power/performance (CPP) metric in check as process complexity spirals upwards has caused inevitable concern. In particular, the cost benefits of shrinking die size tends to go away as the lithography demands double, triple, and even quadruple patterning.
So if GLOBALFOUNDRIES, or any other foundry, wants to keep the customers coming, they have to mitigate the cost increase going to the next node. Taking a hybrid approach such as the 14XM process should be an attractive option for their existing and future customers.
It's interesting to note that TSMC has changed tack slightly and are now saying that they will be using finFETs at 16nm, not 14nm. They are also claiming that their 20nm metal pitch is leading-edge at 64nm, although that's the same as GF's. It's tempting to wonder if TSMC will also use a hybrid approach and transfer their 20nm back-end to the 16nm node, since the arguments are the same. Chenming Hu thinks so , anyway. TSMC are predicting 16nm risk production in 2014.
Profits selling MPUs? No.
Profits licensing technology, aka patents? Yes.
Qualcomms biggest and most profitable division doesnt make products.
Profits selling MPUs? No.
Profits licensing technology, aka patents? Yes.
Qualcomms biggest and most profitable division doesnt make products.
Profits selling MPUs? No.
Profits licensing technology, aka patents? Yes.
Qualcomms biggest and most profitable division doesnt make products.