[Bloomberg] Apple starting process to dump Intel in Macs

[ksec]

Geekbench is an indication of performance. And you can read what was being tested in the individual test. So theoretically speaking if you have the application using the same optimization in both ISA, they should be very similar to the results show in Geekbench. Especially when LLVM is a major pieces that many vendors are working on. But there will be Apps that may be better suited to Intel's Core, not because of x86 ISA but how Intel optimize its uArch. Just the same as ARM64 or Apple's ARM implementation.

Anandtech use Javascript as a benchmarks, and JS benchmarks, both on macOS and iOS safari are being heavily optimized by Apple. That is another indication of performance.

Apple last said 20% of their Mac sales are Desktop, that is everything from iMac to Mac Pro. ( Which is actually higher then a lot of people expected. ) To put this in real number, that is roughly 5M unit / year.
If we assume the normal price of $200M development cost of 7nm HP from TSMC, that is roughly $40 per unit.
I doubt Apple pay anywhere near that number. And there should be a lot of synergy (cost) between 7nm LP and HP, but the numbers are too small for Apple to spend time to invest into. Remember these 5M CPU unit spends from 45W to 100W+, different number of cores, ECC memory support etc. It would make much more sense to use something like Qualcomm Centriq, which has a roadmap to decent single Core performance by 2020.

There was another rumours that Intel will have a new x86 out in 2019/2020. A cleaned up x86 that does not provide 100% backward compatibility.

 

[IRobot23]

AMD

 

[Dolan]

If i can correct this $200M statement... it definitely don't cost so much. Well, it depends...

You can tape-out leading-edge silicon for less than 1M USD. Bigger part of costs will be IP fees, design tools/support and pay for engineers you employ (in listed order). Only salary for team of engineers working on bigger project might go into billions (over years).

Point here is, that Apple already has big part of chain in house. So price for releasing additional chip will be negligible. This is reason, why there is no problem for Intel to release so many different chips. Same reason why even AMD is able to design lots of ASICs in 3 leading edge foundries.

This was regarding price question. After they get into new technology, maxing next chips becomes really cheap.

Regarding performance.: Key here will be TSMC N5 node. They promises great improvement in performance (4 GHz ARM was already demonstrated on 10nm ...under ideal conditions off course).

Plus there will be great transistor budget (200M per mm2). This will allow to spend tens of bilions of transistors on dedicated accelerators and if Apple will be able to integrate it into software ecosystem then it might achieve performance no conventional CPU can match.

Machine with powerful and working rendering or encoding accelerator might be interesting for some people.

 

[beginner99]

Why do you necessarily have to split design? Use the same (or similar) chips and just turn up the clock speed and tdp for laptops.

Macs go from dual core laptops to 18-core workstations. That's why. Unless you just can the upper end of the lineup. that of course is also a possibility. But my main point being Apple will not be able to provide the same lineup for the same price if they ditch intel. Either it will cost a lot more or the lineup will be a lot smaller. It's either way not really a winning proposition to ditch intel.

 

[beginner99]

You said :
"So I would say to cover this range apple would need to make at least 3 designs or completely abandon certain markets"

This is such a strange statement when we have just had the perfect lesson in how to do this with the Ryzen launch. It seems as if not everyone is paying attention.

I thought about that too and mentioned it in another post. Ryzen is just a CPU. Apple would need to also scale the GPU somehow or provide a discrete apple made GPU as AMD doesn't offer Radeon drivers for ARM.

 

[jpiniero]

Macs go from dual core laptops to 18-core workstations. That's why. Unless you just can the upper end of the lineup. that of course is also a possibility. But my main point being Apple will not be able to provide the same lineup for the same price if they ditch intel. Either it will cost a lot more or the lineup will be a lot smaller. It's either way not really a winning proposition to ditch intel.

Yeah but iMac Pro/Mac Pro is only a tiny percentage of desktop sales. The regular iMac uses standard desktop processors.

If they have similar technology in the pipeline I could see Apple using something similar to EMIB to be able to connect multiple CPU tiles, and possibly GPU as well.

 

[moinmoin]

LOL Is Apple a Chinese company yet or are they US and just making all their products in China?

They are a Californian company (as they keep putting "Designed by Apple in California. Assembled in (...)" on their products).
Apple is increasing its margins by externalizing the risks inherent to the manufacturing business.

