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

 

[The Hardcard]

SSD's do wear out eventually. The failure rate during the warranty period is likely very low... but I suppose it could be an issue for people buying used Macs.

IIRC you can boot off an external drive.

Given that SSDs typically go 20 to 30 years with even moderately heavy use, I’m not sure how many people are going to be relying on an M1 series MacBook in 2035.

 

[Jan Olšan]

The upside is super obvious. They ALREADY HAD a great NAND controller they'd purchased from Anobit and integrated into iPhone SoCs years ago. It isn't "super complicated" to continue doing what they had been doing for years. It is also cheaper, since they aren't handing money to a third party for a controller when they already have one. Instead of buying finished SSDs that are marked up by someone else, they are buying the same NAND chips they are already buying in bulk for iPhones and iPads.

No, that is not obvious at all.

Where do we know the controller is great from, exactly? Where has it been compared with anything? What evidence is there to base that claim on? Maynard Handley-style faith that stuff from Apple is always class-leading exception and better than the regular thing? (Joking, this is a serious question: I recall Anobit was something along the lines of "promising startup" or something 15 years ago, but was it ever seen and tested in a product before Apple bought them?. And it's *12 years* ago, what seemed to have been hot then may not be today at all. Let's remember that back when Apple bought Anobit, SandForce seemed to be the star of SSDs. Few years later, they were forgotten...



You say it is cheaper for Apple. Well, also not really obvious, isn't that? They may seem to save on not needing dedicated silicon for the controller, but they may need to add a mini-controller silicon to accompany every NAND die/package anyway. They can't use the general comodity NAND chips everybody else uses due to using that funny NVMe-like interface (or did they drop that since M1?) instead of the usual interfaces normally used between NAND and SSD controllers. This could still end up being somewhat cheaper for Apple due to scale, but also might not. Consumer only suffers regardless, anyway.

Also, you missed the other option Apple had - make their own modular SSD with standard NVMe controller (their own in-house, likely DRAMless/HMB, the SoC integrated one deosn't have dedicated DRAM for mapping tables either) in M.2 form factor, then there is not buying finished SSDs marked up by someone else. Except that's good for the customer, can't have that lol.

 

[igor_kavinski]

Given that SSDs typically go 20 to 30 years with even moderately heavy use, I’m not sure how many people are going to be relying on an M1 series MacBook in 2035.

We are not in the MLC era anymore.

That usage can cause a TLC SSD to fail in less than 10 years.

Cheap QLC ones? Pray you never have to get one.

 

[Jan Olšan]

I'll bet you'd have to really work hard to wear out the SSD on those. Anobit claimed they'd figured out a way to extend the life of NAND using complex proprietary signal processing algorithms and they could achieve 2-3x the life that other controllers do, and that was supposedly one of the big reasons Apple bought them (and one would assume you'd have to provide proof of such claims before Apple will hand you a $400 million check)

People keep worrying about this, but I have yet to see a single report of a Mac that's died because its NAND wore out. I think people don't realize how difficult it is to wear out NAND. Sure, you can come up with some sort of use case where you're writing hundreds of MB per second 24x7 that might wear it out, but if you're one of the dozen people who are going to be doing that to a Macbook you might want to look into buying one of those TB4 m.2 enclosures I mentioned above since even if it was possible to replace Apple's SSDs you'd have to open it up (which is hardly convenient with a Macbook) so the enclosure makes sense for multiple reasons (including performance, if the bandwidth of the built in SSD is a limiter at times you can stripe together multiple drives in your enclosure)

Magical proprietary algorithm improving endurance by 2-3x? Where have we heard that, I wonder...

Around that time LPDC ECC was being developed, has come to market idunno now, in 2016? It facilitates an improvement in endurance that was usually given as 3X+ (actually, it doesn't improve endurance, it just makes reliable operation possible with worse error rates after more erase cycles).

Isn't it quite possible the Anobit magic is pretty much a version of that? Every other SSD controller uses LPDC since TLC became common. That improvement was basically needed (and thus in a sense, "consumed") when moving to TLC NAND, you know, so in a sense, it doesn't really make situation better than it was in 2012 when we had MLC NAND drives.

One thing I learned from being a tech journalist for 12 years: never dwell on literal meanings of statements from ages ago, and never take quantitative numbers as an absolute truth forever, always be aware that context evolves too and can change everything.

Stuff like company saying their stuff is 10x better than XY some 10 years ago (not mentioned: everything else also improved by the same or even better amount, too) is a classic pitfall.

Everybody else having the same technology too, but company XY (or even just the fandom...) calling it by a different very fancy name, therefore it has to be something really different and much better? Another one.

