Solved! ARM Apple High-End CPU - Intel replacement

[Richie Rich]

There is a first rumor about Intel replacement in Apple products:

• ARM based high-end CPU
• 8 cores, no SMT
• IPC +30% over Cortex A77
• desktop performance (Core i7/Ryzen R7) with much lower power consumption
• introduction with new gen MacBook Air in mid 2020 (considering also MacBook PRO and iMac)
• massive AI accelerator

Source Coreteks:

 

[Doug S]

Hence the debates and conflict here. Apple cores will be better for average user. But once you move into more complex stuff like compiling, heavy FP usage or server type usage like databases, then I Intel/AMD will show their advantages. But mist people don't do that on phones, tablets or laptops.

Go check out the A13's ST performance for clang in GB5 and gcc in SPEC2006, and compare with Intel & AMD's fastest CPUs, and tell us again how Intel and AMD will "show their advantages" for compiling. Obviously if you look at the multithread numbers for compiling Apple doesn't do anywhere near as well, because they haven't released anything with more than 4 big cores. Yet.

As for databases, in recent posts I mentioned IBM's POWER architecture and HP's PA-RISC architecture and noted they have huge low level caches (huge L2 in IBM's case in the POWER5/6 and huge L1 in HP's case) You wanna know why? Because they were primarily designed to run databases well, and that's a long-proven way of doing so because the non-locality of most DB access means anything you can do to reduce the average random memory latency will improve your performance when running a database. Not saying Apple's SoCs have been designed to run a database, or would do so as well as Intel's, and you need a lot of I/O capability as well which obviously a phone SoC lacks, but they'd probably do a pretty darn good job with an in-memory database.

FP is another matter, there's a limited role for FP in a phone SoC so those will never be designed to do it well. There is a role for it for in Macs, especially 'Pro' Macs, so I expect it will have more FP capability than the phone SoCs. Probably in the form of something like SVE2, but we'll have to wait and see what Apple does for the Mac SoCs - especially whatever they use in the Mac Pro.

 

[defferoo]

The release the mobile chips before desktop because efficiency matters more for mobile and mobile chips usually have less cores so higher yield. This is especially obvious with the 10nm fiasco.

Server would also profit from efficiency but these dies are much larger and usually yield isn't ideal early on on a new process. On top of that server CPUs simply need more validation than client CPUs.

No this is speculation but if we look at how graviton2 preforms in certain benchmakrs (see phoronix) one can really see the difference between a client first and server first architecture. (graviton2 has extremely slow compile time of linux kernel and database benches are also very slow). So stuff that relies on branch prediction, large caches in general complex stuff, that is were intel/amd shine. Maybe apple too, but that's to be seen.
(I mention compiling because many devs use macbooks and if that takes a large hit, the will probably move away from apple).

You’re right in saying that yields are the major factor in release timing, but that doesn’t change the fact that they design the core for low power usage first, then scale up to high power designs. This is the why the entire industry is focused on PPW, if you can nail that, you get great battery life on low power devices, and great performance on high power ones.

I just don’t agree with your blanket statement that ARM chips can’t run heavy server workloads. We know the Graviton2 is probably hobbled by the amount of cache, leading to issues in certain highly multi-threaded scenarios where each core only has 1.5MB . There are many benchmarks where it is on par or better than Xeon and EPYC competitors. Your idea that ARM ISA chips can’t do “complex stuff” like compiling and databases (which are honestly very routine things...) is just incorrect. There may not be an ARM server chip on the market that competes toe to toe with x86, but given the current trends, we’ll likely see it soon.

 

[blckgrffn]

For a variety of reasons this seems like an ideal time to divorce from Intel, this is likely the least appealing and noncompetitive the Intel lineup has ever been and while it might dip even further in the next 6-12 month before they seem primed to make a comeback. Anyway. If you are setting a baseline for walking away from Intel x86, this is a *great* time to establish it.

As has been pointed out over and over, a next gen A chip on a cutting edge process designed for a tablet is going to be hands down better than a 10nm laptop targeted chip from Intel that likely has twice the power envelope. Better yet, they aren't even putting 10nm chips in their Macbook Pro 16 and Mac Pro models.

That said, I was wondering how Apple will leverage this *to make more money per unit* above and beyond the obvious potential of not giving money to Intel.

My guess is that Apple decides that ~10 hours of battery life continues to the be their target for their Pro Macbooks.

A iPad Pro 12.9 rocks a ~37 Wh battery, the icelake based 13" Macbook Pro has a 58 Wh battery. Both are rated for ~10 hours of use. Is there a reason that an Apple Silicon powered 13" Macbook Pro would now need a battery larger than ~40 Wh for the same battery life?

