Apple rumored to be working on its own non-mobile ARM processor and ditching Intel around 2020

[Nicnac]

What they really should make would be a new macbook air. below 1k would be nice, too. Something that students can get without breaking the bank. 

[LAwLz]

On 10/26/2018 at 11:20 PM, Curufinwe_wins said:

And yes. Without a shadow of a doubt jumping from PPC to x86-64 is dramatically easier than from x86-64 to ARM. Esp since the emulation layers that were occasionally used in the interim back in the day... well they were the best people could come up with and they pretty much sucked. 

Why do you believe that?

 

On 10/26/2018 at 11:20 PM, Curufinwe_wins said:

But more than anything, I think the biggest deal is that pure processing power is a dramatic reduction with design basis processor swaps at the moment, which was unequivocally the opposite 'back in the day'.

Do I understand you correctly. You believe the transistor from x86 to ARM would be more difficult because x86 processors are much more powerful. But PPC to x86 was easier because the x86 processors were more powerful. Correct?

I got some news for you. Apple has already demonstrated that they are capable of producing ARM chips which are more powerful than some of the Intel x86 chips they use.

I think you, and a few other people in this thread, are

1) Dramatically overestimating how good x86 is, especially in the low-power packages like what's found in the MacBook (runs at below 1.5GHz)

2) Dramatically underestimating how good Apple's ARM cores are in a lot of aspects. They are already matching Intel's high-end chips in terms of IPC for some workloads.

3) Underestimating how much extra headroom Apple has. When you think of ARM performance you look at the iPhone and iPad for performance references. Apple could easily make a twice as powerful ARM chips for their laptops. They are not limited to the same chips as they use in their phones or tablets.

 

 

 

On 10/27/2018 at 5:25 AM, warrenr said:

Ehhh, that's not really true.  Windows has been shipping for multiple architectures since the release of NT 3.1, 25+ years ago.  MIPS, Itanium, ARM, Alpha AXP, PowerPC... they've kinda done it all.  Plus, the original Xbox and the 360 are PowerPC systems that run a customized OS that is architecturally very similar to Windows.  Even today, you can get a free version of Windows 10 that runs on ARM devices like the Raspberry Pi.  Microsoft may not be building their own ARM CPUs like Apple is, but they've got the cross-platform story pretty much 100% figured out already.

Completely different scenarios. Microsoft has made OSes which runs on different architectures, but they have so far only made one OS which runs on different architecture and maintains software compatibility (Windows 10 which runs on ARM and x86).

Microsoft never bothered with program compatibility before when they changed to different architectures. Apple did.

 

 

On 10/27/2018 at 5:25 AM, warrenr said:

Also, POSIX isn't really relevant here.  It's an application programming interface... it presents no opinions about how the hardware should work.

It doesn't matter for actually porting the OS over from one platform to another, but it does matter a great deal for retaining compatibility with existing software, which is the main issue.

You seem to think that making the OS run on a different architecture is the challenge here. It isn't even close to being the main challenge. The main challenge is retaining compatibility with all the existing software.

 

On 10/27/2018 at 5:25 AM, warrenr said:

There's going to be a bunch of stuff missing, like optimized graphics drivers.  What, you think Intel is going to be super-cooperative about making Intel Iris work great with ARM? 

Why do you think Apple would use Intel GPUs in this hypothetical ARM variant of their laptop? Apple are developing their own GPUs and could just use that.

They already force people into using their own Metal APIs, which their own GPU architecture already support. GPU compatibility is probably the least of their worries if they were to port MacOS to ARM.

[Curufinwe_wins]

8 hours ago, LAwLz said:

Why do you believe that?

 

Do I understand you correctly. You believe the transistor from x86 to ARM would be more difficult because x86 processors are much more powerful. But PPC to x86 was easier because the x86 processors were more powerful. Correct?

I got some news for you. Apple has already demonstrated that they are capable of producing ARM chips which are more powerful than some of the Intel x86 chips they use.

