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]

https://twitter.com/x/status/1919180817160646666

Interesting comment. Perhaps the 17 series won’t have N2.

Wasn’t that expected to be N3P? N2 always seemed like an iPhone 18 thing based on timing.

At least in my limited understanding of the rumour mill, N2 was never in the cards for Apple A19 / iPhone 17 Pro, as the timing did not align. It was N3P all along. Despite this, there were some dubious claims by some less well known Chinese supply chain leakers that said it would be N2.

 

[jdubs03]

At least in my limited understanding of the rumour mill, N2 was never in the cards for Apple A19 / iPhone 17 Pro, as the timing did not align. It was N3B all along. Despite this, there were some dubious claims by some less well known Chinese supply chain leakers that said it would be N2.

N3P* And there were definitely competing rumors that gave some possibility it was N2. But indeed now that seems very unlikely. So it goes back to my original statement from a few months back: It’s a good opportunity to innovate on the uarch side. They’ll need too if they want to keep a 15% improvement from year to year.

 

[Eug]

N3P*

Thx, fixed. Doh!

 

[Doug S]

At least in my limited understanding of the rumour mill, N2 was never in the cards for Apple A19 / iPhone 17 Pro, as the timing did not align. It was N3P all along. Despite this, there were some dubious claims by some less well known Chinese supply chain leakers that said it would be N2.

Not sure why people are acting surprised about EPYC being the first tapeout. The verification cycle for a chip going in a high end enterprise server is significantly longer than one going in a phone or Macbook Air. The only chance for Apple to use N2 for this year's iPhone would if they were running risk production wafers. That would be more expensive and is probably unnecessary from a sales perspective since Apple mostly competes with itself.

 

[smalM]

There is zero chance for Apple to use N2 this year for anything outside a developement center.
Why do we have this useless discussion at every new major node?

 

[Mopetar]

Why do we have this useless discussion at every new major node?

Is there some other useless discussion you'd prefer we had instead?

 

[poke01]

We got new codenames for M6 and M7.

M6 - Komodo
M7 - Borneo

And Mark Gurman also states an advanced Mac related chip under the name Sotra is under development.

We know Dylan Patel already leaked Sotra and said it was based on N3P. So this chip will be using M5 based IP.

My guess is it’s for the Mac Pro.

Apple working on M6 & M7 chips alongside new AI server processor

While Apple's M5 chip is expected to debut on Mac and iPad, in late 2025, the company's M6 and M7 chips are in the works, in conjunction with a new AI server farm processor.

appleinsider.com

 

[mvprod123]

We got new codenames for M6 and M7.

M6 - Komodo
M7 - Borneo

And Mark Gurman also states an advanced Mac related chip under the name Sotra is under development.

We know Dylan Patel already leaked Sotra and said it was based on N3P. So this chip will be using M5 based IP.

My guess is it’s for the Mac Pro.

Apple working on M6 & M7 chips alongside new AI server processor

While Apple's M5 chip is expected to debut on Mac and iPad, in late 2025, the company's M6 and M7 chips are in the works, in conjunction with a new AI server farm processor.

appleinsider.com

"The iPhone maker is said to be experimenting with chips that have two, four, or even six times as many CPU and GPU cores as the high-end M3 Ultra. If implemented, the new AI chip would likely improve Apple Intelligence features, making them more powerful and competitive."

Mx Extreme is still alive.

 

[Eug]

M4 MacBook Air indoors running looped 3D Mark, in the sun vs in the shade.

 

[jdubs03]

"The iPhone maker is said to be experimenting with chips that have two, four, or even six times as many CPU and GPU cores as the high-end M3 Ultra. If implemented, the new AI chip would likely improve Apple Intelligence features, making them more powerful and competitive."

Mx Extreme is still alive.

Long live THE EXTREME!

 

[name99]

"The iPhone maker is said to be experimenting with chips that have two, four, or even six times as many CPU and GPU cores as the high-end M3 Ultra. If implemented, the new AI chip would likely improve Apple Intelligence features, making them more powerful and competitive."

Mx Extreme is still alive.

Oh, people have no IDEA how ambitious Apple's plans are!

What we're seeing right now is a repeat of the GPU and the modem stories? Remember those: "it's impossible for Apple to create a competitive GPU/modem because <mumble mumble I don't understand how they work so obviously no-one else can either>"?
We're going to hear that same sort of "logic" applied to Apple and large systems.

Meanwhile in the real world look at this beauty!

US20250094093A1 - Optics-Based Distributed Unified Memory System - Google Patents

Techniques are disclosed relating to computing systems that use silicon photonics. In some embodiments, a computing system includes a plurality of compute die packages that include processors configured to execute program instructions that operate on data stored in a distributed memory...

patents.google.com

You don't even have to read much, just look at the diagrams! They just keep going and going, each one a successively higher level of insanity!

Will it ever be manufactured? Well, Apple can cancel plans (eg car). But this sort of thing seems pretty existential for the company. So...