That would be a quantifiable BoM savings and I would think allow for even thinner/lighter models. I would think Apple could easily lean on better performance, thinner & lighter & cooler running chassis as their marketing pillars and then on the back end enjoy better margins in this case.

A Macbook Pro 16 is even more egregious, a 100wh battery for it's 10 hours of use. Let's say we get some form of the Apple Silicon that uses more more power but the savings potential here is huge - we could see a 60-70 Wh battery for the same 10 hours of usage in the larger chassis? What a win.

Comparing how Apple has used it's vertical integration capabilities in the mobile space (largely using smaller batteries compared to Android competition but getting similar/better battery life), I would expect rather than selling laptops with ~15-20 hours of battery life that are the same form factor as they are currently they'll go for bigger margins. I realize they already "dongle books" and it might be hard to make them thinner - but lighter surely?

I admit there might be something really critical I am missing here in comparing iPad Pro and Macbook Pro battery capacities here that makes the math fall apart. If the math holds, I'd find this battery capacity savings without performance regression as compelling as any potential performance increase the new CPUs bring.

 

[defferoo]

hopefully they find a good balance between performance, battery life, and size. honestly the current size of their notebooks are already great, so i would leave that as is. increased performance + increased battery life would be very welcome.

 

[Eug]

In other news, new Intel iMacs are showing up in Geekbenches.
It looks like the latest (last?) Intel iMac will be a Core i9-10910.

I don't know the i9-10910 is because it's not listed at Intel, so it could be an exclusive Apple release.

Geekbench lists this:

Core i9-10910
3.6 GHz base frequency
10-core
20 MB L3

Radeon Pro 5300
1.65 GHz max frequency
20 compute units
4 GB RAM

This will provide an interesting baseline with which to compare the ARM Macs, once we get the CPU scores. I couldn't find that, just the OpenCL score.

 

[DrMrLordX]

10 cores? Apple is using Comet Lake?

 

[Eug]

10 cores? Apple is using Comet Lake?

Why wouldn't they?

Tom's is guessing it's a 95 Watt desktop 3.6/4.7 GHz part, specifically for Apple.

Intel Core i9-10910, AMD Radeon Pro 5300 Leak in Unannounced Apple iMac

Apple getting the good stuff.

www.tomshardware.com

That would be faster than the 65 W 10900. Not as fast as the 125 W 10900K, but not as hot either. Such a 95 W part could fit in the existing iMac form factor (albeit with quick fan ramp ups).

I wonder how hot they'd make ARM desktop chips.

 

[soresu]

The release the mobile chips before desktop because efficiency matters more for mobile and mobile chips usually have less cores so higher yield. This is especially obvious with the 10nm fiasco.

Also because laptops are a huge percentage of modern sales - the desktop is good for us enthusiasts, but the lions share of consumer PC purchases are laptop.

 

[blckgrffn]

10 cores? Apple is using Comet Lake?

In an iMac? How are they fitting the water cooling in there?

In some seriousness, does this mean that we can expect a 10 core replacement for for high end iMacs and other high end Apple SKUs? It seems unlikely they would backtrack on core counts. Possible given really strong cores, even fewer in number, could outpace a larger number of weaker cores, but Apple seems to do a good job of going forward in terms of marketing numbers people claim to understand.

 

[jpiniero]

In an iMac? How are they fitting the water cooling in there?

PL2 probably isn't 255 W like it is on enthusiast boards.

The current model offers the 9900K as an upgrade on the 27", btw.

 

[Eug]

PL2 probably isn't 255 W like it is on enthusiast boards.

The current model offers the 9900K as an upgrade on the 27", btw.

Yes, the 9900K is a 95 W part. Comet Lake doesn't have such a part, so it would make sense to have a semi-custom 95 W 10910 non-K part. Apple could reuse the existing form factor for their last Intel hurrah, and then the rumoured new form factor would be ARM.

That said, I had a so-called 91 W i7-7700K iMac and returned it because heavy work would make it sound like a vacuum cleaner. I bought a 65 W i5-7600 and was much, much happier.

 

[Gideon]

In some seriousness, does this mean that we can expect a 10 core replacement for for high end iMacs and other high end Apple SKUs? I

It could also be 8 big and 8 small cores, similar to Alder Lake

 

[Eug]

In an iMac? How are they fitting the water cooling in there?