I think you, and a few other people in this thread, are

1) Dramatically overestimating how good x86 is, especially in the low-power packages like what's found in the MacBook (runs at below 1.5GHz)

2) Dramatically underestimating how good Apple's ARM cores are in a lot of aspects. They are already matching Intel's high-end chips in terms of IPC for some workloads.

3) Underestimating how much extra headroom Apple has. When you think of ARM performance you look at the iPhone and iPad for performance references. Apple could easily make a twice as powerful ARM chips for their laptops. They are not limited to the same chips as they use in their phones or tablets.

 

 

 

Completely different scenarios. Microsoft has made OSes which runs on different architectures, but they have so far only made one OS which runs on different architecture and maintains software compatibility (Windows 10 which runs on ARM and x86).

Microsoft never bothered with program compatibility before when they changed to different architectures. Apple did.

 

 

It doesn't matter for actually porting the OS over from one platform to another, but it does matter a great deal for retaining compatibility with existing software, which is the main issue.

You seem to think that making the OS run on a different architecture is the challenge here. It isn't even close to being the main challenge. The main challenge is retaining compatibility with all the existing software.

 

Why do you think Apple would use Intel GPUs in this hypothetical ARM variant of their laptop? Apple are developing their own GPUs and could just use that.

They already force people into using their own Metal APIs, which their own GPU architecture already support. GPU compatibility is probably the least of their worries if they were to port MacOS to ARM.

We've seen rather poor ARM performance scaling from the 3W  sustained phones use to the 15-35W design power even moderate laptops are designed for. So yes. I also think people vastly underestimate how much of a difference things like AVX make in software that has been optimized for it (or the dramatic total performance and feature improvements between mobile and intel/amd/nvidia gpu technologies, though that at least maybe they could turnkey from AMD).

 

But actually no, what I was saying was more that as processing power was increasing at a sufficient rate combined with an overall improvement in uArch (with respect to software investment/experience) with the change, the inevitable growing pains of such a shift were dramatically obfuscated by large general preformance uplifts.

[LAwLz]

1 hour ago, Curufinwe_wins said:

We've seen rather poor ARM performance scaling from the 3W  sustained phones use to the 15-35W design power even moderate laptops are designed for.

Who has 15-35 watt ARM designs out?

The Snapdragon 845 is sitting at sub 5 watts during heavy use, and I can't imagine that the 850 which is currently being used in Windows on ARM devices being that much higher (considering it's just a ~100MHz overclock).

 

For what I know, the rumored Snapdragon 1000 will be Qualcomm's first true laptop chip and it is reportedly 20x15mm in size (compared to 12.4x12.4 for the 845) with a 12 watt TDP.

So the package is twice as big, and the power consumption is almost 3 times as high.

 

I would be surprised if we don't see an 3x performance increase compared to the Snapdragon 850 if the rumors are accurate.

But on top of that, Apple currently has around a 3x performance per watt lead over Snapdragon chips. So for rough estimates of how powerful a ~12 watt SoC from Apple could be, take a look at the Snapdragon 850 and then take that performance times 6.

 

 

1 hour ago, Curufinwe_wins said:

I also think people vastly underestimate how much of a difference things like AVX make in software that has been optimized for it (or the dramatic total performance and feature improvements between mobile and intel/amd/nvidia gpu technologies, though that at least maybe they could turnkey from AMD). 

Not sure what you're trying to say here. Yes, SIMD instructions can make a big difference in performance. That's why ARM already has their own instructions for it, like NEON and SVE.

 

 

1 hour ago, Curufinwe_wins said:

But actually no, what I was saying was more that as processing power was increasing at a sufficient rate combined with an overall improvement in uArch (with respect to software investment/experience) with the change, the inevitable growing pains of such a shift were dramatically obfuscated by large general preformance uplifts.

So your argument is that PPC to x86 was easy because the processors were getting better so quickly that the growing pains disappeared very quickly thanks to much better hardware getting released every year?

We see that with ARM today, and that's with very limiting thermal and power constraints which would not be anywhere near as big of an issue in laptops.

[Belgarathian]

With Apple releasing the A12X today, I'd say that it shows the power behind Apple's capabilities.

 

Looking forward to benchmarks since by the looks of things it adds another Vortex core, and doubles the GPU.