Mac Mini Teardown Provides Real-World Look at M1 Chip on Smaller Logic Board

[President-Elect Beachball]

That’s not my experience in 20 years of computer support (10 professionally).

Let me put it this way, the same type of person that can’t change a spark plug or do an oil change out of fear of “messing something up” is the same person who is afraid of even removing a side panel on a desktop that is CLEARLY designed to be removed.

People are afraid of changing physical items they don’t understand. That’s the vast majority of people. It’s sad because I KNOW the average person is capable of these things, they just feel the need to “let the professionals” handle it.

But the thing is that you don't need a majority of people willing to do upgrades for upgradability to be justifiable. The way I see it manufacturers, and particularly Apple, are going out of their way to BLOCK user upgrades and independent repairs with little or no technical reasons to do so. I don't think a majority of users would upgrade their devices if those upgrades were made easy, just that a much larger and significant minority would if able to do so, certainly in the double digits in percentage terms. And those doing it would be the most money-conscious ones, those who want to take the most out of the devices they purchased.

And the initial discussion was about what Apple would do with greater PCIe bandwidth in the entry-level MacMini. They could add more thunderbolt ports or make storage upgradeable. But the percentage of users who would make use of all thunderbolt connectivity available would also be very small, I guess much smaller than those willing to open a panel to add a m.2 drive. Yet we know which path Apple would choose.

 

[NT1440]

But the thing is that you don't need a majority of people willing to do upgrades for upgradability to be justifiable. The way I see it manufacturers, and particularly Apple, are going out of their way to BLOCK user upgrades and independent repairs with little or no technical reasons to do so. I don't think a majority of users would upgrade their devices if those upgrades were made easy, just that a much larger and significant minority would if able to do so, certainly in the double digits in percentage terms. And those doing it would be the most money-conscious ones, those who want to take the most out of the devices they purchased.

And the initial discussion was about what Apple would do with greater PCIe bandwidth in the entry-level MacMini. They could add more thunderbolt ports or make storage upgradeable. But the percentage of users who would make use of all thunderbolt connectivity available would also be very small, I guess much smaller than those willing to open a panel to add a m.2 drive. Yet we know which path Apple would choose.

I can see a storage upgrade being allowed. But as far as RAM goes I’m pretty sure any physically slotted memory would cause a considerable degradation in performance here...that’s something I don’t think you can pull off gracefully with an SoC unless you had some sort of controller feeding into redundant RAM on the chip itself.

That’s my caveman understanding. I’d love if someone like Cmaier could chime in.

 

[poiihy]

More empty space inside means more potential for modifications 😄

 

[deconstruct60]

I am so excited to see what is inside my new Apple Silicon Mac mini! I am especially happy to learn that the M1 Mac mini has 2 Thunderbolt 3 controller chips (one for each Thunderbolt/USB4 port) so they do not need to share bandwidth, nor reduce device speed, between the 2 Thunderbolt 3 40Gbps ports.

No, it doesn't. Those are noted to be ".. JHL8040R controllers ..." in the egp.io article. The 'R' there is extremely significant. Those are not controllers. Those are re-timers.


https://ark.intel.com/content/www/u...251/intel-jhl8040r-thunderbolt-4-retimer.html

Can see that this is a 0.75W solution. It isn't doing much heavy duty signal processing.


Intel's discrete Thunderbolt 4 control is a substantively different product.

https://ark.intel.com/content/www/u...2/intel-jhl8440-thunderbolt-4-controller.html

In terms of costs (~ $10) , size ( 10x10 ) , and power consumed. ( 2.4 W ).


There is distance limit of how far away the thunderbolt controller can be away from the actual physical ports. For Intel gen 10 (Ice Lake ) U processors (and gen11 with TBv4) with embedded controllers and Apple's M1 that is too far. So the signal has to be boosted and "re-timed" to get it to the PHYS port. That what they are. And yes since each port has to be re-timed/boosted then would be one of these per port.

So for example Intel's schematic of their solutions in general is.