 

[LightningZ71]

Maybe it's a combination of intelligent wear leveling, over provisioning in the chips that apple orders, and sticking with TLC, that gives them their effective robustness.

I do know, personally, two people that bought early M1 Mac laptops that had only 8GB RAM and tried to use them pretty heavily. Both were unbootable after about a year and a half. I can only imagine that doing a lot of paging to the SSD took it's toll. It's anecdotal though, so, I can't say that they were all that way.

 

[igor_kavinski]

The only real way to find out if Apple's NAND is better is to find a used M1 Mac Mini (should've around 90% endurance left if it was used normally) and then start thrashing it with constant continuous heavy random 4K I/O and see how many days it lasts. After a few days of regularly checking the value of the life remaining indicator, we can arrive at the rate at which the NAND is degrading and make an educated guess of its possible date of demise. Should be a very interesting experiment to see if Anobit really is innovative technology but I don't want to be the one to sacrifice perfectly good hardware for an experiment, no matter how cool it sounds

 

[igor_kavinski]

Both were unbootable after about a year and a half.

That's roughly a 0.1835% per day decline in the life remaining counter for 256GB. Assuming 5% reserved NAND capacity, that's about 268GB of NAND. Can anyone calculate how much the endurance rating in terabytes written would be from those two values (decline per day and NAND capacity)?

 

[Doug S]

No, that is not obvious at all.

Where do we know the controller is great from, exactly? Where has it been compared with anything? What evidence is there to base that claim on? Maynard Handley-style faith that stuff from Apple is always class-leading exception and better than the regular thing? (Joking, this is a serious question: I recall Anobit was something along the lines of "promising startup" or something 15 years ago, but was it ever seen and tested in a product before Apple bought them?. And it's *12 years* ago, what seemed to have been hot then may not be today at all. Let's remember that back when Apple bought Anobit, SandForce seemed to be the star of SSDs. Few years later, they were forgotten...



You say it is cheaper for Apple. Well, also not really obvious, isn't that? They may seem to save on not needing dedicated silicon for the controller, but they may need to add a mini-controller silicon to accompany every NAND die/package anyway. They can't use the general comodity NAND chips everybody else uses due to using that funny NVMe-like interface (or did they drop that since M1?) instead of the usual interfaces normally used between NAND and SSD controllers. This could still end up being somewhat cheaper for Apple due to scale, but also might not. Consumer only suffers regardless, anyway.

Also, you missed the other option Apple had - make their own modular SSD with standard NVMe controller (their own in-house, likely DRAMless/HMB, the SoC integrated one deosn't have dedicated DRAM for mapping tables either) in M.2 form factor, then there is not buying finished SSDs marked up by someone else. Except that's good for the customer, can't have that lol.

I don't understand why people who are not and never will be Apple customers love to whine and complain about them. If you don't like their stuff don't buy. I won't ever buy anything Samsung makes (actually unrelated to their phones or SSDs) for instance, but you don't see me constantly posting about how their stuff sucks. Get a life.

 

[Doug S]

Maybe it's a combination of intelligent wear leveling, over provisioning in the chips that apple orders, and sticking with TLC, that gives them their effective robustness.

I do know, personally, two people that bought early M1 Mac laptops that had only 8GB RAM and tried to use them pretty heavily. Both were unbootable after about a year and a half. I can only imagine that doing a lot of paging to the SSD took it's toll. It's anecdotal though, so, I can't say that they were all that way.

That doesn't mean the SSD wearing out is what killed them, unless they were monitoring SMART and saw the remaining life decreasing down to zero. Dying after such a short time is a bad thing, but there are other things that can go wrong.

There's no way paging could kill an SSD. It can't write nearly fast enough, and if it could the laptop's overall performance would be so slow as to be unusable.

 

[igor_kavinski]

I don't understand why people who are not and never will be Apple customers love to whine and complain about them.

We whine because Apple is excluding us from being their customers through their anti-consumer practices.

I can't afford a Genoa-X or Turin-X. Do I whine? No. It's for corporations.

I can't afford 18 inch gaming laptops costing $5000. Do I whine? No. Because I can get their performance in a desktop form factor at almost 3.5x less price.

I hear how wonderful using Apple computers is from people who gush about them (mostly in this thread). I want to run some tests on their hardware. Want to use it as rough and as extensively as I use PCs. Can I do that? No! Because I would need to be filthy rich to use expensive hardware, especially since each day I get closer to LOSING that hardware to a landfill. So yes, I believe it's my right to whine. Miracles happen. Someone reports my whining to Tim Cook. He goes, I wonder how many people in total are whining in the world? He gets some lightly loaded team to do an extensive scientific analysis+survey and they come up with a solution to STOP our whining!

