M1 is the A14X but rebranded

[leman]

It's not truly 8 cores. It cannot truly process 8 things simultaneously, it can only do 4 hence the 4 cores. 8 cores != 4c/8t

It can run 8 hardware threads at once, so yes, it can definitely "truly process" 8 things simultaneously. M1 also supports 8 hardware threads at once.

Anyway, this discussion illustrates very well what I mean. What exactly are you counting as cores? The ability to run simultaneous execution contexts (threads)? Independent hardware backends? Independent hardware frontends? Hardware register sets? Execution schedulers? Modular hardware building blocks? Any of these things make sense, this way or another, as cores.

For a consumer, a core, is well, a certain promise of performance. A consumer expects an 8-core CPU to be roughly twice as fast as a 4-core CPU. This is generally true with a symmetric design, not so much with an asymmetric design.

Ah, well, it's all moot in the end. One can make this things arbitrarily complex as well as arbitrarily simple. The only sure thing is that the convenient times when we had a single CPU core and reasoning about these things was easy are long gone

 

[boss.king]

I don't believe Apple will use "M2X" name because M is associated with low-end devices. I believe Apple will call MBP SoCs something like "P2" meaning Pro. Then they could market them as "P2 10-Core Processor. 32-Core GPU". Maybe someday we'll get something like "P4 64-Core Processor. 128-Core GPU" for the Mac Pro.

The M1 is already in two products with the word Pro in their names, so any expressly Pro line of chips after this would be a very muddy distinction. This wouldn't be the first time they make a weird, conflicting decision like this (for example, USB C being their "Pro" port except for on their Pro phones, and M-series chips being for the motion coprocessor before becoming their main chip line) but I feel like they'll want to avoid any extra naming shenanigans wherever they can.

 

[frostbit3]

It can run 8 hardware threads at once, so yes, it can definitely "truly process" 8 things simultaneously. M1 also supports 8 hardware threads at once.

Anyway, this discussion illustrates very well what I mean. What exactly are you counting as cores? The ability to run simultaneous execution contexts (threads)? Independent hardware backends? Independent hardware frontends? Hardware register sets? Execution schedulers? Modular hardware building blocks? Any of these things make sense, this way or another, as cores.

For a consumer, a core, is well, a certain promise of performance. A consumer expects an 8-core CPU to be roughly twice as fast as a 4-core CPU. This is generally true with a symmetric design, not so much with an asymmetric design.

Ah, well, it's all moot in the end. One can make this things arbitrarily complex as well as arbitrarily simple. The only sure thing is that the convenient times when we had a single CPU core and reasoning about these things was easy are long gone

Threads != cores though. The best analogy I can give is that threads are like hands, you are 2 threads 1 core (In this analogy) but you still only have one mouth. You can only eat one thing at a time. You could have 16 hands (threads) but still only one mouth (cores) so it's only going to get you so far (Which is why we don't see 4 core 16 thread CPU's)

There's a reason why Intel has never marketed threads as cores, because they're not. They know that, we know that.

No one has ever said "To define 8 cores you must have 8 cores that can all process things at the exact same speed" because with that logic, as someone pointed out previously in this thread, intel boosts a single core much higher than the others. Would that then be considered single core since it's much faster in that moment? Of course not. A core is a core, it can execute things. Threads do no executing, the cores do. The Intel i7-1165G7 only has 4 cores. It can only execute 4 things simultaneously.

4 Cores > 4 Core/8 Thread > 8 Cores.

 

[senttoschool]

The M1 is already in two products with the word Pro in their names, so any expressly Pro line of chips after this would be a very muddy distinction. This wouldn't be the first time they make a weird, conflicting decision like this (for example, USB C being their "Pro" port except for on their Pro phones, and M-series chips being for the motion coprocessor before becoming their main chip line) but I feel like they'll want to avoid any extra naming shenanigans wherever they can.

Low-end Macs and iPad Pros can share the same "M" SoCs. I think that's good marketing for iPad Pros because it tells consumers that they can do everything that low-end Macs can do performance-wise. No one reasonably expects the iPad Pro to have the same SoC as the MBP 16".

As for the 13" MBP, I think that's going away for good soon and replaced by a 14" MBP.

 

[boss.king]

Low-end Macs and iPad Pros can share the same "M" SoCs. I think that's good marketing for iPad Pros because it tells consumers that they can do everything that low-end Macs can do performance-wise. No one reasonably expects the iPad Pro to have the same SoC as the MBP 16".

As for the 13" MBP, I think that's going away for good soon and replaced by a 14" MBP.

Again, this is a very arbitrary and muddy distinction if you want them to create a Pro line of CPUs.