Examining Intel's Ice Lake Processors: Taking a Bite of the Sunny Cove Microarchitecture

www.anandtech.com

The "retimer" on the diagram above is what we are looking at here on the M1 Mini board. It takes Thunderbolt as a input. (or one of the alternative modes or just plain USB 3.x ... whatever the TB controller sends it. )



All of that says nothing about the backside bus that actually feeds data into the actual TB controller (which is embedded/integrated inside the SoC/CPU package). It is not surprising that Apple would use Intel's re-timers with their own controller because I don't think anyone else ( ASMedia or folks moving to discrete TB controllers ) has done a embedded in processor TB controller. AMD isn't ( or at least is pretty far from getting to one to market. )

I have a discrete TB controller you can just place it within an inch or so of the port and there is little (or no) need for a re-timer at all. That's why nobody else is making them in volume now. Their solutions will not need them when they start to arrive.

There is likely not PCI-e v3 being feed to these retimers at all. Nor is there any "raw" DP being fed to them either. Apple's limitation of just one display coming out of the TB controller complex is very suggestive that they are only putting in one DPv1.4 input. The re-timers can't 'fix' that limitation. Nor could they rebalance the DP stream from port 1 to port 2 .


Apple didn't "double up the bandwidth" here. This is still largely an iPad Pro A14X processor with die and package size limitations. Be happy got the 4 TB muxed lanes out. Apple didn't pump out the raw inputs to the TB controll as that would more than double the number of pins out of the package.

 

[antipodean]

No, it doesn't. Those are noted to be ".. JHL8040R controllers ..." in the egp.io article. The 'R' there is extremely significant. Those are not controllers. Those are re-timers.


https://ark.intel.com/content/www/u...251/intel-jhl8040r-thunderbolt-4-retimer.html

Can see that this is a 0.75W solution. It isn't doing much heavy duty signal processing.


Intel's discrete Thunderbolt 4 control is a substantively different product.

https://ark.intel.com/content/www/u...2/intel-jhl8440-thunderbolt-4-controller.html

In terms of costs (~ $10) , size ( 10x10 ) , and power consumed. ( 2.4 W ).


There is distance limit of how far away the thunderbolt controller can be away from the actual physical ports. For Intel gen 10 (Ice Lake ) U processors (and gen11 with TBv4) with embedded controllers and Apple's M1 that is too far. So the signal has to be boosted and "re-timed" to get it to the PHYS port. That what they are. And yes since each port has to be re-timed/boosted then would be one of these per port.

So for example Intel's schematic of their solutions in general is.

Examining Intel's Ice Lake Processors: Taking a Bite of the Sunny Cove Microarchitecture

www.anandtech.com

The "retimer" on the diagram above is what we are looking at here on the M1 Mini board. It takes Thunderbolt as a input. (or one of the alternative modes or just plain USB 3.x ... whatever the TB controller sends it. )



All of that says nothing about the backside bus that actually feeds data into the actual TB controller (which is embedded/integrated inside the SoC/CPU package). It is not surprising that Apple would use Intel's re-timers with their own controller because I don't think anyone else ( ASMedia or folks moving to discrete TB controllers ) has done a embedded in processor TB controller. AMD isn't ( or at least is pretty far from getting to one to market. )

I have a discrete TB controller you can just place it within an inch or so of the port and there is little (or no) need for a re-timer at all.

There is likely not PCI-e v3 being feed to these retimers at all. Nor is there any "raw" DP being fed to them either. Apple's limitation of just one display coming out of the TB controller complex is very suggestive that they are only putting in one DPv1.4 input. The re-timers can't 'fix' that limitation. Nor could they rebalance the DP stream from port 1 to port 2 .

Thank you for your informative post. I’ve been told the System Report on an M1 mac reports two TB busses.

Does that mean the M1 can send 40 GB/s to each port, but DP could only go to one port at a time?

 

[deconstruct60]

Now THIS is an interesting observation. As far as i’m aware the 2018 Mac mini has 4 ports but only two controllers, so it’s essentially already daisy chained. That would mean you’d get the same performance from this Mac mini as you would from the 4 port Mac mini.

These are not Thunderbolt controllers. They are Thundebolt retimers.
You need a retimer for each port to boost the signal. Both ports are just as far from the SoC ( which is "too far". )

For the other USB 3.x Type-A ports they probably don't need a retimer as it is not as distance sensitive. USB 2.0 certainly is not distance sensitive ( at these span the logic board distances).

