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

M5 Family discussion here:

Page 431 - Discussion - Apple Silicon SoC thread

Page 431 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[jdubs03]

So the perf and power improvements came at esentially no area cost??

That’s interesting because in their publications they were talking about how they dedicated more area of the CPU core to memory enforcement integrity, is was a main focus.

 

[adroc_thurston]

That's still a lot compared to Pegasus.

It's also higher perf/more efficient so there's that.
No free lunch!

 

[digitaldreamer]

Am I to understand that Apple dedicates die space to power control/switching to manage power consumption? So, not all die space is for performance, but also managing power. And, that's where Apple gets its efficiency gains — turning off power when not used/needed.

 

[Mopetar]

All modern CPUs do that to some degree. Apple may be more aggressive at pursuing it than others, but I think a lot of their efficiency comes from their core design and other parts of the SoC being efficient.

Spending a lot of transistors to manage power means having to supply power to a lot of additional transistors. For a lot of cases, trying to determine what to power off is more expensive than just letting it run and do nothing. It's better to find ways to keep everything fed as much as possible so that it's doing useful work while powered on.

There's still room for power management, but it's an easy trap to fall into where you end up spending dollars to save pennies.

 

[smalM]

So the perf and power improvements came at esentially no area cost??

If you believe that TSMC kept quiet about N3P being much denser than they previously claimed...

 

[Doug S]

So the perf and power improvements came at esentially no area cost??

N3P is an optical shrink that's also a bit faster and/or power efficient. If Apple simply took A18P without modification they'd get a good bit of the way towards A19P's improvements and within a couple mm^2 of the size.

If you add in that FinFlex is still kinda new and maybe they weren't as aggressive with it at first and are learning the ropes of how to best use it, that could easily explain the remaining area/power/performance improvement even without any uarch changes. But of course there would have been some uarch changes as well...

 

[poke01]

If you believe that TSMC kept quiet about N3P being much denser than they previously claimed...

N3P offers a 6% density gain. I’d say Apple did a good job especially for A19 Pro GPU

 

[name99]

Am I to understand that Apple dedicates die space to power control/switching to manage power consumption? So, not all die space is for performance, but also managing power. And, that's where Apple gets its efficiency gains — turning off power when not used/needed.

Yes, but...
You can (and people do) place passives (capacitors, less frequently inductors), and the hardware required to isolate voltage domains with a lot of flexibility, so people tend to stick them all over the place, in between "real" logic and memory (and in stranger places, like sandwiched between metal high layers.
Everyone does this, because there are always free square millimeters sitting between where the big logic blocks touch each other.

Apple does better at this in terms of power not because they're using this free area, but because the entire system (each IP block, the SoC as a whole, and the OS) has been designed for low power. It's always easier to start from a good design than to retrofit...
For example EVERY clock and voltage domain needs an adaptor(that shifts the voltage, and waits for the appropriate clock cycle matching) to move a signal to a different domain. Apple has thousands of these different domains! If your design tools began with this assumption, you get all those adaptors for free; but if your tools were not designed with this assumption you may sometimes have to put them in by hand, which limits how aggressively you're going to add new such domains.

BTW another way people use these leftover square mm is by placing experimental logic there that's targeting the next chip or two ahead. Put something there that has been simulated as working, but you want to test it in a real system at real frequencies. Of course no-one outside the HW (and maybe OS) group ever sees it!; but that's one way to run faster.
We know nV do this because they have talked about, but the assumption is that every big company does this.

 

[mvprod123]

https://www.bloomberg.com/news/newsletters/2025-11-16/apple-s-iphone-road-map-iphone-air-2-iphone-18-mac-pro-future-tesla-carplay-mi1q4l2o?srnd=undefined

Apple relegates the high-end Mac Pro desktop to the back burner. Up until the early 2010s, Apple’s Mac Pro was a staple of its computer lineup — a fan favorite among video editors, photographers, designers and other creative professionals.
Then, in 2013, Apple hobbled the desktop computer by releasing a version that customers likened to a trash can. It was a bold but impractical design that the company ultimately apologized for.
The problem wasn’t solely the Mac Pro’s looks: Executives later admitted that the design backed Apple into a “thermal corner,” meaning it was nearly impossible to upgrade with more powerful chips without causing the computer to overheat. The device became the most controversial Mac since the Power Mac G4 Cube in 2000.
Apple’s fix came in 2019 with a completely redesigned Mac Pro tower that evoked the classic aluminum approach from over a decade earlier. At the time, Apple pledged that the Mac Pro would be updated regularly and remain a critical part of the Mac family. But things didn’t play out that way.
The next major update didn’t arrive until 2023, when Apple finally transitioned the desktop to in-house chips with the M2 Ultra Mac Pro. Two years later, that model remains largely unchanged. And it’s been overshadowed by the Mac Studio, which received the M3 Ultra chip earlier this year while the Mac Pro stayed put.
Now here’s the bad news: That doesn’t look set to change anytime soon. There’s no longer an M4 Ultra in the works (a Mac Pro to support it was also nixed), and the next high-end desktop chip will be the M5 Ultra. So far, Apple is only focused on a new Mac Studio for the processor. That suggests the Mac Pro won’t be updated in 2026 in a significant way.
From what I’ve heard inside the company, Apple has largely written off the Mac Pro. The sentiment internally is that the Mac Studio now represents both the present and future of Apple’s professional desktop strategy.

