[RealWorldTech] Kanter tidbit on Apple CPU development

[Arachnotronic]

I came across this post from David Kanter on the RWT forums and I thought it was interesting.

OK, so you are still wrong on this point IPC != performance. Scaling up perf isn't free. It can be done, obviously (since Intel has done so, and they hardly have a monopoly on good design). But there's no guarantee that Apple can get to high frequencies without losing some IPC.

I know several CPU designers at Apple, and they will all tell you that they aren't targeting anything close to Intel performance. Their cores simply cannot scale to 4GHz. They also have no way to use 10A/core, which Intel certainly can. In fact, I bet Intel can probably draw around 15A/core for some workloads.

That is directly from the people who design the CPUs. If you think you know more than they do, let me know.

 

[Glo.]

Thanks very much for this, Arach.

 

[deasd]

I believe there's no solution that operating ARM core closing to 4Ghz, at least until yet.
But I wonder how will be Apple A10X clocked, maybe close enough to 3Ghz.

 

[Arachnotronic]

I believe there's no solution that operating ARM core closing to 4Ghz, at least until yet.
But I wonder how will be Apple A10X clocked, maybe close enough to 3Ghz.

My guess is ~2.8GHz.

 

[Exist50]

Don't exactly see how this isn't anything that wasn't eminently obvious before. Of course Apple isn't trying to make a ~30-165W chip and/or some 20 core server beast. That's not their market. What they are trying to do is make competitive chips in the approximately 0-10W power envelope, and in that range, they do compete with Intel in performance.

 

[witeken]

I came across this post from David Kanter on the RWT forums and I thought it was interesting.

Alright, so that conclusively proves Kaby Lake is superior in performance, but loses out in density. So much for the Intel has to fear Apple and Apple has done something extraordinary and Apple will abandon Intel. Similar but still different design philosophies. Just goes to show why Intel was so proud Skylake scaled from 4.5W to 90W.

But for sure Apple won't put Core in their phones or tablets, they will only want Intel's process technology.

(One can however wonder if the loss of 1-2Ghz is worth the poor density, certainly since most of Intel's profit comes from lower margin (IoT) or low frequency cores (many core Xeons and 15W U SKUs).)

 

[raghu78]

With TSMC 7nm Apple is going to be easily in the 3+ Ghz range. GF estimates A72 running at speeds of 3.5 Ghz or more for their 7nm node.

http://www.eetimes.com/document.asp?piddl_msgid=363476&doc_id=1330467&page_number=2

"An ARM Cortex-A72 core could run at more than 3.5 GHz in the process, the company estimates."

Frankly if Apple wanted they could design a core which scales beyond 4 Ghz. The question is do they want to put the effort to design CPUs for Mac lineup and unify the products under their custom ARM architecture.

 

[Arachnotronic]

With TSMC 7nm Apple is going to be easily in the 3+ Ghz range. GF estimates A72 running at speeds of 3.5 Ghz or more for their 7nm node.

http://www.eetimes.com/document.asp?piddl_msgid=363476&doc_id=1330467&page_number=2

"An ARM Cortex-A72 core could run at more than 3.5 GHz in the process, the company estimates."

Frankly if Apple wanted they could design a core which scales beyond 4 Ghz. The question is do they want to put the effort to design CPUs for Mac lineup and unify the products under their custom ARM architecture.

With enough money, time, and talent, any company can design anything that they want.

 

[raghu78]

With enough money, time, and talent, any company can design anything that they want.

Apple is the company which has no shortage of the above. I believe that eventually Apple will unify its platforms under their custom ARM architecture. Its not a matter of if but when.

 

[Glo.]

Apple is the company which has no shortage of the above. I believe that eventually Apple will unify its platforms under their custom ARM architecture. Its not a matter of if but when.

It is quite funny for me that this is being discussed. We, on Macrumors, have been discussing both possibilities last year: that Nvidia could come to Apple computers again - more than unlikely, and macOS Sierra gives us more and more evidence that they are out for good, and the possibility of using AX chips for powering their computers. Second thing is also unlikely, because of lack very specific parts of kernel extensions for running whole OS on the chip. For us, in current state the sight of Hurricane tidbits in Sierra looks like they will need this chip for encryption, and storing the TouchID fingerprint photo away from the whole system. System can contact with the CPU, and check for validation. Also, APFS technology is directly linked to this chip.

Both can work together, just it is not possible to run macOS currently. In future - who knows. But the chances at this point are floating around 0.1% ranges .

 

[dark zero]

Let's wait for AMD K12 is they manage to get over 3 GHz at 14 nm.

 

[Arachnotronic]

Apple is the company which has no shortage of the above. I believe that eventually Apple will unify its platforms under their custom ARM architecture. Its not a matter of if but when.

