(1) What if TSMC cannot provide the best chips in spite of Apple's design advantages? Would Apple let Intel's chips get ahead and provide a better solution to Apple's competitors (i.e. other OEMs) ...just to keep its "freedom"?
Apples SoC's are *completely* different from others in the market place. They are CPU light (in core # and in clock, but are still competitive) and are GPU heavy. They also include other IP blocks to support various functionality (such as Siri). All this would be lost of if they went for a 'generic' Intel design (or for that matter a generic QCOM design). Apples SoC is tightly woven into it's OS and user experience. Apple *can't * go for a generic design.
As for TSMC, well, they have shown node after node they can deliver some of the best LP processes around (even compared to Intel). Clearly, you are talking about the future so who knows...
Apple *designs* their own SoC's and are using Samsung as the fab.
Intel would (are) compete with Apple (phone/tablet SoC's) - and lets not forget the Ultrabook - clone of MacBook.
I'm not short.
Yes, it's fantasy - TSMC doesn't compete with it's customers.
Do you honestly think Apple would drop the freedom it's gained from designing it's own CPU and SoC and jump into bed with Intel?
Further, the notion that they will be able to INCREASE royalty rates in an environment when fabrication costs per wafer are NOT going down, and
ARM's royalties are based on the % price of the chip when it leaves the fab. If wafer prices go up, so do ARM's royalty revenue.
As more ARM IP is used on the lower nodes (soft and hard IP) you are also seeing the % price *increasing*.
Intel's "Bay Trail" will crush the ARM licensees' hopes of ever getting into the notebook space, "Avoton" will similarly mop the floor with the wannabe, less-than-AMD chumps like Applied Micro who think that they will be able to waltz in with 2 generation old process tech and an anemic R&D budget and take significant server share.
If you take Intel's "Bay Trail" performance numbers at face value, they will struggle to meaningfully beat *existing* ARM based solutions (in CPU). Remember Intel had to put up an low end core based SKU against an ARM based solution in the chromebook. As for notebook's, it really depends what you call a notebook...
As for Applied Micro, yes there is a level of hype but you are underestimating what a clean ISA can do from a design perspective. I would certainly expect X gene to beat Avoton in CPU on a core/thread basis, even two nodes behind. One of the issues Applied Micro is going to have is getting the compilers to spit out decent code:)
These server parts (from the ARM collective) are *not* about CPU performance but *right sized* performance based on the workload . Basically they have huge IO bandwidth potential compared to Intel's SoC's and it's this along with power/price arguments that *might* give them traction.
ARM.s sigular advantage is that ARM provides the only CPU's that everyone can use in their increasingly integrated product designs. If someone wants to design a CPU part they have no choice.
Not true. Even today you still have options, such as MIPS. If you want to leverage the ISA and the ecosystem for mobile, then ARM is a good choice.
ANYone with any capability like Apple and QUalcom will actually desing their own CPU's using only the instruction set.
Indeed. Why do you think that is?
ARM has historically also been low power only because the they have provided low performance CPU's. Power goes up exponentially with performance. You can not fairly compare powerformance per watt at differnet performance levels.
No. ARM's first chips where for high performance PC's/Workstations. The low power profile was an accident rather than by design.
Yes, you can compare performance/watt across the whole spectrum. From the highest performing server chip to the lowest micro controller that is capable of running your workload.
Intel has an enormous manufacturing advantage. Don't you think a 50% power advantage and 50 cost advantage will make a differene in the market?
Perhaps, but Intel has to deliver on this manufacturing advantage - and it needs manufacturing advantage as it's SoC's are that much larger . Historically, Intel's LP processes haven't been as good at the foundries.
What is the problem with Smasungs big/little anyway. Seems to be some problems there. Pretty stupid design compared to Intel standards Intel will better control dynamic voltage and frequency in future chips and better optimize performcnce and power without having to have two different CPU's
Big/Little is a good concept on paper and if it works in the real world then it will have the best power/performance curves of any SoC. It could change the CPU design off future of power constrained devices - if it works:)
loser, you can try to spin it and pretend you are an expert, but Pike is a leading expert in the field. He is clearly saying there are sw issues that arm servers will face.
Actually, when it comes to server workloads (on x86) I am a bit of an expert ;)
The FACT is that most of the workloads that the initial ARM hoard are targeting just *work* out of the box. Yes, there are issues such as a common BIOS model but other points that Pike makes are rather dubious. As an example, no one knows the best approach to manage these microservers. When you've 10K nodes in a cluster (x86 or ARM) the current VM approach doesn't work.
yes I read it and it reads there are software issues. what were you reading bu**head?
and did you *understand* it?
The target market for these micro servers *is* the LAMP stack. Does LAMP run on ARM?
shorts, arm can only compete in microserver market which is 2% of total server data center market. even in this market they will not make any inroads because they are 4 years behind in technology process. Intel's atom avoton will kill arm in this market before it even begin. don't forget, arm is incompatible with current sw. Just look at the surface rt as an example. no company will risk going with arm.
No. There is no software issue (as the workloads are based on the linux stack). While Intel does have a process advantage, the ARM crowd have an integration advantage (ie, networking, IO on the SoC). As for companies not risking it, there are quite a few willing to try...
As for the size of the market (and ARM's/Intel's share), thats a big unknown.
You epitomize the stupidity of the ARM crowd.
If you say so;)
So you think rising prices and slow performance increases for ARM Soc's are going to work out good when competing with Intel chips that will be declining in cost with substantially increasing performance and battery life
If you say so;)
Why is 64bit an issue? Why is software incompatible?
More like 1 TB. Where did you get this wrong idea from ?
Just because a chip can theoretically address lots of memory doesn't mean it can. This chip is limited to 8GB. As for why, perhaps alexander is right, and it's just a simple tech reason, but I suspect Intel is limiting the market for these cheap sever SoC's to protect its margins.
I don't need a stinking analyst to figure this out ...analysts are trailing indicators
how many times did I say : NO more cheap ARM Socs - well it's about to sink in
Looking ahead, at 20nm,
Why is this a problem for ARM? ARM earn royalties based on the price of the chip. If chip prices go up...
If I were running a data center, I would let my competitors take the risk with ARM before I even venture into it. Especially when Intel seems to be in the forefront of delivering the best performance per watt with Avoton and its 14nm follow-ons.
ARM 64-bit is DOA!
Even with 64bit ARM cores, the bulk of Intel's server sales will be safe. These 1/2U servers are for general workloads and Intel's CPU heavy model fits (virtualization etc).
Microservers are a different matter as they are highly specialized and focused on specific workloads. Here 64bit isn't as important. Even Intel's 64bit core can only address 8GB of memory - assuming it's fully loaded.