You are not going to get the same overall aggregate performance out of the ports at all. Apple can feed the two controllers on a Mini 2018 with two x4 PCI-e lanes from the CPU. This Apple embedded TB controller is likely only being feed just one x4 worth of bandwidth. If Apple skimped on DPv1.4 data feeds they are very likely skimping on x4 feeds also. ( both as cheaper and save die space and transistor budget for other stuff. )

Apple's "bigger" M1-midsize variant probably has a bigger transistor budget and doesn't have to constrain itself small enough in size to fit in an iPad Pro. Probably two x4 PCI-e feeds there to two controllers. (and will need 2 more retimers to drive the two more ports. ). Coin toss whether they 'fix' the under provision on DPv1.4 streams. ( I suspect not, but depends upon how much more GPU display output subsystem transistor budget grows and how much memory bandwidth they add. If it is just two more LPDDR4 or LPDDR5 packages then I suspect the monitor limitation will only grow by one . (not jump up to four output total). It still will not pass certification for Thunderbolt v4. )

 

[mr.steevo]

Eh, that's unlikely. The power cable for the Minis is going to stay with the Minis. It's just not the same. (Although I say this as someone who has a couple leftover from my old mini and dead AppleTV 2.)

I get the change on the iPhones. At this point, I keep the chargers in the original boxes when I upgrade. I have so many USB-A power adapters since forever... plus I've got power bricks with them and even a few of the outlets in my house have been changed out. It's become such a standard that everyone has it at this point. (Heck, I even have a couple pairs of EarPods in the original packaging in my "random cables" bin.)

I mean, I actually like that Apple is reducing this stuff. What I dislike is that their commitment to environmental issues is a marking ploy more than anything else. It's still a kinda/sorta win because less plastic waste is a Good Thing™ in my opinion. However, I wish Apple had a more holistic/legitimate approach to environmental consciousness. They're definitely leaps and bounds over Amazon, but... their stance on repairs, for instance, is very anti-environment.

I get why RAM is part of the SoC—there's a legitimate case to be made for performance and power consumption. However, I don't get why the SSDs are glued/soldered in. Same with batteries. It's ridiculous. It takes some nasty solvents to remove that stuff for repair and often they don't even repair, they just replace.

Damn.

You killed my joke.

 

[Jimmy James]

That’s a good theory but I think it just boiled down to trade offs. Do you want to do the extra engineering to let the user buy aftermarket memory that may make the experience worse? Or let the user decide at point of purchase how much memory they’ll want.
Buying a system only to upgrade later doesn’t hold as much value as the risks down the road. But I also think a desktop like the mini or iMac should have user upgradable memory.

I’m using a 2008 MBP that’s only viable because I upgraded parts multiple times. That said, there is no longer the same need for ever increasing ram. The thing that will end-of-life this machine prematurely is likely to be failure or planned obsolescence.

 

[Jimmy James]

What is the expected life of the SSD? I run a MacMini from late 2009. I did a user upgrade on both RAM and HDD. It doesn’t run Catalina or Big Sur but still works fine. I just ordered an M1 version but can’t see it lasting as long if I can’t replace the soldered on storage.

We have 2 old computers that only remain viable because of hdd replacements. I’d hate to toss a computer after even 7 years due to failure. I’d rather repurpose if I’m not using it as a primary machine.

 

[Jimmy James]

Only a fraction? If there is a computer that can alway benefit from expandable storage that's a desktop computer. Heck, they could even engineer nifty little doors that even a toddler could use to install the drives. Why would anyone plonk big money on clunky external thunderbolt storage if they could do just that?

I’m planning to turn this into a family computer with more than one user. With the price of Apple storage it seems frequent on this site that people use external storage. We’ll have a mess of drives and cables on the desk.

 

[sakurarain]

i am excited to see a all-new-designed Mac mini with half size.

 

[Zorori]

They couldn't make a smaller enclosure for this? Really?

Yeah, that's a pretty small board for the case size. I reckon you could easily mod in a bigger fan (if this throttles at all).