Only in your opinion.

 

[Exist50]

Let's wait for AMD K12 is they manage to get over 3 GHz at 14 nm.

I have to wonder if K12 is doing well. They haven't been talking about it at all since the 2015 FAD, and that's certainly worrying for what is supposed to be a pretty core product and a sister to Zen.

 

[Arachnotronic]

I have to wonder if K12 is doing well. They haven't been talking about it at all since the 2015 FAD, and that's certainly worrying for what is supposed to be a pretty core product and a sister to Zen.

K12 is in all likelihood dead.

 

[deasd]

Apple is the company which has no shortage of the above. I believe that eventually Apple will unify its platforms under their custom ARM architecture. Its not a matter of if but when.

I think the current 'platform' are likely be changed rather than unify. Things like Ipad Pro is very closed to Macbook now but we won't call Ipad a notebook platform, the only problem is the time, to wait customers accept it gradually.

 

[Arachnotronic]

Apple is the company which has no shortage of the above. I believe that eventually Apple will unify its platforms under their custom ARM architecture. Its not a matter of if but when.

Just 'cause you have a ton of money doesn't mean you have to spend it in an unwise fashion.

Intel CPUs work just fine for the Mac, I just don't see the value prop of Apple doing its own custom CPU for this relatively small part of Apple's biz. Other than, of course, to satisfy those who wish to see Intel lose business

 

[Exist50]

K12 is in all likelihood dead.

I would hope it at least lives on in the embedded market or something. Would be an awful waste considering the implication that at one point they were putting as much into it as Zen.

Edit: But more on topic, I could see Apple releasing an ARM Macbook (non Pro). The power envelope is right, and with Apple's push to make it purely a consumer device, the compatibility aspect is less of a hurdle. Moreover, it would of course cut Intel's margins from the balance sheet, and would make the custom SoC design costs more palatable.

 

[scannall]

While I do think they are exploring the idea, it may be to push Intel. Intel has been failing at delivery dates for a while for the parts Apple wants to buy. And the next upgrade to the 'Iris Pro' series has been put off to the 1st quarter of 2018. That can't be making the people in Cupertino very happy.

 

[Arachnotronic]

While I do think they are exploring the idea, it may be to push Intel. Intel has been failing at delivery dates for a while for the parts Apple wants to buy. And the next upgrade to the 'Iris Pro' series has been put off to the 1st quarter of 2018. That can't be making the people in Cupertino very happy.

Iris Pro is no more.

 

[deasd]

Intel CPUs work just fine for the Mac, I just don't see the value prop of Apple doing its own custom CPU for this relatively small part of Apple's biz. Other than, of course, to satisfy those who wish to see Intel lose business

Sorry but, you sounds/emotion a bit violent (or distinctive) here......

The problem for Intel/x86 now is when scaling down the architecture and power consumption there's no advantage when against other competitors, while seeing these competitors are scaling up their design(multicore/high clock) to fit more computing and maintaining cost/price acceptable.

We should realize, nowaday desktop (or high-energy consumption) processors are developing painfully slower than before, every generation uplift is getting smaller and miserly. I won't be surprise if the mobile/smallish design eat up the whole share of desktop except high-performance setup in the future.

 

[VirtualLarry]

Intel seems to be doing fine in server. I wouldn't go so far as to say that they don't seem to be making performance or perf/watt improvements, they are. Just in server / HEDT, not in mobile / mainstream desktop.

 

[dark zero]

K12 is in all likelihood dead.

Hardly... the supposed K12 was Seattle. Also Keller made K12 along Zen. So both will come together.

 

[Arachnotronic]

Intel seems to be doing fine in server. I wouldn't go so far as to say that they don't seem to be making performance or perf/watt improvements, they are. Just in server / HEDT, not in mobile / mainstream desktop.

Look at the performance/power figures for IVB-Y, HSW-Y, BDW-Y, SKL-Y, and KBL-Y...

 

[cdimauro]

With TSMC 7nm Apple is going to be easily in the 3+ Ghz range. GF estimates A72 running at speeds of 3.5 Ghz or more for their 7nm node.

http://www.eetimes.com/document.asp?piddl_msgid=363476&doc_id=1330467&page_number=2

"An ARM Cortex-A72 core could run at more than 3.5 GHz in the process, the company estimates."

Frankly if Apple wanted they could design a core which scales beyond 4 Ghz. The question is do they want to put the effort to design CPUs for Mac lineup and unify the products under their custom ARM architecture.

You cannot compare an A72 core with an Apple one: the latter is way more complex.

 

[ShintaiDK]

K12 is in all likelihood dead.

Yep, its dead.