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:

 

[ksec]

I still don't understand why Apple would scrap ARM on notebooks (a larger market) just because they can't produce ARM chips suitable for Mac Pros (a smaller market), whether Apple can produce ARM chips for PC use profitably is still up in the air but the Mac Pro would be well down the priority list of comsiderations.

Given what Apple has done in the past, they'd gladly kill the Mac Pro line if it was such a hassle to switch to ARM.

Had Mac Pro not been announced, this would have been the case.
Had Gerard Williams not left Apple to found a high performance ARM CPU, that would have been the case.

Another factor to consider, the Mac platform are now largely used by a few pro group, iOS Developers, Programmers, graphics design and video editing. Most consumer are no longer on the Mac Platform. They are shifting to iPad and iPhone. iPad has more Active user than Mac.

Basically it doesn't make much sense for such drastic change in this late stage of Mac circle. It is much better to milk it for another 4 years to see how things goes. After all Apple now has AMD as negotiation card, BOM concern isn't as important when Intel was in monopoly.

 

[Mopetar]

Apple still might use one of their own chips at the low end. They don’t need to immediately switch the entire lineup to ARM.

The biggest advantage is on the low end for their MacBook Air or maybe even Mac Mini. Even though they aren’t using top end chips, the low-power chips from Intel still command a price premium. Apple likely gets a deal, but they stand to save a lot from switching to their own SoC.

Either that add a significant chunk of profit per unit sold on what’s likely one of the highest volume Mac products or they can drop prices to gain additional sales. Possibly both at the same time.

 

[jpiniero]

Apple still might use one of their own chips at the low end. They don’t need to immediately switch the entire lineup to ARM.

Apple could easily do the entire laptop lineup, and really the Mini and iMac too.

Have to think it's going to be multiple years before the Mac Pro is refreshed no matter what Apple does, so that's plenty of time for an suitable ARM alternative to arrive by then.

 

[Mopetar]

I think they’d want to leave the pro or other high end models on x86 for a while just so there aren’t issues.

The MacBook Air is just a web browsing computer that does some light work using the kind of office apps that Microsoft and others already have working on ARM.

Moving their full lineup to their own SoCs probably requires a design that would let them do something similar to what AMD has done with Zen. Apple doesn’t sell enough to justify a separate chip.

 

[soresu]

I think they’d want to leave the pro or other high end models on x86 for a while just so there aren’t issues.

At the very least until the lions share of the major MacOS software vendors (Adobe/Autodesk/Avid...) have ARM ports standing up for their most popular packages, and moderately optimised too.

 

[Richie Rich]

Moving their full lineup to their own SoCs probably requires a design that would let them do something similar to what AMD has done with Zen. Apple doesn’t sell enough to justify a separate chip.

Good point. However Apple makes two different chips already:
- small SoC A13 with 2x big cores for iPhone
- big SoC A13X with 4x big cores for iPad (Apple can add inter-chip feature only for this one)

Adding inter-chip links to support 2xSoC configuration is pretty easy.
BTW this configuration with 8x big cores and 8x small core would be very powerfull. Apple's A13 Lightning core @2.6 GHz is little bit faster than Zen2 @4.6 GHz. This would be the most powerful 8xbig core LAPTOP chip ever (assuming old A13). If new upcoming A14 will bring +10% IPC and +10% clocks then Apple will take the absolut performance crown too (sufficient for all desktops except MacPro). And with 4x SoC configuration like Zen1 might be enough even for Mac Pro (IMHO this will happen only when SW for ARM is available, maybe in 2-4 years).

 

[soresu]

IMHO this will happen only when SW for ARM is available, maybe in 2-4 years

Yes, still a sizable stumbling block unfortunately.

There's also a small possibility that Apple may get some Threadripper action in the interim - though I highly doubt it, only that insane 2x2 GPU solution on the Mac Pro has it even crossing my mind.

 

[Thala]

And with 4x SoC configuration like Zen1 might be enough even for Mac Pro (IMHO this will happen only when SW for ARM is available, maybe in 2-4 years).

I dont know where the notion, that it takes time to compile SW for ARM is coming from. It is a decision in the first place and then some work perhaps but not even close to a year.

Speaking of decisions, they will come earlier if you move your entire line-up to ARM and later when you continue to support x86 partially. Imagine Apple would have continued to release PowerPC Macs after the move to Intel - would have only extended the SW transition.

 

[Richie Rich]

I dont know where the notion, that it takes time to compile SW for ARM is coming from. It is a decision in the first place and then some work perhaps but not even close to a year.

Speaking of decisions, they will come earlier if you move your entire line-up to ARM and later when you continue to support x86 partially. Imagine Apple would have continued to release PowerPC Macs after the move to Intel - would have only extended the SW transition.

I totally agree. If there is a company with power to force SW producers to migrate to ARM then it's the Apple. On the other hand there is additional cost of such a migration. I'm sure there is a lot of corporate politics and pressure against this change.