This does leave the Mini in an interesting place for future iterations:

* A smaller Mac Mini? Mac Nano?

* A more powerful Mac Mini? Mac Mini Pro? Basically what reviews like Linus want -- something for server rooms with higher specs? Keep the case, upgrade the cooling and use the chip for the more powerful MacBooks.

 

[MacBH928]

Wow...look at all that empty space. I wonder how many uses could have been for it like adding extra SSDs or a BD drive, a dedicated gpu!

People who say no reason to make it smaller, actually there is a lot of reason. What if it can be so small you can hang it behind your monitor, or tv, or lay it above PS5. Imagine how smaller a server rack would be. After all, Apple is king of thinner and lighter!

I am surprised many people are jumping on the Mac Mini. My understanding this is an entry model for very basic needs and stuff like HTPC. Its like an elementary school lab computer. Somehow people seem to choose it as their work machine. Even Apple won't recommend it as that but whatever floats your boat!

 

[Abazigal]

Yeah, that's a pretty small board for the case size. I reckon you could easily mod in a bigger fan (if this throttles at all).

This does leave the Mini in an interesting place for future iterations:

* A smaller Mac Mini? Mac Nano?

* A more powerful Mac Mini? Mac Mini Pro? Basically what reviews like Linus want -- something for server rooms with higher specs? Keep the case, upgrade the cooling and use the chip for the more powerful MacBooks.

A Mac mini with the form factor of the Apple TV has always been a pipe dream of mine.

 

[theorist9]

Note that the Intel Mac Mini features the same top-end processor as the top-end 21.5" iMac (i7-8700B), which has a 65W TDP. Thus, with its current form-factor, it should be able to accommodate a higher-powered M-series processor. I suspect many would prefer if they kept the same form factor, and offered the expected higher-powered processor as an option.

But I haven't thought about the embedded product market....

 

[deconstruct60]

Thank you for your informative post. I’ve been told the System Report on an M1 mac reports two TB busses.

Does that mean the M1 can send 40 GB/s to each port, but DP could only go to one port at a time?

The re-timers say nothing about how much bandwidth goes to each port that originates inside the rest of the M1 complex. The controller design sets that limit.

The re-timers means that data could transfer from port 1 to port 2 at 40 Gb/s. That is substantively different, bandwidth ( That is thunderbolt network bandwidth). That really isn't much different than what happens in TBv3 ( and previous. ) . There is a difference between "freeway/highway" speeds and "on/off ramp" speeds. From inside that is those are the "on/off ramp" speeds and that is grounded in the inputs to the TB controller ( PCI-e lanes , DP version ) . The speed at which data travels down the TB cables is the "freeway/highway" speed. ramp and freeway don't have to be the same speed (and often are not).


And yes if there is only one DP stream fed into the TB controller inside the M1 , then only one is coming out.

First, the retimers only take what the controller feeds them. If the controller only has one feed than it can only send that one feed to just one of the re-timers. ( The re-timers aren't the limitation. )

Second, There has to be multiple inputs for there to be multiple outputs. There is is a corner case if just mirroring the screen in which the single stream could be "duplicated" ( just copy the same DP data to both ports), but I don't think that is a Thunderbolt standard feature. ( if there were multiple inputs then the M1 systems would not have the one display limitation. They do. So it is quite unlikely that two streams as being feed to the Apple TB controllers. With DPv1.4 (or v1.3) a single compressed stream can feed a 6K display like the XDR (presuming the display can decode the compression.. which the XDR does). Pretty good chance Apple nixed multiple streams because v1.4 says newer hi-res monitors can 'do it all" with just one. )

The DisplayPort standard has some features where can put two logical streams onto a single data stream. I don't think Thunderbolt data conversion process interacts with that well. ( the conversion to TB stream format is multiplexing. Multiplexing on top of multiplexing is at minimally awkward and likely computationally expensive. ) At least, in the context to send those two streams to different destinations downstream. Hence, the one display limitation on the M1 system's TB ports.

Maybe multiple work in daisy chain monitors in alternative mode DP. (take Thunderbolt out of the loop). If Apple implemented the full DPv1.4. ( I'm pretty skeptical that they did though. If they did then the "tech specs" are inaccurate. That wouldn't be the first time. And would involve monitors that Apple isn't particularly keen to sell. ).

