Question Bloomberg: Apple testing SoCs with 16 and 32 high performance cores

[mikegg]

The current ‌M1‌ chip has four high-performance processing cores and four power-saving cores. For its next generation chip targeting MacBook Pro and ‌iMac‌ models, Apple is said to be working on designs with as many as 16 power cores and four efficiency cores.

Apple is also reportedly testing a chip design with as many as 32 high-performance cores for higher-end desktop computers planned for later in 2021, as well as a new half-sized Mac Pro planned to launch by 2022.

Bloomberg: Apple Working on Next-Gen Apple Silicon Chips for MacBook Pro, iMacs, and Mac Pro Due to Launch Next Year

Apple is working on a series of new custom Apple silicon processors to power upgraded versions of the MacBook Pro, new iMacs, and a new Mac Pro for...

www.macrumors.com

 

[mikegg]

If the M1X for the Macbook Pro 16" gets 8 high-performance cores, it will be the undisputed fastest single-threaded and multi-threaded laptop CPU. If it gets 12 high-performance cores, it should rival desktop Zen3 in multi-threaded performance. And if it happens to go up to 16 cores, it should be the fastest non-HEDT CPU.

If Apple does somehow manage to put a SoC with 16 high performance CPU cores into the Macbook Pro 16", it's hard for me to imagine how AMD or Intel will ever manage to catch up in the laptop space.

Apple's ambitions are extremely high if they're already planning a 32-core CPU SoC for the Mac Pro by 2022.

It'll be interesting to see how Apple plans to package the SoCs and if they will go the chiplet approach or stay monolithic. My guess is that they will stay monolithic and just create different SKUs.

On a semi-related note, the same article claims Apple is testing 64-core and 128-core GPUs. I assume that this GPU would have to discrete from the SoC. For reference, the M1 has only 8 GPU cores.

 

[mikegg]

I also wonder if Apple will eventually parlay their SoC designing prowess into designing server CPUs and GPUs. If they could design a 32-core CPU and a 128-core GPU, they're not too far off. Perhaps they will buy Nuvia and enter the cloud business.

Of course, this cloud might be exclusive to Apple's internal use or to Apple customers only.

I presume that Apple isn't stupid and could see the trend of more computing moving from local devices to the cloud.

 

[Thunder 57]

We get it, you love Apple. Let's just wait and see though, OK?

Nice downvote, bozo.

 

[Thunder 57]

So AMD/Intel should just call it quits, right? The M1 has 6 billion+ more transistors than Renoir. You don't think AMD/Intel could spend some transistors on commonly performed tasks?

I'll give you that the performance/watt is very good. But you'd probably downvote me for saying that. Let's just wait and see what happens, or is that worth another downvote?

 

[beginner99]

If Apple does somehow manage to put a SoC with 16 high performance CPU cores into the Macbook Pro 16", it's hard for me to imagine how AMD or Intel will ever manage to catch up in the laptop space.

Apple's ambitions are extremely high if they're already planning a 32-core CPU SoC for the Mac Pro by 2022.

I don't disagree it still makes we wonder why everyone complains about high process cost but supposedly a custom SOC for a relatively niche market of MacPros is financially viable. I can see 16 cores because that can go into iMAcs and possibly larger laptops but 32 cores? MacPro only and I fail to see how that will be profitable. Either process cost are way overblown and just an excuse to raise prices (GPUs and smartphomes most notably, funny, because with x86 CPUs the opposite happened) or a custom SOC for a MacPro is not financially viable (without projected huge growth).

Given that NVs margins are very high and rising and Intels margins are getting lower I think we are getting scammed more or less. Problem is apple is a different ecosystem so x86 camp may react slowly or not at all compared to one of them making this homerun. Intel has run themselves into a corner with the fabs and AMD simply is too small to be competitive to Apple. So yeah I guess you are right it this makes it to market.

 

[beginner99]

You don't think AMD/Intel could spend some transistors on commonly performed tasks?

If you see my history you know I'm not an Apple fan. Yes they could at them but they (AMD/Intel) must agree on the API and vendors must make use of it. Far more complex than if 1 company controls the whole process.

 

[Shmee]

If you could build a PC with one of these apple CPUs, that would be kind of cool. You would need a motherboard with good features though, and it would need to support PCIe gen 4 of course.

 

[Doug S]

I also wonder if Apple will eventually parlay their SoC designing prowess into designing server CPUs and GPUs. If they could design a 32-core CPU and a 128-core GPU, they're not too far off. Perhaps they will buy Nuvia and enter the cloud business.

Why would they buy Nuvia if they are already currently designing 32 core CPUs and 128 core GPUs? What would Nuvia give them that they can't create themselves or wouldn't have already created by the time Nuvia designs hit the market? I'm not sure why some people seem to believe that Apple will/should buy Nuvia - if Gerard Williams III et al wanted to remain Apple employees they would have done so without an end run strategy of startup/buyout.

The reason they left was because either 1) Apple wasn't interested in building what they are going to build (i.e. at the time had no desire to build server class CPUs and maybe still don't) or 2) GW III wanted to be his own boss and design what he wanted, and maybe become much richer along the way than if he remained someone else's employee.

