First results from the 4.11.0 survey

[yum13241]

Mac is not a mobile platform, it mainly gets used by power users.

Explain this new thing called a MacBook. My Dad is the complete opposite of a power user. "Where did my screenshot save?" is a question I still cannot answer to this day.
Also, Linux is POSIX compliant too. Why don't Apple users use Linux then? Never mind the paradigm change, you also can't develop iOS apps on Linux, and neither can you use the convergence of other Apple products with each other on Linux. It just works, until it doesn't. (Once, my cousin wanted to create a new user on her MacBook. She didn't know how to, but I did, so I did but then "fAilEd tO cReAtE a NeW uSeR". With no explanation. At least Windows tries with it's 0xc000000d codes.

[Blzut3]

BTW, I still do not believe any benchmark that ranks Apple's CPUs with highest high end Intel models, that'd be pure magic to pull off on a limited power budget.

They're certainly a bit behind when power limits aren't in play. Although that doesn't really detract from how good the cores are since a lot of people are working with power limits (because laptop).

Mildly interesting related note, Apple's cores are strangely good at running Chocolate Doom. It's a useless benchmark and the collection method is not especially scientific, but a couple of friends and I have been running Chocolate Doom (dummy frame buffer and no sound) on every CPU we can get our hands on for the lols: https://docs.google.com/spreadsheets/d/ ... sp=sharing (See charts on timeline tab.)

No idea why Apple's core is way ahead in IPC in this very specific case, but it doesn't seem to be an ARM vs x86 thing since other ARM cores sit about where you'd expect and most of these tests were gathered with march=native (or equivalent) compiler flags (granted Chocolate Doom doesn't seem to be affected by this until you get to the pre i686 stuff).

[Professor Hastig]

Explain this new thing called a MacBook. My Dad is the complete opposite of a power user. "Where did my screenshot save?" is a question I still cannot answer to this day.

While there are private users that own Macs they mostly get used in the workplace. I have seen entire offices stuffed with Macs for the developers, that can be 20 or more in a single location doing iOS or web development, and due to the requirements of developers these are not the cheap models but the ones that cost more than a fortune to own.

Also, Linux is POSIX compliant too. Why don't Apple users use Linux then?

Yes. Why do they not use Linux? Isn't the reason obvious from the talk that's going on even here? These kinds of developers are not the type that hangs around in universities discussing abstract programming problems while not having a wider view of the world, they are also not the kind of tinkerers that spend endless time configuring their system to their liking, what they want is a robust GUI environment with good development toolchains and proper vendor support. So it's either Mac or Windows, they choose Mac for POSIX. Simple as that.

The private users are just a waste product in the end, so to say, and it mostly seems to be an American thing.

[Professor Hastig]

Mildly interesting related note, Apple's cores are strangely good at running Chocolate Doom. It's a useless benchmark and the collection method is not especially scientific, but a couple of friends and I have been running Chocolate Doom (dummy frame buffer and no sound) on every CPU we can get our hands on for the lols: https://docs.google.com/spreadsheets/d/ ... sp=sharing (See charts on timeline tab.)

So you are benchmarking a software column renderer? If that is true, my guess is that what mostly matters in these numbers is cache performance in extremely bad scenarios.

[Blzut3]

Correct, many of these recent-ish CPUs can fit the whole game somewhere into cache though. Apple's cache layout differences could definitely still be a/the reason.

[Phredreeke]

No idea why Apple's core is way ahead in IPC in this very specific case, but it doesn't seem to be an ARM vs x86 thing since other ARM cores sit about where you'd expect and most of these tests were gathered with march=native (or equivalent) compiler flags (granted Chocolate Doom doesn't seem to be affected by this until you get to the pre i686 stuff).

My understanding is
1. The scheduler in the M cores is really good at ensuring the core's ALUs actually gets utilised without having to resort to SMT.
2. The caches on the M cores are huge. 128kb+192kb L1 per core and 12-16Mb L2 per 4 cores for the performance cores.