Discussion Apple Silicon SoC thread

[Eug]

M1
5 nm
Unified memory architecture - LP-DDR4
16 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 12 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache
(Apple claims the 4 high-effiency cores alone perform like a dual-core Intel MacBook Air)

8-core iGPU (but there is a 7-core variant, likely with one inactive core)
128 execution units
Up to 24576 concurrent threads
2.6 Teraflops
82 Gigatexels/s
41 gigapixels/s

16-core neural engine
Secure Enclave
USB 4

Products:
$999 ($899 edu) 13" MacBook Air (fanless) - 18 hour video playback battery life
$699 Mac mini (with fan)
$1299 ($1199 edu) 13" MacBook Pro (with fan) - 20 hour video playback battery life

Memory options 8 GB and 16 GB. No 32 GB option (unless you go Intel).

It should be noted that the M1 chip in these three Macs is the same (aside from GPU core number). Basically, Apple is taking the same approach which these chips as they do the iPhones and iPads. Just one SKU (excluding the X variants), which is the same across all iDevices (aside from maybe slight clock speed differences occasionally).

EDIT:



M1 Pro 8-core CPU (6+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 16-core GPU
M1 Max 10-core CPU (8+2), 24-core GPU
M1 Max 10-core CPU (8+2), 32-core GPU

M1 Pro and M1 Max discussion here:

Page 78 - Discussion - Apple Silicon SoC thread

Page 78 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M1 Ultra discussion here:

Page 109 - Discussion - Apple Silicon SoC thread

Page 109 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M2 discussion here:

Page 127 - Discussion - Apple Silicon SoC thread

Page 127 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Second Generation 5 nm
Unified memory architecture - LPDDR5, up to 24 GB and 100 GB/s
20 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 16 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache

10-core iGPU (but there is an 8-core variant)
3.6 Teraflops

16-core neural engine
Secure Enclave
USB 4

Hardware acceleration for 8K h.264, h.264, ProRes

M3 Family discussion here:

Page 215 - Discussion - Apple Silicon SoC thread

Page 215 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M4 Family discussion here:

Page 263 - Discussion - Apple Silicon SoC thread

Page 263 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[Eug]

Apparently apple is making a 15" macbook Air next year. That may be the day when Apple kills thr macbook pro 13".

The 15" Air should have better battery life than all macbooks and should handle sustained loads better than the 13" Air.

Yeah, if they do release that 15" MacBook Air, a lot of people are going to be very, very happy. And yes I agree, even fanless, it may do better for sustained performance. I'm not 100% convinced the battery life will be better, but you may be right.

(It won't be a machine for me personally though, since I prefer small laptops.)

 

[moinmoin]

I believe you can't set it manually but turning on low power mode in battery settings automatically sets the M1 clock speed to 2ghz and this should lower package power.

That's unfortunate. The tests I do on Renoir allow one to map out efficiency in small steps which is not possible on M1 then. But it's at least a second data point which is better than just one.

 

[repoman27]

The M2 MacBook Air has a very thin heatspreader, albeit wide. It does cover other chips though, like the SSD. No finned heatsink or anything like that. It reminds me of the heatspreaders I'd see on like circa 2000 DVD players, etc., not a computer.

View attachment 64784

In the first pic you can see the heatspreader flipped up, exposing the thermal paste and the SoC underneath.

View attachment 64783

I think part of the kerfuffle with the thermal solution used for the M2 MacBook Air is that a lot of people don't necessarily understand what they're looking at and tend to have deep-seated expectations based solely on what they're familiar with. The M2 MacBook Air is built more like a clamshell iPad than a traditional notebook PC. So we should be thinking about what the thermal solution in an iPad Pro looks like, rather than what you might find in a typical PC.

Your screenshots capture what's going on perfectly. The thermal solution begins with the M2 package itself, where the bare SoC die is topped with thermal interface material and covered by a partial integrated heat spreader. Thermally conductive paste is applied to the top of the IHS and makes contact with the large metal sheet which sits below the logic board in the chassis. The primary function of that metal sheet is to act as the top of the EMI/RFI shield which isolates most of the components on the logic board and prevents interference. The black sheet covering the metal, which is being peeled back in the second photo, looks an awful lot like a pyrolytic graphite sheet to me. PGS can offer up to 5 times the thermal conductivity of copper and 8 times that of aluminum. That black sheet is probably the real heat spreader, which in turn makes contact with a large portion of the bottom cover for the chassis. That bottom cover is a planar sheet of aluminum which extends to nearly the full dimensions of the device and acts as a giant radiator.