I can imagine Apple supports both ISAs (ARM and x86) for some time. Just to show people how bad and slow x86 CPUs are

 

[jpiniero]

The problem that ARM is going to have, Apple notwithstanding, is that consumers just don't care. Half of the Top 25 in laptop "Best Sellers" at Amazon right now are either aging Atoms or 5 year old MediaTek. There probably isn't much reason to invest that much money in it.

Now Corporate Users don't buy off of Amazon, and they are easily the majority of the market, but they will stick with Intel as long as they continue to use WIndows unless they have a really good reason.... While the shortage is a really good reason, AMD can fill the void there.

 

[scannall]

I totally agree. If there is a company with power to force SW producers to migrate to ARM then it's the Apple. On the other hand there is additional cost of such a migration. I'm sure there is a lot of corporate politics and pressure against this change.

I can imagine Apple supports both ISAs (ARM and x86) for some time. Just to show people how bad and slow x86 CPUs are

Being iOS is OS X with a touch interface, and the desktop bits removed it shouldn't be much trouble really. Especially if they use Apple's compilers and design rules.

 

[Richie Rich]

CPU	SPECint2006 [pts]	Clock [GHz]	PPC [pts/GHz]	PPC increase to 9900K [%]	PPC increase to last gen [%]
Intel Core i9 9900K	54.28	5.0	10.86	0%
AMD Ryzen 3950X - Zen2	50.02	4.6	10.87	+0.2%
Cortex A76	26.65	2.84	9.38	-13.6%
Cortex A77	33.32	2.84	11.73	+8.1%	+25.1% (to A76)
Apple A9 - Twister	21.49 (avg)	1.85	11.61	+7.0%
Apple A10 - Hurricane	29.35 (avg)	2.34	12.54	+15.5%	+8.0% (to A9)
Apple A11 - Monsoon	36.80	2.39	15.40	+41.8%	+22.8%
Apple A12 - Vortex	45.32	2.53	17.91	+65.0%	+16.3%
Apple A13 - Lightning	52.82	2.65	19.93	+83.6%	+11.3%

 

[whm1974]

Why is this thread still alive? It will take many decades if ever for ARM SoC reach AMD and Intel CPU performance, much less crush them. And neither is AMD nor Intel are sitting are their arse.

 

[NostaSeronx]

I heard that the next Apple SoC will be doing "desktop" first(big soc) and phone second(little soc).

Mac mini - Technical Specifications

See all the technical specifications for Mac mini with the M2 or M2 Pro chip.

www.apple.com

^--this but:
Apple SoC 5nm
12 GB LPDDR5-5500(2x64)
4x USB 4.0

 

[whm1974]

I heard that the next Apple SoC will be doing "desktop" first(big soc) and phone second(little soc).

Mac mini - Technical Specifications

See all the technical specifications for Mac mini with the M2 or M2 Pro chip.

www.apple.com

^--this but:
Apple SoC 5nm
12 GB LPDDR5-5500(2x64)
4x USB 4.0

I don't think that DDR5 in any form or USB 4.0 are even out yet in consumer hardware. But I'll double check to keep victims of Apple's RDF off my back.

 

[scannall]

I don't think that DDR5 in any form or USB 4.0 are even out yet in consumer hardware. But I'll double check to keep victims of Apple's RDF off my back.

I also think it's very unlikely. Just pointing out that it doesn't need to be available to the general public for Apple to buy it.

 

[whm1974]

I also think it's very unlikely. Just pointing out that it doesn't need to be available to the general public for Apple to buy it.

The last I've read neither one of them is even available for really high end hardwar. So how is Apple going to buy this if the tech hasn't been finalized yet?

 

[scannall]

The last I've read neither one of them is even available for really high end hardwar. So how is Apple going to buy this if the tech hasn't been finalized yet?

Nah, both are available now to manufacturers. Expect to see more over the next year or two.

 

[whm1974]

Nah, both are available now to manufacturers. Expect to see more over the next year or two.

Right Now??? How can they can be available if the specs haven't been finalized yet?

 

[naukkis]

Right Now??? How can they can be available if the specs haven't been finalized yet?

Lpddr5 is available and even Intel next gen cpu Tigerlake will support it. USB4.0 is Thunderbolt compatible which Apple need, and it's almost finalized so it's possible to design device to support it.

 

[NostaSeronx]

I don't think that DDR5 in any form or USB 4.0 are even out yet in consumer hardware. But I'll double check to keep victims of Apple's RDF off my back.

LPDDR5* => https://www.samsung.com/semiconductor/dram/lpddr5/K3LK2K20BM-BGCN/
SK Hynix/Micron haven't updated their websites yet. However, they have similar products in sampling w/ production this year.

The USB 4.0 controller/switch hasn't been announced yet. It is stated it will be ready by Q3 2020 for high volume.