 

[Jorbanead]

Low-end Macs and iPad Pros can share the same "M" SoCs. I think that's good marketing for iPad Pros because it tells consumers that they can do everything that low-end Macs can do performance-wise. No one reasonably expects the iPad Pro to have the same SoC as the MBP 16".

As for the 13" MBP, I think that's going away for good soon and replaced by a 14" MBP.

I don’t think that’s an issue here. The i3 and the i9 are very different processors. Same with ryzen 3 and Ryzen 9. Ryzen isn’t associated with “low-end” products. It’s associated with AMD. It’s the stuff that comes after that denote where it sits on the performance ladder (9 is better than 3).

That’s why adding differentiating letters or numbers after the M is what could denote it’s power. In the past Apple did this with an X to denote Xtra cores. Xtra power.

Of course it’s all just marketing and Apple has done some weird stuff with their names so who knows!

Also M1 has additional features added to compared to traditional A#X variants like TB controllers and other Mac-specific hardware. So it’s not exactly an A14X based off of that, but again who cares it’s just marketing names and Apple could call it whatever they want.

 

[boss.king]

That’s why adding differentiating letters or numbers after the M is what could denote it’s power. In the past Apple did this with an X to denote Xtra cores. Xtra power.

Don't forget the 12Z, for that Zxtra GPU core.

 

[leman]

Threads != cores though. The best analogy I can give is that threads are like hands, you are 2 threads 1 core (In this analogy) but you still only have one mouth. You can only eat one thing at a time. You could have 16 hands (threads) but still only one mouth (cores) so it's only going to get you so far (Which is why we don't see 4 core 16 thread CPU's)

Your analogy confuses hardware threads and software threads. And yes, there absolutely are CPUs that can execute more than two hardware threads per core - Power9 for example. Also, GPUs can have hundreds of threads in flight simultaneously and execute them to hide data access latency. This is very similar in a way to modern out of order CPUs that reorder instructions on the go.

Threads do no executing, the cores do. The Intel i7-1165G7 only has 4 cores. It can only execute 4 things simultaneously.

I absolutely assure you that this CPU can execute 8 streams of code simultaneously.
That’s what SMT is all about. Each “core” has two instruction pointers and tracks and executes two instruction streams at the same time. Some Intel CPUs even have doubled instruction decoders. And if you start looking at what does “executing”… oh boy. M1 can execute close to 60 instructions simultaneously, and I mean literally simultaneously.

Again, once you get into the rabbit hole of learning how these things actually work, it’s very difficult to get out. In the end, “core” is whatever the engineers (and in many cases, the marketing department) says it is. It’s what the core does that makes it interesting for comparing different CPUs.

 

[JMacHack]

Didn’t AMD recently lose a class action suit over the definition of a “core”? Iirc it was because two “cores shared a floating point unit, and therefore wasn’t a complete core”

it was over Bulldozer I believe.

 

[leman]

Didn’t AMD recently lose a class action suit over the definition of a “core”? Iirc it was because two “cores shared a floating point unit, and therefore wasn’t a complete core”

it was over Bulldozer I believe.

@cmaier would know, he supposedly designed Bulldozer

 

[cmaier]

@cmaier would know, he supposedly designed Bulldozer

I had absolutely nothing to do with that chip. Aside from the fact that it was essentially designed by the Austin team (if i remember correctly - been a long time) (Sunnyvale did even-numbered chips, and Austin did odd), I achieved a lot of notoriety around here by predicting it would suck and explaining why, only to be proven right when the chip actually went on sale.

My time with AMD last through the hammers, and stopped when when we got to vehicular construction equipment

 

[dogslobber]

Everybody is assuming that the higher end machines will have a beefier single ARM based chip in there. Wherein it's cheaper to shovel in a couple of ARM processors as that solves your memory bandwidth issue per chip and reduces development costs. All you need as a high speed interconnect which PCIe gives you for bandwidth between CPU chips. It's been done many times before.

 

[dogslobber]

It can run 8 hardware threads at once, so yes, it can definitely "truly process" 8 things simultaneously. M1 also supports 8 hardware threads at once.

Cores are independent entities which are effectively mini-chips. Hardware threads are reusing the logic units within the same core during their idle period. This works for like workloads but when the pipeline needs to be flushed then everything stalls until the pipeline get filled again.

 

[KPOM]

It can run 8 hardware threads at once, so yes, it can definitely "truly process" 8 things simultaneously. M1 also supports 8 hardware threads at once.

Hyperthreading is just there because x86 instructions are different length. By feeding 8 threads, it is much more likely that all 4 cores will be used close to the maximum. If you send 8 identical-length instructions to a hyper threaded x86 it will take twice as long as 4 identical-length instructions.