A couple other things I noticed while looking at the teardown... In order to reduce z-height, Apple made the PCB in the new MacBook Air entirely single-sided. There are no components on the side of the board that faces up and sits against the bottom of the keyboard tray. It appears to be entirely covered with graphite tape to dissipate heat and prevent hot spots from forming under the keyboard. In order to make the PCB single-sided and still minimize the footprint, Apple had to significantly increase component density on the reverse side of the board. I'm pretty sure they've crossed the line into using substrate-like PCBs for Macs now. This difference, the screen, and the inclusion of MagSafe probably represent most of the $200 cost increase compared to the M1 MacBook Air. A potential side effect of the passive cooling and single-sided PCB is that this might end up being one of the more liquid-damage resistant Macs ever produced.

It also looks like Apple has shifted from using Intel Thunderbolt 4 retimers to either a semi-custom or in-house developed solution. This is quite surprising, because there are now several sources for USB4/Thunderbolt 4 retimers and linear redrivers including long-time Apple suppliers Parade and NXP in addition to the newer startup Kandou. I can't think of any reason why Apple would want to go there for something like a commodity retimer when they could buy off-the-shelf instead.

And finally, looking at the date codes on the chips, many of them were produced in April 2022, however the M2 rolled off the line significantly later in the first week of June. So the delay in the shipping date may have been related to M2 production.

 

[Eug]

The black sheet covering the metal, which is being peeled back in the second photo, looks an awful lot like a pyrolytic graphite sheet to me. PGS can offer up to 5 times the thermal conductivity of copper and 8 times that of aluminum. That black sheet is probably the real heat spreader, which in turn makes contact with a large portion of the bottom cover for the chassis. That bottom cover is a planar sheet of aluminum which extends to nearly the full dimensions of the device and acts as a giant radiator.

I'm not sure, but I don't think that bolded part is correct.

 

[repoman27]

I'm not sure, but I don't think that bolded part is correct.

What makes you say that? I can't imagine Apple would leave that much empty space in a device as thin as this. And if you look at the bottom cover, which is made from machined aluminum, they even went to the trouble of removing a fraction of a mm of material from the area where the M2 and speakers are due to those components being ever so slightly taller.

 

[Heartbreaker]

What makes you say that? I can't imagine Apple would leave that much empty space in a device as thin as this. And if you look at the bottom cover, which is made from machined aluminum, they even went to the trouble of removing a fraction of a mm of material from the area where the M2 and speakers are due to those components being ever so slightly taller.

How are you sure it's making contact with the bottom cover?

If it were, it probably wouldn't throttle. This is what the previously mentioned M1 MBA mod did. Used a thermal pad to make contact with the bottom cover, and ended throttling at the expense of the bottom of the laptop getting too hot.

 

[Eug]

What makes you say that? I can't imagine Apple would leave that much empty space in a device as thin as this. And if you look at the bottom cover, which is made from machined aluminum, they even went to the trouble of removing a fraction of a mm of material from the area where the M2 and speakers are due to those components being ever so slightly taller.

@guidryp explained it.

Neither the fanless 12" MacBook nor the fanless M1 MacBook Air has such contact with the case.

 

[repoman27]

How are you sure it's making contact with the bottom cover?

If it were, it probably wouldn't throttle. This is what the previously mentioned M1 MBA mod did. Used a thermal pad to make contact with the bottom cover, and ended throttling at the expense of the bottom of the laptop getting too hot.

Because you can see the imprint of the bottom cover on the heat spreader in the teardown photos. Also, I understand how valuable interior volume is to Apple and how every tenth of a mm of z-height translates into volume when dealing with an area the size of that heat spreader.

It throttles by design. Apple doesn't implement "turbo". They let the SoC draw as much power as it likes until the temps get out of range for that particular device, then they reduce clock speed. The Apple method will always deliver the maximum performance for a thermally constrained device, which this is.

So sure, you can prevent the SoC from throttling and make the device generally less suitable for most use cases. You can also put the thing in a freezer while running benchmarks if you want to see bigger numbers. I used to do this to get a Mac mini I had to post Geekbench scores that were in the top 5 for all Mac mini models.