In some seriousness, does this mean that we can expect a 10 core replacement for for high end iMacs and other high end Apple SKUs? It seems unlikely they would backtrack on core counts. Possible given really strong cores, even fewer in number, could outpace a larger number of weaker cores, but Apple seems to do a good job of going forward in terms of marketing numbers people claim to understand.

Apple can market however they want and get away with it.

However, Bloomberg has already said Apple will release 12-core (8+4) ARM chips, and is looking at designs with more than 12 cores for later.

Apple Aims to Sell Macs With Its Own Chips Starting in 2021

Apple Inc. is planning to start selling Mac computers with its own main processors by next year, relying on designs that helped popularize the iPhone and iPad, according to people familiar with the matter.

www.bloomberg.com

The first Mac processors will have eight high-performance cores, codenamed Firestorm, and at least four energy-efficient cores, known internally as Icestorm. Apple is exploring Mac processors with more than 12 cores for further in the future, the people said.

EDIT:

[Handwaving mode ON]

Now that I think of it, if Apple could put out a chip that does Geekbench 5 at 1550 single-core, and release that as an 8+4 core part, that means they could potentially release an iMac non-Pro with a Geekbench MT score of well over 10000. Maybe even upwards of 12000 or more.

 

[Doug S]

That said, I was wondering how Apple will leverage this *to make more money per unit* above and beyond the obvious potential of not giving money to Intel.

I don't think this is about making more money, though they will undoubtedly save money not paying for Intel's fat margins, and keep that for themselves.

My hunch has been this:

1) They will keep pricing the same when they swap out an x86 Mac for an ARM Mac
2) The replacement ARM Mac will have better performance than the x86 Mac it replaces
3) The replacement ARM Mac will have better battery life than the x86 Mac it replaces (for notebooks)

That's how they're going to market this - you pay the same price, but you get something that's faster and lasts longer on battery. They'll also tout the stuff like the NPU as bringing capabilities to the PC world that haven't existed previously (at least not as a dedicated part of the silicon, rather than running code on the CPU or GPU) though that depends more on what use developers can come up with for it. Usually Apple has a thing or two up their sleeves, either they will have developed something internally that makes good use of it, or some third party developer will have.

They could have made this transition a few years ago if they were only concerned about making more money, but they wouldn't have been able to make the claim about a performance improvement so keeping the price the same with less performance and only better battery life would be a harder sales job. They waited as long as they did until they knew they could beat the x86 Macs.

So that new 10 core iMac coming out - look for Apple to have something that beats it. Whether that's with 8 cores or 12 cores or what we'll see. The fact there's a new x86 model coming out there is an indication we won't see an ARM version of that right away. I figure the Pro models will go ARM last, as it will take them more time to get those designs ready and they'll want to insure the large and expensive applications those customers use have been updated to ARM first. No one is going to pay $5000 for a Mac Pro and run their main application via Rosetta no matter how good of a job it does.

 

[blckgrffn]

It could also be 8 big and 8 small cores, similar to Alder Lake

For sure! That's too easy for them, I guess. No hurdle at all.

Also @Eug Yeah, after some thought it seemed obvious they would build a part that has 10+ of some assortment of cores (or 12, clearly) and voila, more is better. Marketing everyone can understand! (I know Apple gets special treatment but simple is good )

If the net performance of 8 + 4 cores is approaching 10900k & 3900x levels but at power levels that allow for sustained performance without a wind tunnel, do they need anything better for the non-uber-tower "Pro" models? Again, I am thinking Apple is in this to make money and not give mankind gifts and fewer SKUs is better.

 

[Eug]

Also @Eug Yeah, after some thought it seemed obvious they would build a part that has 10+ of some assortment of cores (or 12, clearly) and voila, more is better. Marketing everyone can understand! (I know Apple gets special treatment but simple is good )

If the net performance of 8 + 4 cores is approaching 10900k & 3900x levels but at power levels that allow for sustained performance without a wind tunnel, do they need anything better for the non-uber-tower "Pro" models? Again, I am thinking Apple is in this to make money and not give mankind gifts and fewer SKUs is better.

The Mac Pro will still need significantly faster performance even if the first ARM high end consumer chip hits 12000 in GB5 MT. And it makes sense for them to keep the Mac Pro, not so much as a high sales item (cuz it isn't) but more so to keep the creative pros in the movie industry and a few other pros happy, and as a boutique marketing item like their high end monitor. They'll want a CPU hitting north of 20000. Will they remove the GPU from such a CPU though? I dunno, maybe.

I don't know what they'd do with the iMac Pro though. If they drop anything, I'm thinking it will be this one.