 

[PeterScott]

If i can correct this $200M statement... it definitely don't cost so much. Well, it depends...

You can tape-out leading-edge silicon for less than 1M USD. Bigger part of costs will be IP fees, design tools/support and pay for engineers you employ (in listed order). Only salary for team of engineers working on bigger project might go into billions (over years).

Point here is, that Apple already has big part of chain in house. So price for releasing additional chip will be negligible. This is reason, why there is no problem for Intel to release so many different chips. Same reason why even AMD is able to design lots of ASICs in 3 leading edge foundries.

This was regarding price question. After they get into new technology, maxing next chips becomes really cheap.

Regarding performance.: Key here will be TSMC N5 node. They promises great improvement in performance (4 GHz ARM was already demonstrated on 10nm ...under ideal conditions off course).

Plus there will be great transistor budget (200M per mm2). This will allow to spend tens of bilions of transistors on dedicated accelerators and if Apple will be able to integrate it into software ecosystem then it might achieve performance no conventional CPU can match.

Machine with powerful and working rendering or encoding accelerator might be interesting for some people.

Constructing the Masks is the expensive part. Each step down in size requires more masking layers to be created.

https://semiengineering.com/racing-to-107nm/
"For those who migrate beyond 16nm/14nm, it will require deep pockets. In total, it will cost $271 million to design a 7nm chip, according to Gartner. In comparison, it costs around $80 million to design a 16nm/14nm chip and $30 million for a 28nm planar device, the research firm said."

This is why everyone tries to make the fewest designs possible, and use them as long as possible, to amortize the obscene up front costs, which more than double with each node shrink.

 

[Dolan]

Its just estimate and key part is to design. As i said, most expensive part of it is pay team of engineers. Tape-out itself is relatively cheap.

Most of companies has more variants in every series (varying core count, gpu size...) I just got you one great example of company with very limited budget managing to release tons of different chips...

 

[PeterScott]

Its just estimate and key part is to design. As i said, most expensive part of it is pay team of engineers. Tape-out itself is relatively cheap.

Most of companies has more variants in every series (varying core count, gpu size...) I just got you one great example of company with very limited budget managing to release tons of different chips...

No, the big growing costs are the Masks, which is a Unique part of the work that has to be done for each chip , by the FAB, not Apple staff. So it is money paid to the FAB.

What example company?

 

[deathBOB]

Macs go from dual core laptops to 18-core workstations. That's why. Unless you just can the upper end of the lineup. that of course is also a possibility. But my main point being Apple will not be able to provide the same lineup for the same price if they ditch intel. Either it will cost a lot more or the lineup will be a lot smaller. It's either way not really a winning proposition to ditch intel.

Alternatively, Apple pays Intel a significant amount for chips across its current line. Cutting that cost from the majority of its Mac products could make it worthwhile to develop additional chips. And high end Macs have prices that could absorb expensive development costs.

I imagine Intel gets a big chunk of the margin on most Macs.

 

[beginner99]

What example company?

I assume AMD is meant with the Zeppelin die. But that is not really a SOC as it lacks GPU. Putting multiple SOCs with a GPU on an MCM and making it work and scale is a whole lot more complex than just CPUs or they then need to have a discrete GPU with drivers that work on ARM uArch. All in all it's not a very exciting prospect.

I only see it working if they can the mac pros. then they could use iPad SOC in lower tier notebooks and create 1 additional design for higher tier and segment the market by cutting that down and by clocks. But I fail to see what actual gain apple has here.

Alternatively, Apple pays Intel a significant amount for chips across its current line. Cutting that cost from the majority of its Mac products could make it worthwhile to develop additional chips. And high end Macs have prices that could absorb expensive development costs.

I think you completely fail to understand how expensive it is to develop a chip. With the volume of the Mac pro 18-core, a custom made chips just for that would probably lead it to cost half a million.

I imagine Intel gets a big chunk of the margin on most Macs.

imagining things is easy but mostly far away from the truth. Just compare a mac to a similar windows laptop (eg same intel chip) and then think again.

 

[ksec]

I imagine Intel gets a big chunk of the margin on most Macs.

It should be "Intel gets a big chunk of the margin on most Macs BOM". Which is true.

I think it is wrong to think Intel has big chunk of Macbook margin, since the MacBook margin belongs to Apple only.

