Apple Transitioning Macs to Custom Chips With 'Incredible' Performance Starting Later This Year

[Gudi]

Or will they look back on this thread like the Apple Newton threads?

Correct me if I’m wrong, but wasn’t the Newton
the very device for which ARM chips were invented in the first place? If so, the Newton is part of the iPhone success story and also the grandfather of this transition to ARM-based Macs.

 

[matram]

On the other hand, if Rosetta 2 allows Parallels and Fusion to run in X86 emulation layer, it will be possible to run Windows for X86 as a VM on the ARM Mac. The performance may be dismal, though, as it will be handled by the emulation and the virtualization layers.

Apple has clearly stated that Rosetta will not support running virtual machines.

 

[Stephen.R]

both OS X 11

... really? (a) it's been called macOS since Sierra - 4 years ago. (b).. how is it possible you didn't know the X is used as the Roman numeral 10.


You might as well say Big Sur will be macOS Eleventy-zero.

 

[djjeff]

The only advantage CISC had in 1995 was intel’s fabs.

The advantages RISC has are the same advantages it always had. X86 requires much more complex instruction decoders with microcode sequencers and microcode ROMs, much more complicated pipelines, much more complicated load/store logic, much more complicated branch prediction to compensate for the more complicated pipelines, etc. All because x86 allows writeable instruction streams, variable length instructions, ALU instructions that directly address memory, etc. For all these reasons RISC cores are much smaller than x86 cores. This allows greater clock frequency if you want it (because electrical signals travel at about 6ps per mm in a chip), less power usage (fewer switching transistors, so less charging of capacitance - power = cap * V squared * frequency), fewer gates between flip flops (which also allows higher clock frequency if you want it), fewer pipe stages (which means less of a penalty when you guess wrong on a conditional branch, etc. ARM also has more general purpose registers, which means fewer load/stores (with their inherent multi-cycle penalties), etc.

And that’s just the start of the technical advantages.

IOW: None. When processor technology was young these were considerations which could have benefit. However as processor technology has matured these benefits all but disappeared.

Take for instance instruction decoding. One of the ideas for RISC was to have an orthogonal instruction set which simplified the decoding logic. Today decoding logic is such a trivial part of a processor it's essentially a non-issue. A modern processor can fetch and decode CISC instructions just as fast as they can RISC instructions. This may have been an inherent benefit when chip real estate was limited. Today, not so much.

 

[Krevnik]

They don’t need a chiplet design. Arm cores are tiny compared to x86 cores.

Need, no. But they aren’t going to be having TSMC manufacturing on the scales of Intel (or AMD). Chiplets would benefit Apple the same way it benefits AMD: yields and pricing.

Apple’s SoCs are already pretty big. While the cores aren’t as large on the die, you are still burning considerable die space on ASICs and low power cores. Going desktop means wanting more die space for more full power cores, more iGPU space, PCIe controllers for TB3/USB4, etc. We’d be going from a 4/4 core design in the A12Z to something in the 32/? range for the (i)Mac Pro.

 

[Stephen.R]

Well he is correct that the Dock app icon was of an iMac with a Windows logo on it...I guess he is implying that this means Windows through Parallels is hopefully going to continue to be a thing??

I kept seeing this comment about "windows icon in the dock" and didn't get the reference but having watched that section the video again I can see what is referred to. It's just the Parallels icon, which happens to be an iMac with a vague representation of the Windows four-squares logo on the screen.

It's literally just an app icon. Nothing more, nothing less.

 

[anthogag]

Put it first in a new Mac mini Pro.

 

[Alan Wynn]

It’s really not so big a market that it’s worth apple’s time. They may, as you suggest, work with some partner to add the ability.

Up until now, Windows on ARM has not really happened because there are not enough customers to make it worth it for most companies to port to it. What is interesting is that this transition will likely create an installed base of ARM-based computers an order of magnitude (or two) larger than what exists today. In addition, if these machines really out spec Intel/AMD machines, it is likely that other vendors will also produce ARM based Windows machines. If one is able to run MicroSoft’s ARM based Windows (which already includes a binary compiler like Rosetta 2), one of two things is likely to happen: there will be critical mass of ARM-based Windows users that will make it worth while for people to support the platform, or we will realize that there were not that many people who cared about it.

It is also possible that these machines may be enough faster to make running transcoded apps acceptable (given that all system calls become native, there is a question as to what the performance hit will be), until this critical mass arrives.