 

[deconstruct60]

Note that the Intel Mac Mini features the same top-end processor as the top-end 21.5" iMac (i7-8700B), which has a 65W TDP. Thus, with its current form-factor, it should be able to accommodate a higher-powered M-series processor. I suspect many would prefer if they kept the same form factor, and offered the expected higher-powered processor as an option.

But I haven't thought about the embedded product market....

Well note that the 21-24" iMac doesn't have a higher end M-series process now either.

All Apple had at the moment is the M1. Apple also only dumped the i3 version of the 2018 Mini. The higher end ones are kept.

There is pretty good chance that when the iMac 21-24 and MBP 16" and four port MBP 13-14" comes that this same case of the Mini will come with new options and an incrementally bigger, higher TDP logic board. This case will be less of a mismatch at that point. ( Mini gets it 4 TB/USB4 ports back and perhaps gets the 10GbE port option back as well if Apple is overlapping with the iMac 21-24" and moving its Ethernet feature out of the previous century ( 1GgE standardized in 1999 ). Or at least 2.5 or 5GbE. 1 GbE is twenty year old tech. Pretty clear we can do better now on > $1K systems in the post 2020 era.

I would expect Apple to save money by using the same case for the two sized motherboard. Similar to the non-Retina and Retina iMac having about the same case. And iMac and iMac Pro with extreme high overlap in industrial design. It is just cheaper for Apple to keep the rest of the case the same and swap out the back panel - logical board assembly that slides into the same case.

Some folks want Apple to chase after the Raspberry Pi like smaller form factor ( or perhaps stuff it into a AppleTV like case ) . I don't think they are likely to go there. If the $699 Mini is saving costs by coasting on MBA/MBP13" two port (and iPad Pro) for parts then just as likely coasting on "bigger' Mac Mini case to save coasts there too. It is about cranking fatter margins on the lower price point.

 

[CJ Dorschel]

I think more to the point is that back in the day even people who were not computer-heads were still open to upgrading their devices.

Computer is sluggish? Here, bam, some more ram, easy-peasy. Hard drive is full? Here, behind this little detachable panel, replace the hard drive with a larger one. Not so easy as you'd still have to backup and reinstall the OS, but easy enough.

Today: computer is sluggish? Hard drive is too small? Buy a new device and shove the old one into a drawer.

Exactly! Thank you. I tend to get “wordy” in my comments as I often anticipate miscommunication.

 

[Anonymous Freak]

They couldn't make a smaller enclosure for this? Really?

They could - but like with the Intel transition, they're starting with familiarity - keep things looking the same so people get something they're used to.

Heck, an AppleTV 4K size device would probably work for the Mac mini if they really wanted. It would just need a noisier fan.

I expect that we'll see the bifurcation of the mini in to "consumer mini" that does shrink in size and keeps the 20W thermal limit and a "pro mini" to replace the still-offered Intel mini that stays the same size, and has more thermal headroom. (It is kind of silly that the M1 mini has an internal 150W power supply, when even under max CPU+GPU load, WHILE supplying maximum power to all Thunderbolt+USB ports, maximum draw is 70 Watts.)

 

[cult hero]

Damn.

You killed my joke.

Haha. I realized that as I got into typing it.

Sorry about that.

The environmental two-faced stuff from Apple is a really sore spot for me. (As you probably noticed, lol.)

 

[athletejmv]

They couldn't make a smaller enclosure for this? Really?

You would think with all the hype about protecting the environment with not including the charging brick for the iPhone 12, reducing packing size, etc they would have been considerate with the Mac mini as well... oh, ya, bottom line ($) always comes first. On a serious note, the M1 in the Mac mini is insanely mind blowing. I'll be selling my 2019 MBP16". I also have 8gb of RAM in the mini that performs better than the 16Gb in the MBP... dumbfounded at this point.I planed on ordering the 16GB (wait time is 1+ months out), but don't see the point now.

 

[poematik13]

I thought the MacMini could drive two monitors??? Somewhere I read three, but I’m having a hard time believing that is true.

You're right, I think it's two. The macbook air/pro can only do 1 external