 

[name99]

There is zero chance for Apple to use N2 this year for anything outside a developement center.
Why do we have this useless discussion at every new major node?

"Anything" is a strong term. For example, could Apple use N2 for the M5 and/or R2, to be released in Vision Pro before the end of the year?

Sure, why not? Volumes are low enough that TSMC capacity is not an issue, while there's also value in reminding people that Vision Pro isn't dead (and maybe even in getting them out before Christmas and Chinese New Year).

 

[Doug S]

We got new codenames for M6 and M7.

M6 - Komodo
M7 - Borneo

And Mark Gurman also states an advanced Mac related chip under the name Sotra is under development.

We know Dylan Patel already leaked Sotra and said it was based on N3P. So this chip will be using M5 based IP.

My guess is it’s for the Mac Pro.

Or its for their AI server farm.

I think it is almost certain those chips with "6x" as many cores are destined for internal use in servers. I can't see Apple building something that's got 6x more cores than M3 Ultra for consumers. It just seems like too niche of a market for Apple to try to win against the entrenched Linux/Windows ownership of the high end workstations that currently are rocking a socket or two of Threadrippers and Xeons. Lot of dollars per system in that market, sure, but you need a lot of hands on salesmanship, people won't show up at the Apple Store and order a couple dozen. Does Apple even have old school technical sales reps, or just the ones with the more edu slant for selling to schools and universities?

 

[poke01]

Or its for their AI server farm.

I think it is almost certain those chips with "6x" as many cores are destined for internal use in servers. I can't see Apple building something that's got 6x more cores than M3 Ultra for consumers. It just seems like too niche of a market for Apple to try to win against the entrenched Linux/Windows ownership of the high end workstations that currently are rocking a socket or two of Threadrippers and Xeons. Lot of dollars per system in that market, sure, but you need a lot of hands on salesmanship, people won't show up at the Apple Store and order a couple dozen. Does Apple even have old school technical sales reps, or just the ones with the more edu slant for selling to schools and universities?

no, the server AI chip codename is "Baltra".
Baltra is one that Mark states will be up 6x Mx Ultra

Sotra is for Mac and will come to Mac if its not cancelled. Sotra is most likely 2 Mx Ultras.

 

[mvprod123]

Oh, people have no IDEA how ambitious Apple's plans are!

What we're seeing right now is a repeat of the GPU and the modem stories? Remember those: "it's impossible for Apple to create a competitive GPU/modem because <mumble mumble I don't understand how they work so obviously no-one else can either>"?
We're going to hear that same sort of "logic" applied to Apple and large systems.

Meanwhile in the real world look at this beauty!

US20250094093A1 - Optics-Based Distributed Unified Memory System - Google Patents

Techniques are disclosed relating to computing systems that use silicon photonics. In some embodiments, a computing system includes a plurality of compute die packages that include processors configured to execute program instructions that operate on data stored in a distributed memory...

patents.google.com

You don't even have to read much, just look at the diagrams! They just keep going and going, each one a successively higher level of insanity!

Will it ever be manufactured? Well, Apple can cancel plans (eg car). But this sort of thing seems pretty existential for the company. So...

"In some embodiments, node 700 includes 4096 LP5X channels, which can provide up to 69,632 TB/s of memory bandwidth (at 8.5 GT/s) and up to 32 TB of memory capacity (16 Gb density, byte-mode, dual-rank). In some embodiments, node 700 also includes a 64 GB SRAM-based memory cache. Assuming each compute die package 400 includes an 18-core CPU and a 40-FSTP GPU, node 700 can include 128 compute die packages 400 with a total of 2304 CPU cores and a 5120-FSTP GPU. For completeness, as each memory die package 100 may support up to 5 TB/s of optical bandwidth, node 700 may support up to 320 TB/s of optical bandwidth.
[0043]
Also note that each compute die package 400 can observe the same latency and bandwidth characteristics to main memory. In doing so, node 700, in essence, is the largest UMA machine ever to have been designed.

Brutal🤯

"18-core CPU and a 40-FSTP GPU"

M5/6 Max die?

Are there any other interesting patents that hint at future architectural solutions in CPU and GPU?

 

[Doug S]

"In some embodiments, node 700 includes 4096 LP5X channels, which can provide up to 69,632 TB/s of memory bandwidth (at 8.5 GT/s) and up to 32 TB of memory capacity (16 Gb density, byte-mode, dual-rank). In some embodiments, node 700 also includes a 64 GB SRAM-based memory cache. Assuming each compute die package 400 includes an 18-core CPU and a 40-FSTP GPU, node 700 can include 128 compute die packages 400 with a total of 2304 CPU cores and a 5120-FSTP GPU. For completeness, as each memory die package 100 may support up to 5 TB/s of optical bandwidth, node 700 may support up to 320 TB/s of optical bandwidth.
[0043]
Also note that each compute die package 400 can observe the same latency and bandwidth characteristics to main memory. In doing so, node 700, in essence, is the largest UMA machine ever to have been designed.

Introducing the new Mac Hyper Nova Pro, preorder now starting at $999,999.

