Observing the A11's Heterogenous Cores

I've always found this experiment to be revealing:

>Specifically, after explaining briefly that I was the editor of a tech news site, I would ask iPhone users “why do you use an iPhone?” and Android phone users “why do you use an Android phone?” My phone was always in my pocket when I asked, and I made sure to never mention a rival platform at all.

In about three months I ended up asking 219 different people this question, including 112 iPhone users and 107 Android users. I wasn’t terribly concerned with the features of each platform that people liked most. I just wanted to see how many people, unprovoked, would mention a rival platform in a negative way when offering their responses to a complete stranger.

http://bgr.com/2016/10/25/iphone-vs-android-survey-takeaways...

I guess you consider me one of those offended, and let me assure you, I am not.

It is just weird if a technical presentation doesn't even mention the environment (lack of a better word) at all, but considers PowerPC multicore history for Mac relevant, which technically I consider not.

There are people out there, that don't consider technology or innovation to be Apple or other-vendor specific, heck, go ahead and stone me, I don't differentiate between Macs and PCs since Apple uses Intel CPUs.

Even further, I totally don't care who invented it.

That's an interesting thought. You make a lot of sense to me, although I don't know that much about hardware design.

I would like to note, though, that the A11's high-efficiency cores are still pretty fast. In this (highly artificial) benchmark, they were able to accomplish more work in the same time than the high-performance cores. Since there's twice as many of them, that means they're at least 50% as fast on this code. So I don't know if this really will be the future, but the "slow" ones aren't useless by any means.

I disagree: we've been playing these games for a long time and I don't see anything on the horizon.

You know how you start a phone call in Japan with Moshi moshi‡? In the 1990s the software in some of the PHS phones would wait until there was enough energy coming through the microphone before powering up the compression hardware in order to save just a little more power. So the "Moshi moshi" would get cut off.

Basically anything to save power -- people hate recharging.

‡ It's like the linguistic equivalent of the old TTY STX (which still survives in ASCII!)

As someone relatively naive on these things, iOS seems like the kind of platform where it's easier to integrate than the more generic case. Much easier for them to optimise for it, since (1) Apps are often using Apple-provided frameworks and you can optimise around those specifically

(2) iPhones are generally performing a single task in the foreground and frequently background tasks. Even with an iPad that moves up to only 2 simultaneous tasks. Seems much easier to optimise the scheduling based on that.

(3) There is a very limited set of hardware developers have to optimise for, literally basically 1 CPU per year. On that basis, any quirks will likely be found in development and can be fixed then rather than having different quirks accross a wider range of CPUs that you might have seen in Android [since 1 and 2 would apply reasonably equally to Android devices]

Even SMP chips today don't make scheduling particularly easy. Some cores share caches and some don't, some are on different sockets, some don't share the same memory banks, some are clock boosted or down clocked...

IOW you think that the folks working on these things are idiots, or lazy, or perhaps that your insights are so rich and rare that they can’t possibly have occurred to any of them?

Has it occurred to you that perhaps all of those things you’re mentioning have already been done, along with all the other easy ones, and the reason that something like this that’s harder to exploit is only just surfacing is because it got put off because it’s so hard to exploit?

Gerard &co. are a lot better at this than you give them credit for.

This is a bad HN comment, not because you don't have anything solid to contribute (I'm sure you do), but because by choosing an uncharitable reading of the OP just in order to smack it down, you make the discussion less specific and substantive.

That's why we have this guideline: "Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize."

https://news.ycombinator.com/newsguidelines.html

> But he described the Apple ecosystem as if nothing outside it even exists.

I did no such thing. I focused on Apple because my blog focuses on Apple, but at no point did I state or imply that nothing outside of Apple exists. I even went out of my way to say that multicore systems have "become the norm in many parts of the computing world."

In no way the author — a vey respected engineer in the Apple community BTW — pretended to provide an exhaustive overview of the multi-core heterogeneous computing world (or assert that Apple invented all this tech on its own).

I honestly think that he just wanted to provide a very brief comparison between A10 and the latest A11, and then go on with his article core subject ;)

Unless he edited the post after you commented, I don’t quite agree with the comparison. He states “in the Apple world” as a qualifier to his history lesson. He doesn’t say “Apple invented multi core CPUs” or “Apple had multi core CPUs first”.

I didn't see the author present any technology as being novel industry-wide.

He did present big.LITTLE architecture as novel to the Apple and iOS ecosystems, which is correct.

Indeed, other big.LITTLE like designs are out there besides the A11. But the article isn't strictly focused on "Apple is a superior innovator."

> My theory was that this should show a strongly bimodal distribution of running times on the A11.

...and then he proceeds to devise a test. This is ultimately how you might verify the functionality of the hardware and the Darwin scheduler. If it were naive, the test might not have behaved this way.

BTW if anyone wants to do similar experimentation with a MUCH cheaper target, you can use the ODROID Xu4 (Exynos 5422, < 100 USD).

Wikipedia says [1] that ARM big.LITTLE "Heterogeneous Multi-Processing (HMP), which enables the use of all physical cores at the same time" has been "implemented in the Samsung Exynos starting with the Exynos 5 Octa series (5420, 5422, 5430)", available in Q3 2013.

1 - https://en.wikipedia.org/wiki/ARM_big.LITTLE#Heterogeneous_m...

>Is the author seriously so deep in the "Apple-is-the-world" bubble that he has not yet heard of BigLittle?

Why the rude comment?

I'm not familiar with the current state of bigLITTLE, but at least initially, bigLITTLE clusters weren't capable of heterogeneous compute; you could shift between cache-coherent clusters (use either all four big cores or all four LITTLE cores) or have virtual cores, each made up of one big and one LITTLE core, with the kernel managing which physical core was actually executing threads assigned to the logical core.

My understanding of the A11 is that it's fully heterogeneous with the capability to run and assign code across any permutation of its physical cores.

It's an Apple-focused blog. The A11 may not be doing anything new when considering the wider world, but it's the first Apple CPU to have it.

Do not confuse "I did not mention this other thing that is not entirely off topic" with "I was not aware of this other thing."

I couldn't find any quote in the article that supports the suggestion that the author isn't/wasn't unaware of big.LITTLE.

Could you cite one that supports your snark?

Well, it's benchmarks are different. It's crushing all the other mobile processors.