If Apple eventually rolls their own CPUs for the cloud it will be a private cloud for operating stuff like iCloud, Siri, maybe a privacy focused Google Search replacement, and perhaps offer supplemental number crunching power to certain classes of apps. Apple is a consumer company, the cloud market is pretty much B2B and B2G only and Apple has never done either well when they've tried. Where they do they partner with someone else who has expertise and sales contacts in that segment, like the companies that provide MDM solutions for devices, or IBM for iPad business apps.

I can't see them having much desire to do cloud gaming - the reason for cloud gaming is to allow gaming on devices that don't have enough GPU horsepower for local gaming. Apple sells higher end devices that will offer higher end graphics performance, which makes cloud gaming rather pointless.

 

[mikegg]

Why would they buy Nuvia if they are already currently designing 32 core CPUs and 128 core GPUs? What would Nuvia give them that they can't create themselves or wouldn't have already created by the time Nuvia designs hit the market? I'm not sure why some people seem to believe that Apple will/should buy Nuvia - if Gerard Williams III et al wanted to remain Apple employees they would have done so without an end run strategy of startup/buyout.

There have been plenty of times in the past where a top employee spun out to start his/her own company and then get bought by the original company.

It's entirely possible that Apple did not want to go into the cloud hardware when Gerard was there but has changed its mind or will change its mind.

Nuvia would give them a head start because designing server CPUs is very different than designing SoCs. And Apple would be bringing back top talent.

Anyway, it's just a thought. It wouldn't surprise me if Apple did it.

 

[mikegg]

We get it, you love Apple. Let's just wait and see though, OK?

Nice downvote, bozo.

Can we have a conversation without you getting all butthurt that Apple is/will/could be faster than AMD for f sakes?

Yes, we get it. You love AMD and you want to own the fastest chips without having to use Apple products. Not being able to in the near future hurts your manlihood. Now let's put that aside and focus on actual CPU discussions.

 

[DrMrLordX]

Can we have a conversation without you getting all butthurt that Apple is/will/could be faster than AMD for f sakes?

Yes, we get it. You love AMD and you want to own the fastest chips without having to use Apple products. Not being able to in the near future hurts your manlihood. Now let's put that aside and focus on actual CPU discussions.

Not a terribly convincing argument.

 

[Kenmitch]

Can we have a conversation without you getting all butthurt that Apple is/will/could be faster than AMD for f sakes?

Yes, we get it. You love AMD and you want to own the fastest chips without having to use Apple products. Not being able to in the near future hurts your manlihood. Now let's put that aside and focus on actual CPU discussions.

Why not post in the Apple Laptops and Desktops sub-forum? Most visitors to the CPU forum aren't really looking for Apple related info. I don't Apple, but if I did most likely I wouldn't be looking in the CPU sub-forum for info.

 

[Doug S]

Why not post in the Apple Laptops and Desktops sub-forum? Most visitors to the CPU forum aren't really looking for Apple related info. I don't Apple, but if I did most likely I wouldn't be looking in the CPU sub-forum for info.

Because you believe that somehow x86 is the only the CPU that should be discussed in the "CPU forum"? Why not discuss x86 CPUs in the "Laptops" or "Peripherals and Components" sub-forum, and just delete this sub-forum? If you don't want to read about anything but x86, why are you on this thread?

It doesn't matter if most visitors to the CPU forum aren't looking to discuss Apple stuff. That's why threads have titles. Click on the ones you want to read, and don't click on the ones you don't want to read. Hint, if the thread title is "Apple testing SoCs with 16 and 32 high performance cores" that's not a thread you want to click on if you personally believe x86 is all that matters.

 

[Kenmitch]

Because you believe that somehow x86 is the only the CPU that should be discussed in the "CPU forum"? Why not discuss x86 CPUs in the "Laptops" or "Peripherals and Components" sub-forum, and just delete this sub-forum? If you don't want to read about anything but x86, why are you on this thread?

It doesn't matter if most visitors to the CPU forum aren't looking to discuss Apple stuff. That's why threads have titles. Click on the ones you want to read, and don't click on the ones you don't want to read. Hint, if the thread title is "Apple testing SoCs with 16 and 32 high performance cores" that's not a thread you want to click on if you personally believe x86 is all that matters.

There really isn't any reason to get upset over what I typed. If you don't like it....Well then take your own advice and ignore it.

 

[Carfax83]

While I would love to see Apple make more aggressive HEDT parts, I'm still not sold on the "wider is better" argument. While the high single threaded performance is impressive, it seems as though it comes with some sacrifices (namely clock speed) and as a result, may not be as suited to desktop workloads. Case in point, I was quite surprised to see how badly the M1 was beaten by even the old Skylake architecture let alone Zen 3, when it came to HEVC software encoding (way more compute intensive than H264) despite the huge advantage in IPC and process node. I realize that the software is beta, but considering the huge gap, I don't think any greater optimizations could completely bridge that gap even on a core for core basis.