@guidryp explained it.

Neither the fanless 12" MacBook nor the fanless M1 MacBook Air has such contact with the case.

Both of those have totally different chassis, logic board, and thermal designs.

 

[Eug]

Because you can see the imprint of the bottom cover on the heat spreader in the teardown photos.

Where? I see no such imprint. Not that there would be much to imprint anyway since it appears to be a flat surface.

Furthermore, one of the cables rests over top of it, between the heat spreader and the bottom case.

 

[repoman27]

The trackpad cable does extend in between the heat spreader and bottom case, but it's also only about 5 hundredths of a mm of black polyimide. The heads of the screws holding down the cable connectors are far more of an issue. The bottom case clearly has an area of reduced thickness where the M2 is located. I might have been on crack when I thought I could see the imprint from that on the heat spreader, but the surface of the heat spreader is definitely not uniform. Although in some of the video shots it also looks more like conventional graphite tape.



Regardless, even if there is a very small gap between the heat spreader and bottom case, the device is pretty much sealed—there's no airflow happening. That air will quickly heat up and the heat will be dissipated through the aluminum bottom case which will still act as a radiator.

 

[Eug]

Regardless, even if there is a very small gap between the heat spreader and bottom case, the device is pretty much sealed—there's no airflow happening. That air will quickly heat up and the heat will be dissipated through the aluminum bottom case which will still act as a radiator.

Yes, that is presumably what happens with the 12" MacBook and M1 MacBook Air as well.

With the 12" MacBook, if I put the unit on a granite counter, it throttles less (4%) than if I put it on a wood table (7%), after about half an hour. However, in both instances, it appears to throttle less than the M2 MacBook Air in terms of percentages.



I was hoping the M2 MacBook Air would throttle in the 10-15% range, just because. However, it's more like 15-25% in Dave2D's tests.

 

[Heartbreaker]

Regardless, even if there is a very small gap between the heat spreader and bottom case, the device is pretty much sealed—there's no airflow happening. That air will quickly heat up and the heat will be dissipated through the aluminum bottom case which will still act as a radiator.

Try using your CPU heatsink with no thermal compound and you will see how even the tiniest air gap kills heat transfer.

The whole point here is to slow down heat transfer, so the heat spreader heats faster, and throttling occurs.

You want throttling to occur, because 30 watts long term would heat the case to scorching levels.

Throttling keeps the case from getting too hot. It's pretty much the only choice with a passive mobile design.

 

[repoman27]

Try using your CPU heatsink with no thermal compound and you will see how even the tiniest air gap kills heat transfer.

The whole point here is to slow down heat transfer, so the heat spreader heats faster, and throttling occurs.

You want throttling to occur, because 30 watts long term would heat the case to scorching levels.

Throttling keeps the case from getting too hot. It's pretty much the only choice with a passive mobile design.

Yeah, air makes a better insulator than conductor, but it's also the fluid primarily responsible for carrying heat away from your heatsink. Once the heat makes it to the fins of your heatsink, the next stop is the air inside the case. Proper contact and excellent thermal conductivity are necessary to wick heat away from point sources present in the silicon as quickly as possible and spread it out over as much surface area as possible.

The point I was trying to make, which seems to be entirely lost here, is that the flimsy piece of metal is not the heatsink, it's just an RFI can. The black sheet of material stuck to it is the actual heat sink. And it doesn't appear that any of the people commenting on Apple's thermal solutions have any idea what a thin sheet of graphite is capable of:

There is no question that there is direct contact throughout the stack from the M2 (and many of the other components) to that point. There is also no question that the bottom case acts as the primary radiator for the device and the top case / keyboard as the secondary. Depending on the ambient temperature and given that the heat is spread out as evenly as possible, that bottom case can probably dissipate a considerable amount of heat without the skin temperature at any given point exceeding ~45º C. You don't want the SoC to throttle until your thermal reservoir is saturated, which in turn could lead to skin temperatures exceeding critical temps.

 

[Heartbreaker]

The point I was trying to make, which seems to be entirely lost here, is that the flimsy piece of metal is not the heatsink, it's just an RFI can. The black sheet of material stuck to it is the actual heat sink. And it doesn't appear that any of the people commenting on Apple's thermal solutions have any idea what a thin sheet of graphite is capable of:

Those demos aren't that impressive. The copper strip looks thicker and would have more thermal mass and will take longer to heat and cool because of that alone. It would have to be sheets of equal thermal mass to really see legitimate differences. Thermal coefficient would have that much effect on it cooling, as that is primarily radiation, which is quite slow.