The Apple SoC on TSMC's 4FF will support DDR5 SO-DIMMs/DIMMs. While, the 5FF one only does LPDDR5. The second one is more aggressive replacing more intel Macs than the first.

 

[whm1974]

LPDDR5* => https://www.samsung.com/semiconductor/dram/lpddr5/K3LK2K20BM-BGCN/
SK Hynix/Micron haven't updated their websites yet. However, they have similar products in sampling w/ production this year.

The USB 4.0 controller/switch hasn't been announced yet. It is stated it will be ready by Q3 2020 for high volume.

The Apple SoC on TSMC's 4FF will support DDR5 SO-DIMMs/DIMMs. While, the 5FF one only does LPDDR5. The second one is more aggressive replacing more intel Macs than the first.

So how big a boast can we expect from the 1st Generation DDR5 memory? And how fast will USB 4 be?

 

[scannall]

Right Now??? How can they can be available if the specs haven't been finalized yet?

Specs have been final since last year.
August 2019 for USB 4, though DDR5 hasn't been finalized it's close. So you were right on that one. Thought it had been already.

 

[Richie Rich]

Why is this thread still alive? It will take many decades if ever for ARM SoC reach AMD and Intel CPU performance, much less crush them. And neither is AMD nor Intel are sitting are their arse.

I'm afraid you didn't realized that 16c Ryzen has lower SPECint score (50.02) than Apple A13 (52.82). This means that Ryzen has lower absolute performance despite running at 4.6-4.7 GHz (way higher than A13's 2.65 GHz). And that's the problem for x86, they run at peak voltage/clock already and cannot clock any faster. Apple can scale their uarch up to 4GHz (+50%) if there is need for max performance.

YEAR	INTEL	GeekBench 5.1 [pts; pts/GHz]	Apple	GeekBench 5.1 [pts; pts/GHz]	PPC difference to INTEL
2013	4770K 3.9GHz	966; 247.69	A7	270; 207.69	-16.15%
2018	9900K 5.0GHz	1384; 276.80	A12	1116; 441.11	+59.36%
	PPC increase	+11.75%		+112.38%
	PPC per year	+1.96%		+18.73%

In 2013 INTEL was leader in terms of PPC/IPC. However Apple's first 64-bit ARM CPU was not far behind.
While INTEL had long period of IPC stagnation Apple was bringing +18% IPC increase every year in average. That's pretty amazing what they achieved.


And some Geek Bench 5.1 comparison:

                                 CPU              GeekBench 5,1 [pts]             Clock                PPC [pts/GHz]      PPC increase to 9900K [%]PPC increase to last gen
                      Intel 9900K                                   1384                     5,00                   276,80                 +0,00   
                      AMD 3950X                                     1317                     4,60                   286,30                 +3,43   
                      ARM Cortex A76                                 720                     2,84                   253,52                 -8,41   
                      Apple A7                                       270                     1,30                   207,69                -24,97   
                      Apple A8                                       323                     1,40                   230,71                -16,65                   +11,08   
                      Apple A9                                       564                     1,85                   304,86                +10,14                   +32,14   
                      Apple A10                                      770                     2,34                   329,06                +18,88                    +7,94   
                      Apple A11                                      933                     2,39                   390,38                +41,03                   +18,63   
                      Apple A12                                     1116                     2,53                   441,11                +59,36                   +13,00   
                      Apple A13                                     1332                     2,65                   502,64                +81,59                   +13,95

 

[whm1974]

I'm afraid you didn't realized that 16c Ryzen has lower SPECint score (50.02) than Apple A13 (52.82). This means that Ryzen has lower absolute performance despite running at 4.6-4.7 GHz (way higher than A13's 2.65 GHz). And that's the problem for x86, they run at peak voltage/clock already and cannot clock any faster. Apple can scale their uarch up to 4GHz (+50%) if there is need for max performance.

YEAR	INTEL	GeekBench 5.1 [pts; pts/GHz]	Apple	GeekBench 5.1 [pts; pts/GHz]	PPC difference to INTEL
2013	4770K 3.9GHz	966; 247.69	A7	270; 207.69	-16.15%
2018	9900K 5.0GHz	1384; 276.80	A12	1116; 441.11	+59.36%
	PPC increase	+11.75%		+112.38%
	PPC per year	+1.96%		+18.73%

In 2013 INTEL was leader in terms of PPC/IPC. However Apple's first 64-bit ARM CPU was not far behind.
While INTEL had long period of IPC stagnation Apple was bringing +18% IPC increase every year in average. That's pretty amazing what they achieved.

Can Apple's A13 get real work done? If it can then why hasn't Apple offered the A13 in an actual HEDT Workstation? Why haven't they come out with a new Mac Pro using the Wonder SoC?

All Apple's A13 SOC is in are iPhones and iPads only.