 

[ksec]

No, the big growing costs are the Masks

Let's just say it is both design and mask. And lots of other IP, testing, QA etc I.e it is EXPENSIVE!

And that 200 millions actually get you two node, 10nm and 7nm. Along with 7nm+ etc. 7nm is a long node so it isnt too bad of an investment with enough volume.

 

[Zstream]

They are a Californian company (as they keep putting "Designed by Apple in California. Assembled in (...)" on their products).
Apple is increasing its margins by externalizing the risks inherent to the manufacturing business.

In other words... screwing American taxpayers

 

[scannall]

In other words... screwing American taxpayers

How so? Apple pays more taxes than just about everyone. Largest US Corporate Tax Bills

 

[PeterScott]

I assume AMD is meant with the Zeppelin die. But that is not really a SOC as it lacks GPU. Putting multiple SOCs with a GPU on an MCM and making it work and scale is a whole lot more complex than just CPUs or they then need to have a discrete GPU with drivers that work on ARM uArch. All in all it's not a very exciting prospect.

That's an example of a company reusing the same chip in multiple applications. He said a company with low budget that "release tons of different chips".

In other words some kind of evidence that the price building different chips was negligible (when available info says the opposite). Maybe he went off to look for the example.

 

[chrisjames61]

History. When they switched from 68000 series CPU's to Power PC, they stopped supporting the 68000's just a few years later. Then, when they switched from Power PC to Intel, they stopped supporting the Power PC processors in just a few Mac OS X point releases.

The first PPC Macs arrived in March 1994. It wasn't till late 1998 that OS 8.5 came out and 68K Macs were left behind. Snow Leopard had updates till late 2013. Not bad since the last PPC Macs were built seven years earlier.

 

[jpiniero]

The first PPC Macs arrived in March 1994. It wasn't till late 1998 that OS 8.5 came out and 68K Macs were left behind. Snow Leopard had updates till late 2013. Not bad since the last PPC Macs were built seven years earlier.

Leopard was the last OSX release to support PPC, not SL.

 

[IntelUser2000]

There was another rumours that Intel will have a new x86 out in 2019/2020. A cleaned up x86 that does not provide 100% backward compatibility.

They are working to eliminate BIOS support and move to UEFI. It also seems logical to eliminate 16/32-bit support. This isn't a rumor, its being worked on.

What is a rumor is a CPU that is supposedly being worked on that can decode ARM instructions to x86.

 

[PeterScott]

They are working to eliminate BIOS support and move to UEFI. It also seems logical to eliminate 16/32-bit support. This isn't a rumor, its being worked on.

You probably shouldn't lump those together. Intel has indicated they are removing BIOS support from UEFI by 2020. That is a fact. That is also likely just a chipset change, nothing to do with the CPU.

16/32 bit support is part of the CPU, and would be a major change that Intel has made no indication of at all. That would be entirely rumor based. The source of the Intel Streamlining x86 is this C&B story:
https://www.bitsandchips.it/english...-rumor-even-intel-is-studying-a-new-x86-uarch

Which has nothing that elevates it beyond the random nonsense rumor that goes by on a regular basis.

What is a rumor is a CPU that is supposedly being worked on that can decode ARM instructions to x86.

That is almost certainly a total BS rumor( Googled, can't find this one). I would bet ZERO chance of this one being true. It really makes no sense for Intel to build something like this.

 

[IntelUser2000]

That is almost certainly a total BS rumor( Googled, can't find this one). I would bet ZERO chance of this one being true. It really makes no sense for Intel to build something like this.

This is from the CPC article.

 

[PeterScott]

This is from the CPC article.

Is there a link?

 

[jpiniero]

^^

IIRC, the Sea of Cores patent talked about executing x86 instructions on non-x86 cores. Vice-versa would make sense as well. But I think this is an actual real ARM core either on die or off on it's own tile that the x86 core could utilize.

The specific thing I saw being removed was the MMX unit. Now, removing 16/32 bit support would make a ton of sense to Intel's server customers who don't want or need it. But if you break that, you may as well do a clean break. Maybe that's a part of it, that Intel made it obvious that the Lakes are just going to be milked with minor improvements at best and the Rapids won't be client-friendly...

 

[IntelUser2000]

Is there a link?

Yea sure.

http://temlib.org/pub/Canard/Canard_Intel.pdf

I'd still count this as a rumor.