I don’t think in the terms of reference for the development of another mobile system-on-chip was a requirement to break the multi-core competitors in the benchmarks. The main problem was still the same: to make the following smart phone and tablet with the best in the world. Wanted as better, it turned out as always. Formally, the new a chip used 4-core processor. And amazing stories about the 2-core processors, superior 6-, 8 – and even 10-core CPU competitors like gone. But the world is more complicated than it seems.
It is unpleasantly to lose, especially if the prize Fund is billions of dollars, yet in the accounts of millions of seekers mobile adventure. And those who lost, often behave badly. Claim that it was rigged. Creating the processor at one or two specific devices to achieve killer benefits snap (and who they were forbidden to do the same, if it’s easy?)
“Victory breeds hatred, the defeated live in sorrow” – wrote the Holy book of the Buddhists, by unknown author in the 6th century BC.
Suggested solution: simply give up the struggle immediately and the victory, and from defeat and you will be happy.
Apple was accused of old-fashioned, in that it does not use the fashionable advanced technology, or starts to use them when they are about to expire. A strange kind of company produces beautifully designed stuff, expensive, deciding for those who spends the money and how they buy – and is a leader in the industry. Magic?
In the Middle ages Apple has long been just burned.
The processor in Apple’s A10 Fusion, being a 4-core, behaved in tests and in real life as 2-core. And still left behind 6 – and 8-core rivals. And disgustingly happy Phil Schiller was limited, as always, a minimum of information.
And architecture used in Apple’s A10 Fusion is destined to remain the peak of fashion less than a year. ARM Holdings presented the architecture of big.LITTLE in October 2011. In October 2012 ARM announced the first practical implementation of these ideas. At the end of 2013 big.LITTLE became a reality. In 2016, it was “old news,” technology of fermented and insisted – and now the wizards of Central California drew attention to her.
Or, rather, has seen fit to use her for their own purposes. Anyway, Apple’s A10 Fusion replaced the Apple A9 system-on-chip still present in the tops but pushed back with first place, and Apple was once again the leader.
And six months after the release of Apple’s A10 Fusion in may 2017, ARM unveiled the successor of the big.LITTLE, architecutre DynamIQ.
This continuation of the series Pro processors from Apple. Previous parts:
The first part: In the shadow of the Apple A4;
Second part: the birth of the “Apple” of the processor;
Part three: Returning to the beginning of beginnings: Apple A6/A6X (Swift);
The fourth part: the Cyclone comes to replace the Swift (in the Apple A7);
Fifth part: one “NeXT” or coprocessor for fitness (Apple M7);
Sixth part: turns into Typhoon Cyclone (Apple A8);
Seventh part: the Apple S1: riddle enclosed in a mystery;
Eighth part: Apple A8X: the Graphics in the zone of special attention;
Ninth part: Apple A9 – the deceiver, the scoundrel and chipset;
Tenth: Apple A9X: Nuclear physics?.
Big LITTLE architecture
In 2011 in the sector of processors for mobile devices has developed a revolutionary situation. On the one hand, not to lose in the most severe competitive struggle, their computing power had to increase. And for everything in the world you have to pay. Big power is the elongation of the conveyor, increasing the size of caches at all levels, multiplying the number of computing units. And the rapid increase in the number of transistors on a chip that consume energy even in “sleep”state.
The peak load for which it was collected the whole army, was episodic. Most of the time the many millions of microscopic switches ate, dozed, and waited. More precisely, the moment. Battery discharged faster, waste of millions of transistors (their heat) needed to somehow “take out”, otherwise the system was applied to irreparable damage.
Anyway, with something similar in microelectronics have already encountered. Intel and AMD have found their own solutions to common challenges, about how they succeeded in the presentation for those who use processors, was well and widely known. But the specific nature of these decisions, of course, was the most guarded secret of the industry leaders.
ARM Holdings, proposed a solution: to use the processors in different cores, with different performance and energy consumption. “Large”, for peak loads and “small” for everything else. Architecture called big.LITTLE, and spoke about it in October 2011. The obstacles to realizing a world of universal dreams were many: a processor with this architecture we had to teach on time and very fast to turn on and off large or small nuclei. This is just the tip of the iceberg, in fact to all of this mechanic to work, the processors were required to master a lot of difficult tricks.
The reverse side is an ambitious task like this is a godsend for talented engineers, excellent opportunity to learn, to try their strength, and in the end it’s very interesting. And yet – each project makes the world a little smarter. And at the same time silly (here, too big.LITTLE), but let’s not about sad.
To play these games, the cores of both types was required to develop anew, taking into account the participation in this game.
Rapidly-quickly switch between modes took 20 thousand CPU cycles. Even at clock frequencies of 500 MHz (in 2011 met and such) is 400 microseconds.
Initially it was assumed that switching will be “clustered” in each unit of time that the processor will use only the “big core”, or just “small”. 20 thousand cycles is the price for this approach.
In the process of using this architecture, there were two more ways of its application. The switch could have happened independently, in a single pair big-small, depending on the demands of your executing this pair of tasks. And even, cunning and not very elegant way, big and small were allowed sometimes to work at the same time.
