Is ARM the death knell for the hackintosh?

[Maximara]

We can always hope that x86 hegemony will be broken, but that would require someone to make a fast enough consumer non-x86 CPU. So far Apple is the only company that does it, and they will not license their architecture to anyone.

It's not so much a fast enough non-x86 CPU but translation (rather then emulation) software. If Rosetta 2 emulated like ARM Windows does than it would blow goats with x86 code as well. Apple can do translation because they only have to support a limited amount of software.

Microsoft is having people moan and groan about not being able to run 16-bit programs on Windows 10...in 2020! This what the much "acclaimed" backward compatibility has resulted in - a user base that wants to live in the freaking dark age of computers. The only thing more ridiculous is if the Windows users were complaining about not being able to use punch cards.

 

[Joelist]

Well, since the only thing about Apple Silicon that is ARM is the ISA; and even there Apple has additional instructions I think yes it is dead. Apple has custom blocks in the SOC for a host of functions and each will have their own custom instructions.

 

[Hexley]

What would be the point of running a Hackintosh in a post-M1 world? So you can have worse performance and compatibility issues? lol.

Exactly my thoughts! If the 10W TDP M1 mobile chip can outperform a 125W TDP Core i9 desktop chip .... what more a 125W TDP Apple Silicon chip?

It would be class leading in both CPU and discreet GPU performance. Redefining performance of any iGPU.

But then again alot of people lack the foresight of this eventuality. But then for decades people have only seen an evolution rather than an revolution in performance.

It's like eGPU support coming to a future macOS system update. Is it that impactful to its users today or by early 2021 event?

Let us push some figures that are not privy to most Mac fans.

Apple globally shipped an estimated 18.35m Macs in 2019.

M1 Macs represents ~80% (~14.68m) of all Macs shipped as they are the cheapest with a starting price of $999 and $699. They are the also the smallest and lightest relative to other Macs.

How many percent of entry level M1 buyers have the buying power for an eGPU + discreet GPU more powerful and more expensive than the M1 Macs? Does it exceed ~20% (~3.67m)? Or is it more like ~1% (~185.35k)?

eGPU upport is more relevant to the ~20% (~3.67m) of higher-end Macs as they tend to be more expensive than the eGPU + discreet GPU that will be using a more powerful future Apple Silicon chip. There's a paradox though. Will these higher-end Macs even need a eGPU with a iGPU that matches or even exceeds discreet GPUs?

Best to talk about this after the early 2021 Mac event.

 

[JouniS]

What would be the point of running a Hackintosh in a post-M1 world? So you can have worse performance and compatibility issues? lol.

If user-upgradeable memory/storage/GPUs are not supported in anything less than a Mac Pro, there is going to be a price range where non-Apple hardware is more cost-effective. The range will probably overlap both high-end iMacs/MBPs and low-end Mac Pros.

 

[leman]

It's not so much a fast enough non-x86 CPU but translation (rather then emulation) software. If Rosetta 2 emulated like ARM Windows does than it would blow goats with x86 code as well. Apple can do translation because they only have to support a limited amount of software.

Not just that, A14 has dedicated hardware to emulate x86 execution environment.

 

[Maximara]

Not just that, A14 has dedicated hardware to emulate x86 execution environment.

First I have heard of this. Source?

 

[leman]

First I have heard of this. Source?

https://twitter.com/x/status/1288660944885821442

More discussion here: https://www.realworldtech.com/forum/?threadid=193883&curpostid=193883

Another thing Apple CPUs can do is switch the memory page per process. The native page size is 16KB, but it can switch to the 4KB page when running Rosetta 2.

 

[Maximara]

https://twitter.com/x/status/1288660944885821442

More discussion here: https://www.realworldtech.com/forum/?threadid=193883&curpostid=193883

Another thing Apple CPUs can do is switch the memory page per process. The native page size is 16KB, but it can switch to the 4KB page when running Rosetta 2.

Ok exactly what does the above have to do with the statement "Not just that, A14 has dedicated hardware to emulate x86 execution environment."

Note I am reading that as 'the A14 chip has an internal part that emulates the execution of x86 code.'

Buried in the discussion is the statement:
"A14 (t8101)'s ID_AA64PFR1_EL1: 0x0000000000000020 (bits [11:8] are 0b0000, meaning that MTE is not implemented in the hardware)"

The fact you have to install Rosetta 2 the first time you run x86 code means there is nothing internal to the M1 (which is different from an A14 in several key details) that translates (I really wish the people reporting this would get a clue on the difference between an emulator and a translator) x86 code. If what you said was correct then you wouldn't have to download Rosetta 2 as it would already be in the M1.

 

[leman]

Ok exactly what does the above have to do with the statement "Not just that, A14 has dedicated hardware to emulate x86 execution environment."

Note I am reading that as 'the A14 chip has an internal part that emulates the execution of x86 code.'

You are highlighting the wrong part 😉 the keyword is “execution environment”, not “execution”. The x86 to ARM transpilation is obviously done in software, why would one waste silicon on that. But transpilation alone is not sufficient. You need to guarantee that the code behavior is identical, or you might run into weird bugs. Apple has dedicated hardware in their CPUs to emulate the aspects of x86-64 behavior where it is not compatible with ARM64.

By the way, this is why the Windows on ARM execution of x86 code is fundamentally broken. You just can’t emulate Intel memory ordering on ARM in software without incurring an absolutely ridiculous performance penalty.

 

[Maximara]

By the way, this is why the Windows on ARM execution of x86 code is fundamentally broken. You just can’t emulate Intel memory ordering on ARM in software without incurring an absolutely ridiculous performance penalty.

Apple M1 Mac running Windows 10 ARM is embarrassing for Surface Pro X just added some more "fun" to the pile. The fact this is using QEMU, which not known for being the snappiest emulator out there ,and yet "Windows 10 on ARM actually runs faster – a lot faster – on Apple’s new M1 ARM-based chip than it does on Microsoft’s rival SQ2 ARM CPU which powers the Surface Pro X." (Apple’s M1 chip embarrasses Microsoft’s Surface Pro X by running Windows 10 on ARM much faster) shows just how much things have changed

While neither article gives any hint on how ARM Windows handles x86 code (likely because even with this set up it likely still blows goats) it just goes to show how good the hardware-software interaction on the M1 Macs is.