I can just hear some of the people laughing their heads off in this thread. Yeah, keep laughing you cruel, heartless people.

 

[igor_kavinski]

There's no way paging could kill an SSD. It can't write nearly fast enough, and if it could the laptop's overall performance would be so slow as to be unusable.

Most of paging is 4K writes. That can kill an SSD pretty fast.

 

[Jan Olšan]

I don't understand why people who are not and never will be Apple customers love to whine and complain about them. If you don't like their stuff don't buy. I won't ever buy anything Samsung makes (actually unrelated to their phones or SSDs) for instance, but you don't see me constantly posting about how their stuff sucks. Get a life.

How was that whining about Apple products?

If people toss around misconceptions not really based on facts or good sources, is it bad when the misconceptions get corrected?
In case we are actually discussing and repeating stuff that is based on nothing but brand cult, isn't that like, really bad?

Point of the comment was: you proposed the Apple design is "obviously" cheaper and the controller is great, but based on what information you believe that, exactly?

Most of paging is 4K writes. That can kill an SSD pretty fast.

Probably 16K on Apple OS.

 

[LightningZ71]

I don't have specifics on which NAND cells that Apple used on the M1 Macbook Airs, but, with 256GB in size, assuming proper wear leveling, and a standard life cycle of about 1000 cycles written before you get an appreciable percentage of failed cells, that's roughly 2.5 Petabytes written to the drive before you achieve that. Divide that by 540 days (about a year and a half) and you get about 4.6Terrabytes written per day. At 8 hours a day of usage, that's under 200MB/sec written.

That's not unachievable for systems that are paging heavily, constantly, for 18 months. Reduce the duty cycle by 1/2 for taking weekends off, vacations, and lunch breaks, that's still 400MB/sec. Still well within the interface capabilities of the chip. But, that's an obscene amount of usage and frankly abusively misusing the device. Remember, 90% of that usage would be from paging due to memory overflow.

 

[Eug]

FWIW...

This the OEM Apple Samsung 128 GB SSD I got in the used 2015 MacBook Pro I bought in 2021:



This is the used OEM Apple Samsung 256 GB SSD I bought to upgrade it:

 

[poke01]

Btw, SSD failure is covered by Apple Care but I don’t have to buy it as there’s really good consumer laws in Australia. Which Apple cannot refuse so in that regard I have no worries.

For people in the States and Asia yeah better buy Apple Care if you are paranoid.

 

[Jan Olšan]

I don't have specifics on which NAND cells that Apple used on the M1 Macbook Airs, but, with 256GB in size, assuming proper wear leveling, and a standard life cycle of about 1000 cycles written before you get an appreciable percentage of failed cells, that's roughly 2.5 Petabytes written to the drive before you achieve that. Divide that by 540 days (about a year and a half) and you get about 4.6Terrabytes written per day. At 8 hours a day of usage, that's under 200MB/sec written.

That's not unachievable for systems that are paging heavily, constantly, for 18 months. Reduce the duty cycle by 1/2 for taking weekends off, vacations, and lunch breaks, that's still 400MB/sec. Still well within the interface capabilities of the chip. But, that's an obscene amount of usage and frankly abusively misusing the device. Remember, 90% of that usage would be from paging due to memory overflow.

After HDDs stopped being used as system storage, developers lost feedback telling them when they overtax the drive by various writes (saving huge browser state periodically...), so rogue apps causing unnatural levels of wear are a risk. I recall spotify being an offender, browsers...

We also don't want to have provisioning for just 1.5 years. Frankly, I have a huge problem with drive killing a computer after 5 years (out of warranty of course). I'm not sure if I would be fine with 10, even, but that may be mainly hobby collector mind speaking, at that point. 200 MB/s may not be that much and probably the amount that apps could inadvertently cause without anybody noticing much, with modern NVMe drives. Also you must not only think about swapping. OS and app components or games keep getting updated all the time, erase cycles are going to get eaten by that. Users fill their drive with downloaded movies, then deleting them and downloading others, or doing that when cutting their video on their small SSDs.

Also it should be noted that Apple (or was it just the evangelists?) often painted the swapping as a huge feature of Macs given how fast and good at substituting in the limited RAM capacity it is supposed to be on their processors + macOS. That story suggests Apple computers would swap more, so when there is simultanous move to fixed storage that can't be easily replaced, it makes huge lot of sense to be proactive about the possible risks. It was presented as a magical solution to RAM capacity, and when you say A, B has to be said too.