In other words, the Mac may be the reason that Windows on ARM actually happens, just as the iMac was the reason that USB finally happened. While USB had been available on Windows, the drivers were buggy and very few devices existed given that all existing (and new) machines released had PS/2 ports and RS-232 ports, that were cheaper for manufacturers to support. The introduction of the iMac with its lack of ADB ports, serial ports, and a floppy instantly created critical mass for USB peripherals and pushed Microsoft to make its drivers work.

 

[Stephen.R]

most likely looking to nail things down with its App Store only method.

That's nothing.

I heard they're going to require all payments to be made in the form of virgin sacrifices, or alternatively recording a prayer to Cthulhu.

How far fetched is this thread going to get with conspiracy theories? Maybe I should just start making **** up (I clearly wouldn't be the first person here to do so) and claim that Windows on Arm is evidence Arm chips can only run 32bit x86 binaries, so macOS Big Sur/etc will obviously have the same limitation, so now none of the Catalina (i.e. 64bit required) apps will work.

 

[cgsnipinva]

IOW: None. When processor technology was young these were considerations which could have benefit. However as processor technology has matured these benefits all but disappeared.

Take for instance instruction decoding. One of the ideas for RISC was to have an orthogonal instruction set which simplified the decoding logic. Today decoding logic is such a trivial part of a processor it's essentially a non-issue. A modern processor can fetch and decode CISC instructions just as fast as they can RISC instructions. This may have been an inherent benefit when chip real estate was limited. Today, not so much.

That is not necessarily true. RISC runs an instruction every clock cycle. CISC chips can have instructions that take 12-20 clock cycles. In fact - Intel has implemented microcode processors on their chips to break down complex instructions into their component instructions to increase efficiency. Additionally, the x86 architecture requires some of this processing overhead that tales space on the chip where it doesn't on the RISC. WIth Apple spending a decade on their custom implementation - they have probably squeezed out more performance.

The biggest issue with RISC is how it uses memory - which historically has been inefficient - but Apple's implementation of RISC/ARM seems to be efficient with memory. Given the power and efficiency and the increased commonality for software development - there appears to be more advantages than we are recognizing.

 

[solardudesf]

Just curious. What does this mean for the Mac Pro? Will they offer an ARM-based Mac with an A-Series chip and ECC? Not sure I would want to spend $10k+ on an Intel Mac Pro that already has its clock ticking (seems doubtful to me they will support it for 7-10 years as I usually expect).

 

[Gerdi]

What are these inherent advantages? Can I expect the same list as what was provided back in 1995?

Inherent advantages? Compared to x86-64, ARMv8:
1) ís Load/store architecture
2) has a fixed length instruction set
3) has more architectural GP registers
4) has a weakly ordered memory model
5) Does not need pre-decode into uops on a larger scale.

All of this enables you to create ARM based designs which will significantly outperform x86-64 based designs within the same power and area envelope.

 

[cgsnipinva]

It’s naive to think that Apple, now using an architecture that has an inherent advantage over x86, and now free to CHOOSE to use whatever fab is executing on all cylinders, can’t keep up with Intel which has to execute perfectly to even be competitive. Hell, if Intel gets its act together in the fabs, you don’t think they’d happily make Apple’s arm chips in order to soak up fab volume? Of course they would.

The fabless model is a huge part of apple’s advantage - look at what getting rid of fabs did for AMD once they jettisoned Global Foundries (their spinoff).

Good point - Apple has been able to focus its investments on design and engineering of their custom SoC which has yielded in a competitive ARM implementation that gives them an advantage. They probably have been working on the Mac Implementation for 2-3 years. Hence they have been able to complile MacOS and Mac Apps into an ARM version.

However, I wouldn't be surprised to see Apple invest in TSMC to have a closely aligned manufacturer to maintain supply of chips.

 

[ArtOfWarfare]

Apple Silicon is rumored to have performance 70-100% greater than Intel x86; why would the price come down?

Performance isn't something that most consumers care about anymore. They use an Office Suite and browse the web. They want a bigger screen and a full keyboard, otherwise the iPhone/iPad would be good enough.

Price is something consumers care about. If Apple's price on the CPU is reduced by, say, $300, by moving away from Intel, then they can pass on a price reduction of $200 to the customer while improving profits per sale.

 

[Krevnik]

I see, yes I agree the high end is unproven. I'm expecting at least 2-3 years more development to optimize performance at the high end but hopefully I'm wrong. Let's put it this way, I think the low end (MacBook Air etc.) will show more dramatic gains sooner than the high end (Mac Pro). Also, there's more volume at the low end and (even though the Mac Pro is very expensive) I assume also more profit, so they'll probably be more motivated to sink engineering resources into the low end. Also, low end machines probably need less radical designs and can share more with the iOS processors and still achieve good performance and battery life.