 

[IvanKaramazov]

[Handwaving mode ON]

Now that I think of it, if Apple could put out a chip that does Geekbench 5 at 1550 single-core, and release that as an 8+4 core part, that means they could potentially release an iMac non-Pro with a Geekbench MT score of well over 10000. Maybe even upwards of 12000 or more.

Everyone else here is more educated on such things than I am, but unless I'm missing something, Eug, I expect that hypothetical 8+4 chip you're speculating about is headed for the 13" Macbook Pro.

The A13 runs a bit over 1300 single core, and TSMC estimates a 15% performance improvement just from the move from 7nm to 5nm. Assuming the A14 in the iPhone hits that mark exactly it will be 1500ish SC (as you've speculated), and historically the iPad variants have had the same single core performance with more cores. An A14X with 4+4 might feasibly (now I'm handwaving) be in the ballpark of 1500 / 6000 (too high?), and that presumably around 8-10 watts. Apple could probably keep an 8+4 variant well within the 28w processor budget of the current 13" MBP, in which case even assuming the Mac processors are just iPad chips with more cores and a fan we might (maybe?) expect 1500 / 10000+ in the smaller MBP.

 

[blckgrffn]

The Mac Pro will still need significantly faster performance even if the first ARM high end consumer chip hits 12000 in GB5 MT. And it makes sense for them to keep the Mac Pro, not so much as a high sales item (cuz it isn't) but more so to keep the creative pros in the movie industry and a few other pros happy, and as a boutique marketing item like their high end monitor. They'll want a CPU hitting north of 20000. Will they remove the GPU from such a CPU though? I dunno, maybe.

I don't know what they'd do with the iMac Pro though. If they drop anything, I'm thinking it will be this one.

Right, if the Mac Pro is to be succeeded it likely needs more than what the A14 incarnation is that fills the 16” MacBook.

That said, if there is one Apple x86 product sold right up to the deadline of 2 yrs wouldn’t it be the Mac Pro? Just made available for purchase and realistically it’s CPU configuration wasn’t that impressive (compared to its other facets) even when it was finally available for purchase.

If that is the fact isn’t it conceivable we are looking beyond A14 for the replacement? And if the CPU bogey stays a 1S Xeon on 14nm it seems like that a high end desktop A series SKU might get there in 2022? Maybe the larger challenge is being able to host a competitive amount of RAM? By then it seems we should have even higher densities (ddr 5?) available, which might make for another easy “win”.

Like you pointed out, the other question is if they keep external dGPUs and accelerators or if they make it all part of the package. 🤷‍♂️ Seems like at least putting them on PCIe busses keeps up charging for expanding on the option list.

Also by then it’s a pretty locked in affair with Arm and MacOS. Even if you don’t like that Apple “only” puts 16 high performance CPUs in your $45k tower, if the Mac Pro is what you need for MacOS and all the frame buffer and accelerator cards, what are you going to do?

 

[Eug]

Everyone else here is more educated on such things than I am, but unless I'm missing something, Eug, I expect that hypothetical 8+4 chip you're speculating about is headed for the 13" Macbook Pro.

The A13 runs a bit over 1300 single core, and TSMC estimates a 15% performance improvement just from the move from 7nm to 5nm. Assuming the A14 in the iPhone hits that mark exactly it will be 1500ish SC (as you've speculated), and historically the iPad variants have had the same single core performance with more cores. An A14X with 4+4 might feasibly (now I'm handwaving) be in the ballpark of 1500 / 6000 (too high?), and that presumably around 8-10 watts. Apple could probably keep an 8+4 variant well within the 28w processor budget of the current 13" MBP, in which case even assuming the Mac processors are just iPad chips with more cores and a fan we might (maybe?) expect 1500 / 10000+ in the smaller MBP.

Yes, I think the MacBook Pro will get 8+4, but I also think iMacs will get 8+4. Perhaps some of them will be higher clocked. ie. Apple finally bins their chips to a significant extent.

However, I get the impression that Apple doesn't like the current Wattages in their laptops. The current 16" MacBook Pro is a vacuum cleaner, as are their high end iMacs.

I wouldn't be surprised to see Apple going with fanless 12" MacBooks (again) and fanless 13" MacBook Airs (or whatever they decide to call them), and also go with lower wattages in their MacBook Pros. Similarly, I might expect the iMacs to lose the 95 Watt category completely.

Perhaps I'm biased though, because I really hate loud computers. However, I say this not just because of the fan noise, but also because I expect Apple is going with new form factors with their ARM Macs, especially the iMac.