 

[igor_kavinski]

But, that's an obscene amount of usage and frankly abusively misusing the device. Remember, 90% of that usage would be from paging due to memory overflow.

Pretty normal for anyone that is used to opening a lot of tabs when researching something in a browser.

 

[igor_kavinski]

Also it should be noted that Apple (or was it just the evangelists?) often painted the swapping as a huge feature of Macs given how fast and good at substituting in the limited RAM capacity it is supposed to be on their processors + macOS.

Only in their world is a major problem (paging) regarded as a "feature".

 

[Eug]

Btw, SSD failure is covered by Apple Care but I don’t have to buy it as there’s really good consumer laws in Australia. Which Apple cannot refuse so in that regard I have no worries.

For people in the States and Asia yeah better buy Apple Care if you are paranoid.

Here in Canada I never buy AppleCare either. My credit card (like many credit cards in the US and elsewhere) includes an additional 1 year warranty on non-refurb products. And in fact, my credit card even includes theft and accidental damage insurance on phones and tablets for 2 years.

Anyhow, since we are not super heavy users, I have no worries about the SSDs. In my limited experience the OEM Mac SSDs have done much better than the spinning HDs in Macs. I also don't worry about the memory either. I've never had OEM memory in a Mac fail. I've sometimes had third party RAM fail though.

I'm not saying Apple NAND and RAM chips are magic, because I know they do fail from time to time, but they just aren't on my main list of worries for Macs. Bigger issues have been stuff like failing keyboards, etc.

 

[The Hardcard]

Pretty normal for anyone that is used to opening a lot of tabs when researching something in a browser.

You have to be trolling. It can’t be possible that you truly believe that web browsing can cause 4.6 TB of swap per day. Not even in a year.

 

[The Hardcard]

Only in their world is a major problem (paging) regarded as a "feature".

Could you provide a link to anyone on the planet experiencing this “major” problem? In the course of reading this thread, I did about 3 dozen pages of Googling and could not find a single reference to anyone who swapped their storage to failure.

I used several searches with various terms, but I’m not perfect. I would appreciate any links or search terms you can provide.

 

[Eug]

You have to be trolling. It can’t be possible that you truly believe that web browsing can cause 4.6 TB of swap per day. Not even in a year.

For that 2015 MacBook Pro I mentioned above, the two SSDs had these stats:

Apple OEM drive for this model that I bought used
256 GB - 97524322 MB writes in entire lifetime of 31540 hours.
That works out to 3092 MB writes per hour, or 23.2 GB per 7.5 hour day.
Assuming a 1900 hour work year, that's about 5.9 TB per year.
Wear levelling index 67%.

Apple OEM drive that came with the Mac
128 GB - 4179231 MB writes in its entire lifetime of (only) 877 hours.
That works out to 4765 MB writes per hour, or 35.7 GB per 7.5 hour day.
Assuming a 1900 hour work year, that's about 9 TB per year.
Wear levelling index 94%.

BTW, I'm not sure why this 2015 model had so few hours on it. However, I suspect it may have been a refurb unit, because the thing looked brand new when I got it in 2021, and when I opened it up to swap in the 256 GB drive, there was no dust in there at all. The battery only had a cycle count of 29 too.

 

[Eug]

The perils of CPU architecture migrations: I think I had an issue with Rosetta 2.

On my M1 Mac mini I'm running PowerPoint 2016 because that's the latest version I have a perpetual Mac licence for. Word 2016 works perfectly for what I need it for, and so does Excel 2016. With PowerPoint 2016 I can create and edit presentations just fine on Apple Silicon, but during presentations sometimes the slides get displayed with no content. If I exit that slide and go back to it one or a few times, then the content comes back, but obviously that's not ideal.

The same presentation works fine in Office 365 for iPad and Office 2016 on Intel Macs running Ventura. The Keynote version of the same presentation also works fine on my iPad Pro M4, my M1 Mac mini, and my Core m3 MacBook.

 

[igor_kavinski]

I'm not saying Apple NAND and RAM chips are magic, because I know they do fail from time to time, but they just aren't on my main list of worries for Macs. Bigger issues have been stuff like failing keyboards, etc.

You opted not to use their magic keyboard and magic mouse so I don't know how you have the gall to complain /s

 

[igor_kavinski]

With PowerPoint 2016 I can create and edit presentations just fine on Apple Silicon, but during presentations sometimes the slides get displayed with no content. If I exit that slide and go back to it one or a few times, then the content comes back, but obviously that's not ideal.

Maybe you can install the latest service pack?