If that is indeed, PGS, then I would see the heat spreader as both the metal layer and the PGS layer together. The metal adds thermal mass, while the PGS spreads heat more evenly and faster. This would prevent early hot spotting. While it's technically interesting if it is PGS, it doesn't change the purpose. It's a big heat spreader.

There is also no question that the bottom case acts as the primary radiator for the device and the top case / keyboard as the secondary. Depending on the ambient temperature and given that the heat is spread out as evenly as possible, that bottom case can probably dissipate a considerable amount of heat without the skin temperature at any given point exceeding ~45º C. You don't want the SoC to throttle until your thermal reservoir is saturated, which in turn could lead to skin temperatures exceeding critical temps.

Sure, ultimately most of the heat will leave through the bottom of the case. But the whole point I was making, is that there is no physical contact between the heat spreader, and the bottom of the case. You effectively have a small insulating air layer between them. It has to transfer heat between the heat spreader and the case by a much slower radiation/limited air convection method.

You have enough thermal mass in the heat spreader to absorb spikes, but once you do long running work, the heat spreader reaches the design point temperature to throttle. This is essential to keep too much heat from dumping into the case.

 

[repoman27]

I was hoping the M2 MacBook Air would throttle in the 10-15% range, just because. However, it's more like 15-25% in Dave2D's tests.

Well, yeah, but the 2017 MacBooks also had base frequencies ranging from 1.1 to 1.3 GHz... Was your throttling actual thermal throttling to below base clocks, or just consecutive runs being limited to diminishing turbo time?

If it makes you feel any better, Dave2D did 20 runs of R23 vs. your 10 runs of R15, and his MBA was only down 13% compared the the MBP after 10 runs.

 

[Heartbreaker]

If it makes you feel any better, Dave2D did 20 runs of R23 vs. your 10 runs of R15, and his MBA was only down 13% compared the the MBP after 10 runs.

I really don't see that even 20% performance loss in these kinds of tasks really mattering. If you are doing long renders/encodes taking 30 minutes+, you probably aren't watching the clock waiting for it.

 

[Eug]

Was your throttling actual thermal throttling to below base clocks, or just consecutive runs being limited to diminishing turbo time?

I didn't actually monitor that but I presume it's the latter.

If it makes you feel any better, Dave2D did 20 runs of R23 vs. your 10 runs of R15, and his MBA was only down 13% compared the the MBP after 10 runs.

I did 25 runs of R15 but each run takes so long on that machine that it took over an hour.

Plus my result was invalid for the granite test at run 11. I moved the machine over about a foot on the granite counter and the performance went UP. It was because that part of the counter was not pre-heated by the computer and acted as a better heatsink. I later tried doing that again on purpose at run 23, and yep, moving the computer over made the performance go back up.

Since R15 didn't have an auto-loop feature, I had to do the test looping manually, and didn't feel like repeating them, and since 10 tests already took close to half an hour, I decided to limit the test to 10. Plus it was other people testing the i5 and i7. I figured it'd be easier to recruit people to benchmark if it didn't involve an hour of babysitting a computer.

These are my results over the 25 runs on the granite. The two increases are when I moved the computer. Note also that the performance increase extended over two runs, not just one.



In contrast, here are Notebookcheck's results.



Dunno why their results were so much worse. Higher ambient temp? Lemon? They had more throttling and then after a certain point the performance just tanked.

I really don't see that even 20% performance loss in these kinds of tasks really mattering. If you are doing long renders/encodes taking 30 minutes+, you probably aren't watching the clock waiting for it.

You're probably right, but my own personal rule of thumb is that when the performance difference is <15% I don't really care that much. However, when the performance difference is >25% I may care. 15-25% may be a grey zone though. Depends on the activity of course, but like I said, it's just a general rule of thumb.

 

[poke01]

This M2 Air design was exactly an iPad Pro design but it's clamshell based.
What happens when Apple moves to a 3nm based Mx chip?

I think the M2 Air was designed for future chip designs after all Apple's case designs last 4-5 years before redesigning.


With the M2 Air it's clear that Apple wants people who want do sustained loads to move to the Pro Macs.