It seems that Apple will abandon the Mac Pro.

 

[poke01]

https://www.bloomberg.com/news/newsletters/2025-11-16/apple-s-iphone-road-map-iphone-air-2-iphone-18-mac-pro-future-tesla-carplay-mi1q4l2o?srnd=undefined

Apple relegates the high-end Mac Pro desktop to the back burner. Up until the early 2010s, Apple’s Mac Pro was a staple of its computer lineup — a fan favorite among video editors, photographers, designers and other creative professionals.
Then, in 2013, Apple hobbled the desktop computer by releasing a version that customers likened to a trash can. It was a bold but impractical design that the company ultimately apologized for.
The problem wasn’t solely the Mac Pro’s looks: Executives later admitted that the design backed Apple into a “thermal corner,” meaning it was nearly impossible to upgrade with more powerful chips without causing the computer to overheat. The device became the most controversial Mac since the Power Mac G4 Cube in 2000.
Apple’s fix came in 2019 with a completely redesigned Mac Pro tower that evoked the classic aluminum approach from over a decade earlier. At the time, Apple pledged that the Mac Pro would be updated regularly and remain a critical part of the Mac family. But things didn’t play out that way.
The next major update didn’t arrive until 2023, when Apple finally transitioned the desktop to in-house chips with the M2 Ultra Mac Pro. Two years later, that model remains largely unchanged. And it’s been overshadowed by the Mac Studio, which received the M3 Ultra chip earlier this year while the Mac Pro stayed put.
Now here’s the bad news: That doesn’t look set to change anytime soon. There’s no longer an M4 Ultra in the works (a Mac Pro to support it was also nixed), and the next high-end desktop chip will be the M5 Ultra. So far, Apple is only focused on a new Mac Studio for the processor. That suggests the Mac Pro won’t be updated in 2026 in a significant way.
From what I’ve heard inside the company, Apple has largely written off the Mac Pro. The sentiment internally is that the Mac Studio now represents both the present and future of Apple’s professional desktop strategy.

lol they can’t get chiplets to scale up with the SoC approach.

 

[mvprod123]

lol they can’t get chiplets to scale up with the SoC approach.

It's not about scalability. The idea behind the Mac Pro included RAM replacement and external GPU connectivity. Unified memory architecture limits this. Fortunately for them and unfortunately for enthusiasts, there is no point in investing resources in a product that accounts for barely 1-2% of total Mac sales.

 

[okoroezenwa]

lol they can’t get chiplets to scale up with the SoC approach.

How do you know?

 

[poke01]

https://twitter.com/x/status/1989855761326116988

I can see 1TB RAM in the Mac Studio M5 Ultra.

https://twitter.com/x/status/1989855761326116988

 

[Doug S]

It's not about scalability. The idea behind the Mac Pro included RAM replacement and external GPU connectivity. Unified memory architecture limits this. Fortunately for them and unfortunately for enthusiasts, there is no point in investing resources in a product that accounts for barely 1-2% of total Mac sales.

Yep the only real difference between the Mac Pro and the Mac Studio is the PCIe slots. Since PCIe GPUs are not an option, how much is out there that people would plug into a Mac Pro that TB 5 is not good enough for? Even stuff where it doesn't quite make the cut, like 100 GbE or the fastest PCIe 5 NVMe SSDs it is honestly close enough that it won't be limiting almost anyone's real world workflows.

Before the Studio there was still a niche for Mac Pro even for people who didn't need PCIe, because it was the only way to get a Max in a desktop or an Ultra in anything. Perhaps the existence of the Studio shrunk Mac Pro's niche so small they no longer see a market for it?

While I highly (highly) doubt they would ever do it, if they really wanted to serve the entire Mac Pro market (and even expand it somewhat) with Studio they could offer a version that supports PCIe/CXL externally via a QSFP interface. Then you can use an external PCIe enclosure, add CXL RAM, etc.

 

[mvprod123]

https://twitter.com/x/status/1989855761326116988

I can see 1TB RAM in the Mac Studio M5 Ultra.

https://twitter.com/x/status/1989855761326116988

I'm not sure about the 1TB in the M5 Ultra, but it seems that the M5 Max may get a 256GB RAM option.

 

[johnsonwax]

Yep the only real difference between the Mac Pro and the Mac Studio is the PCIe slots. Since PCIe GPUs are not an option, how much is out there that people would plug into a Mac Pro that TB 5 is not good enough for? Even stuff where it doesn't quite make the cut, like 100 GbE or the fastest PCIe 5 NVMe SSDs it is honestly close enough that it won't be limiting almost anyone's real world workflows.