This ties in to my argument from the other thread. Single threaded performance while extremely important due to how it contributes to the overall throughput of an architecture, is not very relevant in and of itself in desktop/workstation/server workloads because those workloads have long been progressing towards more and more parallelism.

To further my point, many of the benchmarks that the Apple proponents like to use as examples of the M1's prowess like Geekbench and Spec all run much faster in multithreaded mode than they do in single threaded mode. In fact, it's not even close. So while Apple's M1 is highly impressive (especially from a perf/watt perspective), it seems to come up short against modern x86-64 architectures like Zen 3 that prioritize high clock speeds, SMT and wide SIMD in heavy workloads.

Increasing the core count will of course go a long way in evening the odds, but it's possible that they may have to adapt their cache hierarchy which could end up affecting single threaded performance.

 

[naukkis]

HEVC does have hand-tuned assembly for x86 which makes the difference.

M1 is still just a low-power Ipad chip, chips for high performance laptops and desktops aren't yet released.

 

[Bam360]

Clock speed alone is meaningless, it doesn't make sense to say that low clock speed is bad for desktop, what matters is clock speed plus IPC. Who cares if Apple can't get past 3.5GHz on a more aggressive desktop chip (probable) if IPC is like 50% higher, on average, than the best alternative.
HEVC may be a case of terrible optimization (likely based of what has been said about hand-tuning and not using SIMD on ARM) or simply a bad workload for this Apple architecture, kind of like what Zen was for gaming (in comparison with most other workioads).

 

[Doug S]

HEVC does have hand-tuned assembly for x86 which makes the difference.

M1 is still just a low-power Ipad chip, chips for high performance laptops and desktops aren't yet released.

Doesn't matter, a year or two from now when Mac software has hand tuned SIMD code for both sides instead of just x86, and Macs with many more cores are available to compete on the biggest multithreaded tasks, the naysayers will find something else even more niche that the x86 excels at and point to that as the one true test.

Eventually they'll be forced to bring back microbenchmarks that are essentially measures of clock speed like Dhrystone, because that will be all that they have left where x86 wins convincingly.

 

[DrMrLordX]

the naysayers will find something else even more niche

You think software encoding is niche?

 

[Bam360]

We also need to remind ourselves x265 uses better the AVX 256-bit capabilities compared to x264, I remember Zen 8-core losing against Skylake 6-core in handbrake x265, whereas x264 was a different story and the delta was closer to other workloads like Cinebench, Zen2 got an amazing improvement in handbrake x265.
What I mean is that, even when optimized properly, it should come as no surprise that in workloads that can utilize full 256-bit capabilities, Apple has lower performance compared to other workloads. If Apple cares about these workloads, then it should implement SVE with 256-bit at the very least sooner than later.

 

[Carfax83]

HEVC does have hand-tuned assembly for x86 which makes the difference.

HEVC also has hand tuned assembly for ARM64 since version 3.4. I'm not saying the amount of optimization is as extensive as it is for x86-64, but it does have some.

Faster ARM64 encodes using x265 - x265

 

[Carfax83]

Clock speed alone is meaningless, it doesn't make sense to say that low clock speed is bad for desktop, what matters is clock speed plus IPC. Who cares if Apple can't get past 3.5GHz on a more aggressive desktop chip (probable) if IPC is like 50% higher, on average, than the best alternative.

I agree, clock speed+ IPC = performance. Just like how clock speed isn't an endless well of performance due to diminishing returns, so it is with IPC.

What matters is performance. AMD and Intel seem to target 5ghz, and can still squeeze out double digit IPC gains. That's why I said that I'm not fully sold on the wider is better argument.

HEVC may be a case of terrible optimization (likely based of what has been said about hand-tuning and not using SIMD on ARM) or simply a bad workload for this Apple architecture, kind of like what Zen was for gaming (in comparison with most other workioads).

See above. HEVC does have ARM64 hand tuned assembly optimizations.

 

[Carfax83]

We also need to remind ourselves x265 uses better the AVX 256-bit capabilities compared to x264, I remember Zen 8-core losing against Skylake 6-core in handbrake x265, whereas x264 was a different story and the delta was closer to other workloads like Cinebench, Zen2 got an amazing improvement in handbrake x265.
What I mean is that, even when optimized properly, it should come as no surprise that in workloads that can utilize full 256-bit capabilities, Apple has lower performance compared to other workloads. If Apple cares about these workloads, then it should implement SVE with 256-bit at the very least sooner than later.

The M1 has 4x 128 bit NEON, so isn't that similar to 2x 256 bit AVX in throughput?

 

[Carfax83]

Doesn't matter, a year or two from now when Mac software has hand tuned SIMD code for both sides instead of just x86, and Macs with many more cores are available to compete on the biggest multithreaded tasks, the naysayers will find something else even more niche that the x86 excels at and point to that as the one true test.

Eventually they'll be forced to bring back microbenchmarks that are essentially measures of clock speed like Dhrystone, because that will be all that they have left where x86 wins convincingly.

So with Zen 4 projected to have an IPC gain of at least 20% or greater and Golden Cove with 25% while presumably still targeting 5ghz, you think that x86-64 will not be able to compete favorably?