 

[thingstoponder]

[*]This means the low end Macs will not receive latest core designs first each year. It never made much sense for low end Macs to destroy MBPs in single core performance for 9+ months every year.

I don’t think the release time of the first AS Macs is any indication of future ones. Like I really doubt the low end iMac will be updated 6 months after the other low end Macs won’t he same chip every year. Maybe 3 months later at the most in the future. It seems the iMac was delayed, because people like Kuo were saying it would be one of the first day 1 Macs.

I think once all the devices are transitioned and redesigned and there’s no COVID the updates will be much quicker and you won’t have to wait half a year between Macs with the same chip. Although I do think the higher end ones will come after the lower end ones each year. But not 9 months, more like 3-4. I think the Mac Pro will be on its own cycle though and won’t be a yearly thing.

 

[thingstoponder]

I think that it's been fairly obvious that the M1 is just the A14X wearing a mustache.

View attachment 1789427

What I would point out is that Apple did tweak it to include things like virtualization support and optimizations for Rosetta. Those particular capabilities wouldn't have been useful in an iPad Pro, at least in its current incarnation.

The a14 supports virtualization too now. They didn’t add it to the M1 so much as added it to the cores. I agree that the M1 would have existed without the AS transition and would be called A14X although it might not be exactly the same. Maybe they don’t add thunderbolt to it if it’s just an iPad chip. At the least they wouldn’t have put 2 Thunderbolt ports in it instead of 1.

 

[quarkysg]

You could have 16 hands (threads) but still only one mouth (cores) so it's only going to get you so far (Which is why we don't see 4 core 16 thread CPU's)

The Sun SPARC CPU can have from 2 to 32 threads of execution per core. So such CPUs exists.

 

[rui no onna]

the max speeds of a given processor are often limited to single-core performance, and the CPU as a whole has to throttle down under multi-core load to maintain operating temperatures.

Pretty sure the A12X/Z and M1 does this as well.

 

[Jorbanead]

Don't forget the 12Z, for that Zxtra GPU core.

The Z is for Zoooom GPU go fast.

 

[cmaier]

Pretty sure the A12X/Z and M1 does this as well.

I don’t think so. Any evidence for that?

 

[rui no onna]

I don’t think so. Any evidence for that?

Real-world: fanless MacBook Air slower than MacBook Pro during extended load.

The 2018 Apple iPad Pro (11-Inch) Review: Doubling Down On Performance

www.anandtech.com

 

[Ethosik]

I absolutely assure you that this CPU can execute 8 streams of code simultaneously.

A 4 core, 8 thread CPU will not be as efficient as an 8 core, 8 thread CPU - using a piece of software that utilizes 8 cores. The hyperthreading uses context switching, sharing of cache. Hyperthreading ultimately allows running two things on the same core, but those two things are still on that one core. Whereas a full 8 core, 8 thread (non hyperthreaded) CPU will be better due to having 8 full cores available.

Its not appropriate to call a 4 core, 8 thread CPU as an 8 core CPU.

 

[cmaier]

Real-world: fanless MacBook Air slower than MacBook Pro during extended load.

The 2018 Apple iPad Pro (11-Inch) Review: Doubling Down On Performance

www.anandtech.com

View attachment 1789849

Ok, i think you and the person you were responding to were referring to different issues. Full CPU throttling is different than what, for example, Intel does, where if you want to use one core at full clock you need to reduce the clock on all the other cores. One major difference is that with M1, it does NOT have to throttle in machines with fans, whereas, if you had the hot-spotting issue the person you responded to was talking about, you likely would have to.

 

[rui no onna]

Ok, i think you and the person you were responding to were referring to different issues. Full CPU throttling is different than what, for example, Intel does, where if you want to use one core at full clock you need to reduce the clock on all the other cores. One major difference is that with M1, it does NOT have to throttle in machines with fans, whereas, if you had the hot-spotting issue the person you responded to was talking about, you likely would have to.

Iirc, Intel does allow max clocks as long as TDP allows (ever since Haswell?). I know I've seen top burst rates on all cores at times on desktop. It just can't sustain the performance for long. Probably not something you ever see on mobile though.

 

[cmaier]

Iirc, Intel does allow max clocks as long as TDP allows (ever since Haswell?). I know I've seen top burst rates on all cores at times on desktop. It just can't sustain the performance for long. Probably not something you ever see on mobile though.

Yeah, not every intel chip behaves that way - it’s been awhile since i was up-to-date on all that. They even had to migrate threads between cores because they had an issue with local hot spotting.