Linked by David Adams on Sun 8th May 2011 04:47 UTC
Apple The Apple/ARM rumor du jour is that Apple will transition its entire portable Mac line to ARM-based CPUs, dropping Intel altogether. Sources speaking to Semi Accurate claim this is a "done deal," and the move should happen by 2013, when a 64-bit ARM A15 core becomes available. While a future generation of Apple's A5 processor could make some sense for something akin to the MacBook Air, the claim that Apple will ditch Intel wholesale for ARM just doesn't add up.
Permalink for comment 472318
To read all comments associated with this story, please click here.
Unsurprising
by torbenm on Mon 9th May 2011 09:54 UTC
torbenm
Member since:
2007-04-23

One of the reasons Apple stated for changing from PPC to x86 was power use, so if they still find this important, ARM is a logical choice for a new transition, now that they are powerful enough to power laptops and servers. Also, Apple has greater experience than most in changing processors: They started the Mac series with 68K, then switched to PPC and later to x86. They have now used x86 longer than either 68K or PPC, so you could argue that they are overdue by now.

But the main point about this is that Apple have become increasingly careful not to rely too much on a specific processor architecture, so the cost of changing is far lower for Apple than it is for Microsoft, who nevertheless will make Win8 run on ARM.

These days, most of the heavy loads on a computer are run on graphics processors and other coprocessors, so the performance of the main CPU is less than it used to be. It is far more important that you can get the coprocessors you want. Here ARM has an obvious advantage over x86: It is fairly easy to get an ARM license to produce an ARM core surrounded by whatever coprocessors you like, an option Apple took for its A4 chip. With x86, you have to take whatever SoCs Intel, AMD and a few other make.

As for running "old" code, Apple showed with its previous transitions that emulation isn't a problem: Most time is spent in library routines, which can run natively. And with JIT, emulation of the rest can be done with relatively low overhead (around x2). ARM and x86 share the use of arithmetic flags, so emulation of flags, which is a problem when emulating x86 on, say, PPC or MIPS is almost a non-issue when emulating on ARM. In the future Apple will ship more and more code in the LLVM format, which is compiled at load time to native code, so they can ship the same code to both x86 and ARM-based computers.

As other's have mentioned, using a SoC of their own design and under their complete control, Apple can prevent porting of future versions of MacOS to non-Apple machines. As Apple makes its money from hardware sales (at least with the Mac line), this is something they are eager to do.

The main problem with ARM so far has been performance and support of more than 4MB of memory. The A15 core solves both of these problems, even if it isn't a true 64-bit core. But 64 bits is mainly an issue for memory -- the performance gain is modest. For x86, the main performance gain from 64-bit processors is due to a different FP instruction set; the integer performance isn't that much greater.

Reply Score: 0