On Apple Announcing the ARM Mac Transition at WWDC This Month(daringfireball.net)
daringfireball.net
On Apple Announcing the ARM Mac Transition at WWDC This Month
https://daringfireball.net/2020/06/on_apple_announcing_the_mac_arm_transition_at_wwdc
21 comments
I'm excited about this. Can't wait to see what performance looks like. As for the effect on sales, it's probably a wash. For people who know and care about these things, I'd imagine a 50/50 split between people racing to get the last rev of Intel macs to wait out the transition until things stabilize and the other half waiting for the new stuff. macOS app devs won't really have a choice, but iOS devs will.
I believe this will enable new form factors given PPW and higher performant systems free of power and thermal constraints on desktop. It may take some time, but I think this will lead to significantly greater long term sales and market share. Apples lead is rather significant in this space and the only computer with that soc will be a Mac.
I really believe that the lack of x86 is going to be a big buying problem for a lot of people who need to run software on Windows. Without Boot Camp or x86 virtual machines its going to be a problem to justify the machine with a large group of people.
I suspect that for every customer they lose for someone needing to boot natively they will gain more than one from performance, cost and form factor flexibility enabling much more desirable machines. Batteries add both size and weight so you can get thinner and lighter and still have a functional keyboard.
Why do people believe Apple will price these machines cheaper than their Intel offerings?
Depends on what you're trying to do. x86 emulation is probably more than fast enough for a lot of types of software. And since Apple is designing the CPU, presumably they could add instructions to accelerate emulation if they wanted to.
I wonder if the image, video and audio processing software vendors will make of this - I would assume a lot of optimizations are tied to the underlying hardware and written in assembly?
This seems to be a golden opportunity for Adobe pay back in some way what Steve Jobs did to them by not supporting flash on iPhone
This seems to be a golden opportunity for Adobe pay back in some way what Steve Jobs did to them by not supporting flash on iPhone
For vendors using metal it could be seamless. There would also be gains had through ML and neural engine in image, video and audio processing. Not to mention dedicated video hardware decode. Time will tell, but I don’t see adobe being punitive there. They make money on software subscriptions and are already bringing native photoshop to iOS.
Why not just require apps to be distributed as LLVM bitcode, as is done for the Apple watch? That would give Apple the flexibility to change architectures--or even to create its own instruction set, now that it's making its own chips.
LLVM bitcode is not that portable in reality. Just like Plan9 assembly intermediary bitcode files are pretty much single architecture, due to containing lots and lots of low-level assumptions about the target architecture, not to mention possible incompatibilities in extended intrinsics (for example for vector units).
Most of the portability problems for retargetting to another little endian 64-bit architecture aren't the easy ones of bit widths and integer types, but rather more subtle things like a more relaxed memory model and different performance characteristics.
It's not worth the overhead of attempting to ship bitcode. Apple has shipped products on 6502, 68000, PPC, x86, x86-64, ARM, and ARM64. The transitions have been annoying for the lost software, but that's not an unusual occurrence for an OSX upgrade either.
It's not worth the overhead of attempting to ship bitcode. Apple has shipped products on 6502, 68000, PPC, x86, x86-64, ARM, and ARM64. The transitions have been annoying for the lost software, but that's not an unusual occurrence for an OSX upgrade either.
I sincerely doubt this will show up on anything but entry level products until developers have time to migrate their apps.
In the past, there has always been a period where the current and future architectures were both sold.
The interesting thing going forward is the question of working with Microsoft to get Windows on ARM ported over (assuming Microsoft finally gets the promised x64 app compatibility working) or to fork something like QEMU to allow for x86 virtual machines.
Or both, I guess.
In the past, there has always been a period where the current and future architectures were both sold.
The interesting thing going forward is the question of working with Microsoft to get Windows on ARM ported over (assuming Microsoft finally gets the promised x64 app compatibility working) or to fork something like QEMU to allow for x86 virtual machines.
Or both, I guess.
Can they source enough chips to do this in the near term? Meanwhile, another major fabless designer (AMD) is ramping up too.
Apple ships close to 200 million iPhones a year all with their custom ARM CPUs, fabricated by TSMC. They ship about a tenth as many Macs. I'm pretty sure they can handle that growth, especially as one of TSMC's biggest customers.
And the move to 5nm for the A14 (sampling since 2019) will yield more chips per wafer. Either way, Apple is very forward looking on supply chain and has been planning this for a decade. Remember when they owned the majority of the nand market due to long term supplier contracts?
Interesting thought. Will the functional yield really be much higher? Because there could be more wasted chips?
Smart move...
How would you otherwise convince existing customers to throw away working systems and buy new ones?
I don't think that we should expect new ARM64 to be 10 times better than x64...
The only idea I can think of is if chip will have bult-in hardware-accelerated OpenVG or any other 2D API to replace severely outdated CoreGraphics ...
How would you otherwise convince existing customers to throw away working systems and buy new ones?
I don't think that we should expect new ARM64 to be 10 times better than x64...
The only idea I can think of is if chip will have bult-in hardware-accelerated OpenVG or any other 2D API to replace severely outdated CoreGraphics ...