 

[defferoo]

Yes, I think the MacBook Pro will get 8+4, but I also think iMacs will get 8+4. Perhaps some of them will be higher clocked. ie. Apple finally bins their chips to a significant extent.

However, I get the impression that Apple doesn't like the current Wattages in their laptops. The current 16" MacBook Pro is a vacuum cleaner, as are their high end iMacs.

I wouldn't be surprised to see Apple going with fanless 12" MacBooks (again) and fanless 13" MacBook Airs (or whatever they decide to call them), and also go with lower wattages in their MacBook Pros. Similarly, I might expect the iMacs to lose the 95 Watt category completely.

Perhaps I'm biased though, because I really hate loud computers. However, I say this not just because of the fan noise, but also because I expect Apple is going with new form factors with their ARM Macs, especially the iMac.

i think a fanless 12” Macbook or 13” Macbook air is definitely in the cards. Just throw in a 4+4 A14X and it’ll already be much faster than the current Macbook Air.

An 8+4 13” Macbook pro would probably be my ideal machine in terms of performance and size. Finally time to say goodbye to Intel’s TDP games over the last few years as a result of trying to squeeze more out of 14nm.

 

[Eug]

i think a fanless 12” Macbook or 13” Macbook air is definitely in the cards. Just throw in a 4+4 A14X and it’ll already be much faster than the current Macbook Air.

Actually, I think they could actually get away with an A14 non-X for the MacBook and MacBook Air. They could play with clock speeds if necessary.

Remember, the A13 in the frickin' iPhone already scores higher than the Core i7-1060NG7 in the 2020 MacBook Air (which gets complaints because of fan noise).

An 8+4 13” Macbook pro would probably be my ideal machine in terms of performance and size. Finally time to say goodbye to Intel’s TDP games over the last few years as a result of trying to squeeze more out of 14nm.

Maybe:

A14 (2+4): MacBook and MacBook Air
A14X (4+4): 14" MacBook Pro and 24" iMac
A14? (8+4): 16" MacBook Pro and 29" iMac

 

[JasonLD]

An 8+4 13” Macbook pro would probably be my ideal machine in terms of performance and size. Finally time to say goodbye to Intel’s TDP games over the last few years as a result of trying to squeeze more out of 14nm.

13" will probably remain 4+4 simply due to product segmentation. Unless they have 12+4 configuration for higher end Macbook pro but that doesn't sound likely at this point.

 

[Doug S]

Right, if the Mac Pro is to be succeeded it likely needs more than what the A14 incarnation is that fills the 16” MacBook.

I don't think anyone disputes that. The open questions are:

1) will it be A14 based or come far enough in the future that it is A15 or possibly even A16 based?

2) will it be a monolithic 'big chip with lots of cores' like Intel or a 'chiplet' strategy like AMD?

3) if it is a chiplet strategy, will they design the chips that go into smaller Macs like the hypothetical 8+4 chip going into a Macbook Pro in such a way that they can function in that role as a standalone but also work as a chiplet with one or more others to create 16, 24 and 32 core variations?

If Apple plans to use their own GPU in the Mac Pro (something that we have no clue about, but have to recognize as a real possibility) there wouldn't be room for enough CPU cores and a big enough GPU on one die, though they could have two big dies - one with lots of CPU cores and one with lots of GPU units. All dies might have 32 cores but enable only 16 of them in a lower end model etc.

 

[beginner99]

Go check out the A13's ST performance for clang in GB5 and gcc in SPEC2006, and compare with Intel & AMD's fastest CPUs, and tell us again how Intel and AMD will "show their advantages" for compiling.

Sadly I can't find this anymore but I took that straight from Linus Torvalds where he says that these benches (the compile parts) are crap and way too small to test anything meaningful. A real test would be the compilation of the Linux kernel.
Since I can't judge it myself I'm gonna go with Torvalds over a random internet stranger.

Your idea that ARM ISA chips can’t do “complex stuff” like compiling and databases

I never said "ARM ISA chip". I said graviton2 which like Apple cores are ultimatley cpu cores made for lower power and client use. Of course you can make an ARM chip just like intel/amd.

 

[Doug S]

Sadly I can't find this anymore but I took that straight from Linus Torvalds where he says that these benches (the compile parts) are crap and way too small to test anything meaningful. A real test would be the compilation of the Linux kernel.
Since I can't judge it myself I'm gonna go with Torvalds over a random internet stranger.

Fine you can continue with your misplaced doubt about Apple's ARM chips until the new Macs are released in six months and you're forced to eat your words.