Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Apple: No Macintosh Forks. But the iPad...

mondaynote.com

Post reply on HN
Print view View original

161–166 of 166 posts

Re: Apple: No Macintosh Forks. But the iPad...

#161

Post by eecc »

GordonS wrote: post #87
eecc wrote: post #72

Frankly I’m surprised no one hasn’t built an external core/ram expansion docking station yet. With OS support for Numa and a broker to schedule tasks on Big/Little cores all the pieces are there. My guess is the halo effect around the cloud that put these devices back into the drawers; when you can sync files across two conventional independent devices, risk aversion wins. But people might become aware and weary of r…

> Frankly I’m surprised no one hasn’t built an external core/ram expansion docking station yet I imagine that the distance and all the connectivity required between the CPU and RAM sitting in an external docking station would destroy performance.

Well, the point is that current interconnects are reducing this latency significantly, people are buying GPUs in external boxes today

Re: Apple: No Macintosh Forks. But the iPad...

#162

Post by eecc »

yjftsjthsd-h wrote: post #137
eecc wrote: post #72

Frankly I’m surprised no one hasn’t built an external core/ram expansion docking station yet. With OS support for Numa and a broker to schedule tasks on Big/Little cores all the pieces are there. My guess is the halo effect around the cloud that put these devices back into the drawers; when you can sync files across two conventional independent devices, risk aversion wins. But people might become aware and weary of r…

Suddenly disconnecting a dock today is a minor annoyance; worst case, you lose some file descriptors/mountpoints, and network connections. If you plug in processors , accidentally unplugging outright drops processes, and likely panics the kernel.

Likely? Depends how the kernel mediates access to the detachable CPU facility.

Hot plug CPU has been around forever, as well as Apple's GCD.

Access to the pluggable CPU/RAM complex doesn't have to be transparent and irrecoverable on error.

Re: Apple: No Macintosh Forks. But the iPad...

#163

Post by SkyMarshal »

reitzensteinm wrote: post #46
SkyMarshal wrote: post #36

Earlier quoted context omitted.

I can't see Apple using any ARM chip other than their own. If they commit to converting OS X to ARM then they'll either produce it in-house or, less likely, buy AMD (or maybe just buy AMD's Zen tech only). One of the main reasons for Apple to move to ARM is to control more of their technology stack. It's the direction they've been moving in for a while now and this would be part of that strategy.

The article claims Apple would need to keep x86 around indefinitely for the Mac Pro - I'm just saying that makes no sense. Apple would definitely prefer to only use its own chips, but if they're unable to, using ARM Zen from AMD makes a million times more sense than x86 Intel. I think you may be underestimating Apple's willingness to let the Mac Pro be an ugly duckling. This is a platform that they left stagnant for…

>but if they're unable to, using ARM Zen from AMD makes a million times more sense than x86 Intel.

How do you figure? Transitioning from x86 to ARM would be a major, expensive undertaking. If it's justified at all it would mainly be by Apple gaining control over that part of their stack. I doubt they would make such an effort and investment only to switch from the current third party, Intel, to a different third party, AMD.

But otherwise yes they're really not likely to make the transition at all. Given how small a proportion of their overall sales Mac sales are, they're most likely to just leave it on x86 indefinitely.

Re: Apple: No Macintosh Forks. But the iPad...

#164

Post by ksec »

ken wrote: post #160
ksec wrote: post #148

Earlier quoted context omitted.

You will need to buy different set of tools to work and design for TSMC 's node. And while it is still from TSMC, the node for Desktop ( High Performance ) will be different to Phone / Tablet ( Low Power ). The cost of designing and testing a 5nm chips is roughly a billion. For iPhone that is spread out over 200M unit over its life time, or roughly $5 per chip excluding the actual cost of wafer. For Mac you are looki…

> the node for Desktop ( High Performance ) will be different to Phone / Tablet ( Low Power ) The iPhone CPUs are already competitive with Intel chips. They've got incredible single-core performance, great memory bandwidth, and solid 64-bit and SIMD support. What specific feature of a "high performance" node would they require that they aren't using for iPhones now? They've been pushing the limits on allegedly "mobil…

>The iPhone CPUs are already competitive with Intel chips.

In selected benchmarks. That is like saying the Qualcomm ARM Sever Chips are already competitive with Intel Chips. But it is pretty close, at least to the point for most consumer usage it doesn't matter.

>The iPhone CPUs are already competitive with Intel chips. >What specific feature of a "high performance" node

At sub 10W ( The TDP for iPhone and iPad Pro ), not at 15W+ or 250W on the Mac Pro. You will require different tools, nodes, design guidelines operating at those TDP. No one is saying Apple cant do it, the question is if it make any financial sense doing it.

>AFAICT, nobody in the world has a production 5nm chip yet. How can you put a price on something that nobody has successfully produced yet?

https://semiengineering.com/5nm-vs-3nm/

The lead time for designing a chip / SoC with new node is roughly 2-3 years. Apple are already working ( finalising ) on 5nm SoC for next year, and they already knew the basic costing, ( excluding yield issues ). 3nm Costing Projection are also already known and are being worked on as we speak.

>If I were Apple and I were confident that just throwing $1B at it would solve all the engineering problems, I would absolutely do it. $40 per Mac for the best CPU (and independence from Intel) sounds like a steal.

Quoting myself excluding the actual cost of wafer. and yield. Apple would have made the jump if it really only cost them $40.

Re: Apple: No Macintosh Forks. But the iPad...

#165

Post by reitzensteinm »

SkyMarshal wrote: post #163
reitzensteinm wrote: post #46

Earlier quoted context omitted.

The article claims Apple would need to keep x86 around indefinitely for the Mac Pro - I'm just saying that makes no sense. Apple would definitely prefer to only use its own chips, but if they're unable to, using ARM Zen from AMD makes a million times more sense than x86 Intel. I think you may be underestimating Apple's willingness to let the Mac Pro be an ugly duckling. This is a platform that they left stagnant for…

>but if they're unable to, using ARM Zen from AMD makes a million times more sense than x86 Intel. How do you figure? Transitioning from x86 to ARM would be a major, expensive undertaking. If it's justified at all it would mainly be by Apple gaining control over that part of their stack. I doubt they would make such an effort and investment only to switch from the current third party, Intel, to a different third part…

I'm speaking specifically about the Mac Pro, which Apple doesn't really care about. I get the sense you think I'm talking about using Zen in their laptops etc.

They're not going to build a chip that competes with high end Xeons just to field the new Mac Pro.

Given that, in a world where MacBook Pros are powered by A15s, Apple can either cut the Mac Pro, keep it as their only x86 product or put an ARM Zen in it.

I'm saying keeping it as x86 will straight up never happen. ARM Zen or cancellation are the plausible options.

Re: Apple: No Macintosh Forks. But the iPad...

#166

Post by jart »

gridlockd wrote: post #108
jart wrote: post #19

Earlier quoted context omitted.

Is the struggle for microprocessor supremacy still even relevant? I'd ask, "who's going to build the best quantum processor?" rather than, "who's going to dethrone x86?"

> Is the struggle for microprocessor supremacy still even relevant? Absolutely. Today, it's not about performance, but about performance/watt. Simply using a web browser is a huge power drain. Even if that wasn't the case, getting rid of the "Intel dependency" and the "x86 premium" would make sense from a business standpoint.

x86 is a free/open isa. The patents are expired up to i486 and x86_64 patents should expire soon. A guy from Oracle not too long ago implement i186 in Verilog as a hobby project. At least we know that if Intel ever gets too authoritarian, we can do that for i386.
Post reply on HN