 

[name99]

Or its for their AI server farm.

I think it is almost certain those chips with "6x" as many cores are destined for internal use in servers. I can't see Apple building something that's got 6x more cores than M3 Ultra for consumers. It just seems like too niche of a market for Apple to try to win against the entrenched Linux/Windows ownership of the high end workstations that currently are rocking a socket or two of Threadrippers and Xeons. Lot of dollars per system in that market, sure, but you need a lot of hands on salesmanship, people won't show up at the Apple Store and order a couple dozen. Does Apple even have old school technical sales reps, or just the ones with the more edu slant for selling to schools and universities?

Of course they are for servers!

I don't get why people seem to feel that's not a legitimate target. How many DGX's is nVidia selling to consumers?

 

[name99]

"In some embodiments, node 700 includes 4096 LP5X channels, which can provide up to 69,632 TB/s of memory bandwidth (at 8.5 GT/s) and up to 32 TB of memory capacity (16 Gb density, byte-mode, dual-rank). In some embodiments, node 700 also includes a 64 GB SRAM-based memory cache. Assuming each compute die package 400 includes an 18-core CPU and a 40-FSTP GPU, node 700 can include 128 compute die packages 400 with a total of 2304 CPU cores and a 5120-FSTP GPU. For completeness, as each memory die package 100 may support up to 5 TB/s of optical bandwidth, node 700 may support up to 320 TB/s of optical bandwidth.
[0043]
Also note that each compute die package 400 can observe the same latency and bandwidth characteristics to main memory. In doing so, node 700, in essence, is the largest UMA machine ever to have been designed.

Brutal🤯

"18-core CPU and a 40-FSTP GPU"

M5/6 Max die?

Are there any other interesting patents that hint at future architectural solutions in CPU and GPU?

Of course there are. I've discussed many of them in my PDFs.
The big problem is that as you get further from existing hardware, it becomes harder to judge how "real" something is.
Slight tweak to what goes in an iPhone? Of course that will happen soon.
Apple version of something they have never sold before? Who knows when that will ship, or if they will even tell us that something like that has been installed?

Even some basic things are not obvious. Apple has a lot of patents in the field of manipulating point space clouds (basically constructing a 3D scene from stereo camera or lidar, though there are many variants and many aspects to it). What's this for?
Something something Vision Pro? Sure, sounds plausible. But what EXACTLY...

 

[jpiniero]

Of course they are for servers!

I don't get why people seem to feel that's not a legitimate target. How many DGX's is nVidia selling to consumers?

Apple isn't getting back into servers.

Maybe they will use internally but that's about it. Maybe AWS.

 

[adroc_thurston]

Apple isn't getting back into servers.

I hope they try. It would be Very Funny.

 

[Mopetar]

Apple isn't getting back into servers.

Maybe they will use internally but that's about it. Maybe AWS.

They don't really need to. The people who really want them will buy hundreds of minis and find some way to stick them all in a rack. Anyone who wants or needs a beefy single box solution can buy a Pro and do the same.

The bigger limiting factor is the software ecosystem. There's a lot of stuff that's x86 code and even if Apple can emulate it, the performance hit is too large to make it cost effective. Your customer pool is more limited and only the largest providers could take on that kind of niche to buy expensive Apple servers as opposed to just buying hundreds of Minis.

 

[oak8292]

I hope they try. It would be Very Funny.

Home cooking? Apple may be buying enough servers annually to build their own. They have 8 data server centers and they are building two more and will need to keep these up to date every 4-6 years depending on energy efficiency improvements.

Apple’s Data Center Locations: Enabling Growth in Services

Apple’s data center locations help power the company’s growth in services, including iCloud, with company needing more capacity located close to users.

dgtlinfra.com

Amazon web services has been installing about 50% ARM processors in their cloud over the past two years. Apple is a user of AWS and could already be developing their services on ARM.

 

[Mopetar]

Isn't the bulk of Apple's data center activity centered around content distribution or cloud storage? Most of what they're doing doesn't seem like it needs a significant number of servers or computational power since it's mostly storage. It seems like they're trying to do most of their AI stuff on device.

 

[Doug S]

Isn't the bulk of Apple's data center activity centered around content distribution or cloud storage? Most of what they're doing doesn't seem like it needs a significant number of servers or computational power since it's mostly storage. It seems like they're trying to do most of their AI stuff on device.

Is Apple doing its own iCloud data storage or is that still be outsourced to other cloud providers? iCloud may be big but it is a drop in the ocean compared to the scale of Amazon or Microsoft's cloud data storage. I'm not sure they DIY it less expensively than they could negotiate it with third parties.

As more countries are likely to have requirements that data has to be stored in their borders, Apple datacenters acting as a front end to talk to end user devices in "iCloud protocol" and passing the data off to third parties to store makes more sense than building infrastructure every time a country passes a law requiring local data storage.

 

[jpiniero]

Is Apple doing its own iCloud data storage or is that still be outsourced to other cloud providers?

The answer might be both.