 

[Doug S]

This M2 Air design was exactly an iPad Pro design but it's clamshell based.
What happens when Apple moves to a 3nm based Mx chip?

I think the M2 Air was designed for future chip designs after all Apple's case designs last 4-5 years before redesigning.
View attachment 64887

With the M2 Air it's clear that Apple wants people who want do sustained loads to move to the Pro Macs.

The M2 Air has an on board accelerometer. Since they abandoned HDDs years ago, why is that in there? I would guess it is because they use the exact same board for the iPad Pro.

If you're using the same SKU for the board you're probably using the same cooling, and like I said before Apple can either clock it lower than the M2 MBP or they can give you the benefit of its default clock for as long as that lasts. Maybe they could design a better cooling solution for the M2 Air but it may be the best they can do for the iPad Pro given the limitations on its form factor and desire to avoid unnecessary weight.

Anyway most people buying Air are casual users who honestly would not notice the difference if an A12Z was swapped in for the M2.

You can get more performance, better display etc. with the MBP plus the option for even more performance with the 'Pro' line, but you have the tradeoff of higher weight, higher price and fan noise. Same type of tradeoffs you have in the PC world where you can buy a laptop with a 'U' series CPU that is lower performance but doesn't need much of a fan, versus the 'H' series that need a hell of a lot of cooling and still can't run full speed very long.

 

[Eug]

You can get more performance, better display etc. with the MBP plus the option for even more performance with the 'Pro' line, but you have the tradeoff of higher weight, higher price and fan noise.

That’s just it. You get a worse display, worse camera, polarizing Touch Bar, worse speakers, and no MagSafe by going with the 13” MacBook Pro.

 

[poke01]

That’s just it. You get a worse display, worse camera, polarizing Touch Bar, worse speakers, and no MagSafe by going with the 13” MacBook Pro.

I think the person meant the real Pro macbook, the 14" model.

 

[MadRat]

It sounds like Apple relies on your lap to cool the frame. Adding an airgap under the machine probably has little benefit, as it may have been designed around minimal airflow.

Maybe they should have used A16 in the 13" Macbook Air. Hide one in middle of lid and other under the chassis. The one in the lid can concentrate on graphics while one in chassis handles main non-graphics workload. Surely two A16 can compete with M2 on both cost and performance, while also being able to sustain max workloads.

 

[igor_kavinski]

It sounds like Apple relies on your lap to cool the frame.

They are doing the world a favor. More lap heat = low sperm count = fewer babies. Even better for the babies who don't come into this messed-up world.

 

[lobz]

They are doing the world a favor. More lap heat = low sperm count = fewer babies. Even better for the babies who don't come into this messed-up world.

That's where I should say I REALLY hope you'll use them on your lap a lot, but I'm afraid the cancel police would think I actually mean it 🤣

 

[repoman27]

The M2 Air has an on board accelerometer. Since they abandoned HDDs years ago, why is that in there? I would guess it is because they use the exact same board for the iPad Pro.

If you're using the same SKU for the board you're probably using the same cooling, and like I said before Apple can either clock it lower than the M2 MBP or they can give you the benefit of its default clock for as long as that lasts. Maybe they could design a better cooling solution for the M2 Air but it may be the best they can do for the iPad Pro given the limitations on its form factor and desire to avoid unnecessary weight.

Anyway most people buying Air are casual users who honestly would not notice the difference if an A12Z was swapped in for the M2.

The accelerometer is for spatial audio. There is not even a remote possibility that Apple could or ever would use the same logic board or thermal solution for the MacBook Air and iPad Pro.

<rant> The M1 in the iPad Air is clocked at 3.2 GHz, same as all the other M1 devices. Did nobody notice that sustained performance for that generation was not the same across the line from iPad Air to Mac mini? Why is this suddenly a thing with the M2? Why does everyone understand that the A15 in the iPhone mini throttles before it does in the iPhone Pro Max or iPads, but loses their mind when it happens in the new MacBook Air? Throttling under sustained load is an intentional design decision by Apple. They are fully capable of making an M2 MacBook that doesn't throttle, and they did—it's called the MacBook Pro. But the design goals for the new MacBook *Air* were silent, thin, and light, not "never throttles". Everyone on the internet seems to believe that the thermal solution in the MacBook Air is somehow deficient, when it's most likely exceptional at meeting Apple's design requirements. </rant>