Mostly capture cards and I/O cards. the Mac Pro is mainly a product for the video market, and every time there's an external bus fast enough to do the job, those things usually move to that bus. TB5 isn't fast enough to stream 60FPS 8K RAW over it so if that's your critical need, you need a card for that, or to split that over two interfaces.

Essentially every generation of Thunderbolt knocks out probably half of the remaining use case for internal PCI, so the Mac Pro increasingly appeals to a corner case of the preceding corner case market. Some places will favor the reliability of a card screwed into the chassis, but others will favor the ability to plug that into a laptop and take it on set, in the field, etc. Apple works pretty closely with the industry, so if Apple kills the Mac Pro it'll be because the industry says that they don't really need the cards any longer. I mean, Apple built the Afterburner card exclusively to keep that industry on board in the late days of x86 Macs and then built all of those features into Apple Silicon. If they do kill off the Mac Pro (something I've been expecting for a while) expect some new features in Studio or some external hardware that fills the needed gap. Mac Pros used to be common in the audio market and I think they're fine with Studio now as their needs haven't scaled as fast as Thunderbolt/Apple Silicon has.

One of the annoying issues for both of these markets is they do still rely a lot on rack mounting their hardware and Apple doesn't exactly cater to that need. I mean, you can do it, but it kind of sucks. I've sort of wondered what Apple's move back into server hardware (if only for their internal needs) might crack open. Once you're designing for your own data center, might you put a rack product back for sale since you've already designed it, set up manufacturing, etc? I wonder to what extent Apple's internal AI servers also meet what the video market's needs that aren't served by Studio and whether that might be the better angle to approach the problem from. Doubt they'd have PCI slots in their internal products though unless they too are reliant on 3rd party FC cards, that kind of thing.

 

[Doug S]

Mostly capture cards and I/O cards. the Mac Pro is mainly a product for the video market, and every time there's an external bus fast enough to do the job, those things usually move to that bus. TB5 isn't fast enough to stream 60FPS 8K RAW over it so if that's your critical need, you need a card for that, or to split that over two interfaces.

Isn't 60 fps 8K raw ~100 Gbps? TB5 can do that, it does either 80 Gbps bidirectionally or 120 Gbps one way and 40 Gbps the other.

 

[poke01]

https://twitter.com/x/status/1988969455284818325

also first time Apple has used Trace cache in its CPU?

 

[mvprod123]

also first time Apple has used Trace cache in its CPU?

yes

There is speculation that Apple has implemented its Trace Cache technology patent in the A19/A19 Pro (M5?) CPUs. https://patents.google.com/patent/US20250021332A1

 

[Mopetar]

The only downside to something like the Studio as a Pro replacement is that it's not at all upgradable. For the most part it doesn't need to be as most anything a user might want to add (aside from RAM) could be connected externally via USB. However, I don't think it would kill Apple to add a slot or two for adding additional storage internally.

I'll personally miss the iconic case design. Even the impractical trash can at least looked cool. The Mac Studio is a tidy minimalist design, but the G5 and early Xeon Pros were handsome looking machines. I suppose a person could always gut one of those and stick a bunch of Studios inside of the case.

 

[LightningZ71]

At this point, the only way that we're likely to see an upgradeable MAC again is if they add CXL attached memory to the studio at some point in the future. If the industry REALLY needs 100+ GB/sec bandwidth for video production work, I'm absolutely sure that Apple could find a way to do TB5 dual link to use the unidirectional boost speed one way on one cable and the opposite on the other. They might coin it as some other standard or brand, but it's doable, and is likely far cheaper than doing a VERY small market Mac Pro desktop.

 

[name99]

yes

Just so people don't get the wrong idea, let me be very very clear that what Apple has implemented is a *GENUINE* *TRACE* cache. Not a µOp cache, not a MOP cache, not the various abominations that other people (most importantly Intel) have implemented and called a Trace Cache.

The original Trace Cache academic proposal existed to solve one, and only one, specific problem, namely achieving enough Fetch bandwidth as a design gets wider and wider. Apple's implementation exists to solve that exact same problem.
You will not understand why it exists if you imagine that it's trying to do the same thing as Intel was doing, or ARM (not any more, but designs a few years ago) or Veyron in it's current versions...

 

[StinkyPinky]

I'm sure the three people that bought the Mac Pro last year are upset at the latest news. The factory worker in Texas that assembles them (part time, on Fridays) will need to find a new role.

 

[Doug S]

In the very bottom left of that A19P die photo is a tiny block that's been annotated as "modem". Modulating and demodulating what?

 

[mvprod123]

Cool

You can turn a cluster of Macs into an AI supercomputer in macOS Tahoe 26.2

Apple is making it easier to turn a bunch of Macs into an AI supercomputer with Thunderbolt 5 clustering.

www.engadget.com