I agree.

Honestly, I’m interested in a Mac Mini or 16” MBP using ARM. The Mac Mini could probably get a dGPU again in its current power budget. The MBP could probably quiet down for CPU-heavy loads, and be easier to cool when running dGPU loads.

And those would honestly be competitive using A12-derived silicon. Die yields being the risk there as it grows in size to accommodate more cores and more I/O that these computers need compared to the iPad.

It’s just not obvious to me that you can take the A12Z and get something in the 32-core range meant to handle Workstation level loads without a lot of work. Apple has done a lot of good work with the A series, but past performance is never a guarantee of future results in the CPU space.

Apple has said they will transition the entire Mac lineup to custom silicon within two years. That includes Mac Pro. if they say that, it means they have a roadmap and are confident they can pull it off.

This is also the same company that released the 2013 Mac Pro and let it rot. In their own words it was because they misjudged the direction of the market. So there’s the optimist in me that agrees with you, but there’s also the skeptic that points in the direction of the Mac Pro’s history over the last 7 years.

As I said, I think it’s fair to be skeptical at this point about the high end. At least until Apple gives us something more substantial than a DTK using existing silicon, and silence about the roadmap.

 

[sirozha]

Apple has clearly stated that Rosetta will not support running virtual machines.

I really do not understand how Apple can jettison Windows for X86 from its new ARM-based Max platform. This will seriously reduce their sales in the enterprise, where Windows is still king. They will have to come with a solution for running Windows for X86 on the ARM architecture one way or another.
maybe their own built in hypervisor will handle the X86 émulation for Windows.
[automerge]1592925883[/automerge]

That would be totally stupid - because this way you would emulate the whole Windows code in addition to application code. When you run Windows on ARM - only the application code - not the Windows code - need to be emulated, which gives as much better experience.

Windows on ARM is dead. It has no applications, not even MS office.
[automerge]1592926195[/automerge]

They never said anything about running anything other than the A12Z, which is part of the selling point. Everything we saw yesterday was on A12Z.

This was on done on stage. Nice way to obfuscate, though. I’m talking about the X86 games emulated under Rosetta 2 in their secret lab. Nothing was mentioned about the silicon they ran that emulation on other than it was the Apple silicon. I’m 99%!sure it was A14X or whatever they are planning for the ARM Macs at the end of 2020.

 

[ssgbryan]

Yeah, don’t let proof get in the way of your opinions.

Here’s my PhD thesis, if that helps? https://www.ecse.rpi.edu/frisc/theses/MaierThesis/
[automerge]1592922013[/automerge]

Ok. 10+ isn’t as bad as 14++++++. (Is it? Is it like a b-flat is the same as an a-sharp? And shouldn’t intel use minuses instead? Like 14- ?)

Still won’t be competitive with TSMC of course.

I wouldn't be too sure about that - 2020 is going to suck for Intel, but Golden Cove will be a major jump.

 

[Gerdi]

Windows on ARM is dead. It has no applications, not even MS office.

Stop your ill informed nonsense. Windows ARM has a triple WoW layer, which allows it to run 32 and 64 bit ARM apps as well as x86 apps. In addition it has WSL2, which allows it to run Linux apps.
And of course i can just install Office right from the store.
The SQ1 in the Surface Pro X outperforms any other 7W Windows device including the freshly announced Lakemont chip from Intel - and that is with 3 year old Cortex A-76 cores.

 

[Wando64]

Windows on ARM is dead. It has no applications, not even MS office.

It is my understanding that Windows 10 ARM can run almost all standard windows applications including, of course, Office.
What would be the point of Microsoft crippling its own hardware line.

I think you are confusing it with Windows RT.

 

[falainber]

In 2 years TSMC will be on N5P, and on the cusp of N3.

And Intel will be on 14++++++.

And more cores is better than more threads per core. And AMD will have no fab advantage over Apple.

And Intel’s new architecture will be no better in terms of CPI than its current one.

I really doubt AMD or Intel can compete with Apple.
[automerge]1592915495[/automerge]


Oh yeah? When are they going to get rid of FireWire ports, SD card slots, USB-A ports, and floppy disks if you’re so smart?

I would expect better from you. You know perfectly well that "3" and "5" in the process name is just a label without much technical substance in it. That's not to say that TSMC is not ahead of Intel right now but the numbers don't tell the full story (not clear if they tell any story). The way the industry treats the process names right now, Intel can name their next process 2nm and jump ahead of TSMC.
[automerge]1592926751[/automerge]

It is my understanding that Windows 10 ARM can run almost all standard windows applications including, of course, Office.
What would be the point of Microsoft crippling its own hardware line.