Each of the three methods, their pros and cons. Apple engineers chose the easiest and most economical way: switch all at once, at any time working either “large” or “small”. Practical number of engines under this approach is equal to half of their physical numbers but on the order of less than reasons to fail.
2-core 4-core processor is Apple’s A10 Fusion and some of his relatives. Practice – criterion of truth. Top positions in tests (and in the most realistic, in particular) made a speculation about the backwardness and mental deficiency Apple groundless. But I assure you, Apple engineers have spent time and effort not on it.
Architecture of big.LITTLE – the most important and noticeable difference in the Apple A10 processor than its predecessor. But not the only one. Around this architecture arose such a number of prejudices and misconceptions that had given her so much time. Sorry, if that.
Artifact from the Fruit of the galaxy
In addition to Apple engineers, and ARM, cut off in full in the process of creating Apple A10, and users got their hands on the product of high technology and inspiration, there were other lucky ones who have experienced with what incomparable pleasure from solving very difficult puzzles of nature. About Apple’s A10 Fusion in the concert hall named bill Graham said many words and said practically nothing.
You still spend time on vulgar puzzles and crosswords?
Decipher the details and prove the correctness of his interpretation was very difficult, it requires a high qualification and deep knowledge. In this way we made some mistakes which were corrected over time.
Even sophisticated and expensive equipment, including a powerful electron microscope cannot give a final answer to all questions.
Decoding the first images from “the back side of the man-made moon” was incorrect, that it published in most of the essays about the new system-on-chip:
To honor for the guys from Chipworks, they doubted its correctness. To recognize its accurate interfered with the illogic and lack of elegance resulting picture. Apple handwriting. So – the question marks.
Hard work mind, discussions, comparison of fragments of opened landscape exoplanets Apple’s A10 Fusion with other landscapes, the exact details of which are given in the technical documentation of other worlds (other companies) helped to reveal the true nature of things.
A faithful transcript:
This is also a hypothesis. Apple did not comment on the opening of voluntary researchers, but this time the decoding is more likely to better agree with the test results, and 99.9% true.
In our world a feat!
The hurricane and the gentle breeze
The secret was not only the location of objects on the crystal A10. Almost all of them. The processor was designed by Apple for Apple. To excite the interest among the masses about him, it was impossible to tell, but the most interesting.
The rest is the result of hard work of dozens (hundreds?) good minds, proven and rechecked all the rules of real science. Thank you to them.
In the world processor (code name T8010, apparently) called Cyclone-4. Fourth generation 64-bit processors from Apple. At the same time found out what was extracted from the plist files in previous versions of iOS. It was the name of the processor cores. And Swift and Cyclone are the names of cores. While he was exactly the same, it didn’t matter. It was different now.
Large and voracious kernel was called Hurricane. Hurricane. But one of the types of cores in the Cyclone-4 was the name of another British fighter of WWII, the hurricane. It really was a large kernel, on chip, each of them held of 4.18 sq. mm.
The maximum clock frequency of Hurricane – of 2.34 GHz (in adverse conditions, when overheating, for example, the frequency braked).
Little was called Zephyr. Aside from the obvious, this word has other meanings, for example – “light breeze”, “breeze”. The area of each of the small – 0,78 sq. mm.
The maximum clock frequency of Zephyr and 1.05 GHz.
Inside the case that protects the crystal from damage and corrosion, used new technology the layout from TSMC, InFO. In the body, in addition to the crystal housed RAM made by Samsung. Or 2 GB LPDDR4, in a version for the iPhone 7, or 3 Gigabytes in the iPhone 7 Plus.
Cover housing A10 Fusion for Apple iPhone 7 Plus:
On the housing cover in one of the cryptograms designated amount of memory. K3RG1G10CM-YGCH in the case of a configuration with 2GB and K3RG4G40MM-YGCH – in the case of 3 GB.
System-on-chip was produced by TSMC on FinFET technology 16 nm. Samsung did not communicate – one Chipset Apple were useless.
The GPU was, oddly enough, almost the same as in A9, PowerVR Series 7XT GT7600 Plus – that’s only “Plus” reported about some of his differences. Same 6 cores, PowerVR which persistently continued to be called clusters, recognizable pattern of nuclei-clusters on the landscape.
This is an advanced Apple variant of the PowerVR Series 7XT GT7600, 2 times more efficient and consume slightly less energy than the basis of the GPU.
In Apple documentation processor is designated as APL1W24 documentation TSMC as 339S00255, 339S00258 and probably something else – it was produced in several variants, in the beginning there were 2, then added to other. Symbols are present on the lid of the processor. 339S00258 likely represents a variant for the iPhone 7 Plus.
The second level cache (for CPU only) – 3 MB cache in the third level (for the entire SoC) to 4 Megabytes. Exactly like the Apple A9.
And iPhone 7/7+ with Apple’s A10 Fusion inside is really “tearing” of competitors, with 6 – and 8-core processors (if you agree, join in our Telegram chat.
With tests use only one core, everything is clear: the 2-core (in its essence) of the CPU or just couldn’t be. But the tests use all the cores available – please note – forced to remember the magic.
To be continued