Ashraf Eassa, The Motley Fool:
Back in 2013, Apple introduced the A7 system on a chip (SoC) as part of its then-flagship smartphone, the iPhone 5s.
The A7 was the first 64-bit ARM processor.
Now, Intel’s chips, at the time, ran at much higher frequencies (in excess of 3 gigahertz), but what the strong per-gigahertz performance of the A7 chip signaled to me was that Apple had built a very impressive base from which to build up in future smartphone chips.
And:
While Apple is great at chip design, it doesn’t manufacture its own chips — it outsources production to third parties. Apple’s A-series chips through the A7 were manufactured exclusively by Samsung, and Taiwan Semiconductor Manufacturing Company was the exclusive manufacturer of the A8, A10, and A11 chips. TSMC and Samsung reportedly split the orders for the A9.
Both TSMC and Samsung have delivered new manufacturing technologies at a breakneck pace. The performance, power consumption, and economic viability of a chip are determined heavily by the technologies upon which it’s manufactured.
Apple is relentless. As they constantly innovate in chip design, they can take advantage of whatever manufacturing advances bring the most performance advantages for each particular chip generation.
The result:
I believe that when Apple introduces its next iPhone in about four months, it will deliver equal or better CPU performance to Intel’s best notebook processors designed to consume 15 watts but at a fraction of the power consumption.
Amazing.
Is this a case of Intel resting on its laurels or did Intel completely misjudge Apple and the phone market?
For all his douchiness, Eric Schmidt certainly didn’t underestimate Apple.
Intel suffered from extreme myopia and love of its x86 and CISC ISAs, presumably as it is more proprietary to them. Don’t forget, Intel had XScale (some of the highest-performing ARMs 15 years ago) and sold it!
Instead, they put their efforts into Itanium, which was even more complex (and a total dud), and the M (now Core M) line, which they believed would overcome the complexity via process improvements and allow them to enter mobile using x86.
However, x86 chips are just way too freaking complex to shrink. I do agree they underestimated the mobile market and how poorly x86 would fit.
Apple’s chip work is their most underrated asset. They are way ahead of everyone else right now.
I wonder if Apple will make a push into ARM server chips – maybe even just for their own internal use within iCloud data centres.
The A7 was the first 64-bit ARM processor
First mobile SoC one, especially in mass production.
AppliedMicro was demoing a 64 bit ARMv8A in 2011 on FPGAs; the same day ARM announced the architecture, in fact.
APM shipped it in silicon in 2013, same as the A7 – and had samples in customers’ hands before the iPhone shipped it, so they probably technically win on Real Silicon.
(Apple’s CPU is better in terms of power and efficiency p28mm vs. 40mm process, for one].
But it’s not the first.)
“However, x86 chips are just way too freaking complex to shrink”
Howso?
(I mean, they’ve all been RISC internally since the Pentium Pro, and just translate in microcode.
As I understand it, ARM hasn’t actually been a “simple” architecture in several generations; looks a lot like an ARMv8A SoC should be “about as complex” as e.g. an Atom or i3.)
Sure, the core of x86 chips are now RISC by internal translation of x86 instructions into something RISCier. But it’s still a ton of complexity in the design of both this translation and the requirements it imposes on the RISC core.
Intel has a ton of hardware experts; I have a tough time imagining the problem is just that “Intel can’t do 10nm and Apple’s contractors can,” but rather getting Intel’s die with its size, power requirements, and heat into 10nm is fiendishly hard.
“I believe that when Apple introduces its next iPhone in about four months, it will deliver equal or better CPU performance to Intel’s best notebook processors designed to consume 15 watts but at a fraction of the power consumption.”
… and then in 2019…
Mind boggles where they will be in 3-5 years and if anyone really thinks most Macs wont be on ARM by then.