I think you are confusing it with Windows RT.

You are correct in that Windows 10 ARM can do all of this but the problem is that nobody uses it. So, it might be dead in that sense (probably too early to draw the conclusions but so far the start was not very encouraging)

 

[cmaier]

IOW: None. When processor technology was young these were considerations which could have benefit. However as processor technology has matured these benefits all but disappeared.

Take for instance instruction decoding. One of the ideas for RISC was to have an orthogonal instruction set which simplified the decoding logic. Today decoding logic is such a trivial part of a processor it's essentially a non-issue. A modern processor can fetch and decode CISC instructions just as fast as they can RISC instructions. This may have been an inherent benefit when chip real estate was limited. Today, not so much.

How is decoding a trivial part of the processor? Have you designed an instruction decoder for x86-64? Because i assure you it took a fairly large team a year and a half to do it for every x86 chip I ever worked on, and these things are HUGE compared to ARM instruction decoders.

And chip real estate is ALWAYS limited. Each of those transistors has leakage current. Each of those transistors takes up space forcing you to place other blocks that want to talk to each other farther apart. Each of those transistors burns dynamic power for every instruction that is decoded. And that decoder has an entire state machine in it - essentially a little computer all by itself - to handle x86 decoding. One CISC instruction can break down into a dozen little simple instructions. You have to keep track that these all belong to the same CISC instruction so that when they issue out-of-order and a mispredicted branch occurs you know how to revert state back to a clean state. You have to have multiple pipeline stages just to cope with al that. None of that is necessary in ARM.

You are speaking without any actual experience building these things.

 

[bigtomato]

I think the first '24' inch arm computer from apple will be a giant iPad, hope they make it touch capable !

 

[Stephen.R]

Windows ARM has a triple WoW layer

Like, "wow, this is a piece of crap" ?

which allows it to run 32 and 64 bit ARM apps as well as x86 apps

Specifically, only 32bit x86 apps. So it can run native arm binaries, and 32bit x86 apps.

And of course i can just install Office right from the store.

Yes, x86 Office. In 10 years of Windows on Arm existing, they've apparently never considered it worthwhile giving their customers a native binary.

In addition it has WSL2, which allows it to run Linux apps.

It's just a VM running Linux, and it requires Hyper-V to be enabled, which means you can't run any other Virtualisation tools at the same time. So if you want to run a VMware instance and use WSL at the same time, you're SOL.

What would be the point of Microsoft crippling its own hardware line.

Given the aforementioned 10 years of Windows on Arm with zero native office binaries, I'm not sure whether to assume it's apathy or incompetence?

 

[darkslide29]

Can you elaborate on how this is different than intel powered macs?

Love MacOS, especially with handoff when I'm using my iPhone. But I bootcamp to Windows often for gaming, since Mac gaming support is so low. I spec'd out an i7 Mac mini and added 32gb of ram, with an eGPU to play high-end games without the need of having to build a separate PC, and I've been really pleased with how well it works. It's so nice to have this little box be able to do MacOS and my gaming.

I get that people will say that if my reason for disappointment is not being able to run Windows, then one can argue I'm not using MacOS for the right reasons. I also get that I am in a small minority. Still a bummer for my uses though.

 

[sirozha]

Stop your ill informed nonsense. Windows ARM has a triple WoW layer, which allows it to run 32 and 64 bit ARM apps as well as x86 apps. In addition it has WSL2, which allows it to run Linux apps.
And of course i can just install Office right from the store.
The SQ1 in the Surface Pro X outperforms any other 7W Windows device including the freshly announced Lakemont chip from Intel - and that is with 3 year old Cortex A-76 cores.

Thanks for for your rude comment. I was wrong. Windows for ARM runs 32-bit code written for the Intel architecture in emulation. No 64-bit Windows apps written for the Intel architecture are supported in emulation layer and the support is not expected until 2021. So, Windows for ARM will run as a VM and then it will emulate X86 within the VM to run Office. Great solution. The only way this would work decently is if Apple allowed Windows for ARM to be installed natively in Boot Camp.

This is turning into a complete disaster: Native ARM apps for MacOS, X86 macOS apps running on Rosetta 2 emulation layer, iOS apps running natively, iOS apps recompiled in Catalyst for macOS, virtualized Linux for ARM with native ARM apps, virtualized Windows for ARM with native Windows ARM apps, Windows apps written for the Intel architecture running in emulation layer in virtualized Windows for ARM.

What’s the name of that new macOS? Big Slur? Remember, it was coined by me here.