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

 

[poke01]

Wasn't AVX2 locked behind patents ? Have those expired?

Apparently they expire this year.

 

[okoroezenwa]

macOS requires an M3 chip or later for NV.

isNestedVirtualizationSupported | Apple Developer Documentation

A Boolean value that describes whether the platform configuration supports nested virtualization.

developer.apple.com

Boo 😒

 

[poke01]

Boo 😒

Agree, M2 supports NV. At least it’s not locked to M4…

 

[Doug S]

Agree, M2 supports NV. At least it’s not locked to M4…

Maybe M2 "supports" it but the implementation had a few issues that were fixed in M3? It is not that uncommon with something complex like that to take a rev or two to get it right. Especially since NV is kind of a specialized thing outside of servers. I'm surprised they support it at all, or that there are people here who are eager to do it.

I'm curious, what are you planning on using it for on a Mac?

 

[poke01]

I'm curious, what are you planning on using it for on a Mac?

Just a nice thing to have. Not planning on using it myself for work but for playing with vms.

 

[FlameTail]

Does the 6P cluster in M3 Pro/M3 Max have a bigger AMX unit than the 4P cluster in M3?

 

[SarahKerrigan]

Looks like Rosetta still doesn't support F16C or the BMI ops, which is causing some compatibility issues with games that use those in addition to AVX2.

Remember, kids, check all of the ISA features you're planning to use. Don't assume Extension A implies Extension B unless it's mandatory.

 

[poke01]

Looks like Rosetta still doesn't support F16C or the BMI ops, which is causing some compatibility issues with games that use those in addition to AVX2.

Remember, kids, check all of the ISA features you're planning to use. Don't assume Extension A implies Extension B unless it's mandatory.

Apple devs know feedback has been sent. Hopefully they add them

 

[poke01]

Also looks like Apple using M2 Ultras for servers is true. Some doubted but it looks like it’s real. The 3nm servers are likely M4 Ultras.
Makes me wonder if the M4 Extreme is real too?

 

[Doug S]

Also looks like Apple using M2 Ultras for servers is true. Some doubted but it looks like it’s real. The 3nm servers are likely M4 Ultras.
Makes me wonder if the M4 Extreme is real too?

Maynard mentioned something about this and I did a quick skim of the transcript but didn't see anything about servers. Do you have a link?

He was implying one of the things Apple was doing is something I had suggested in RWT 7-8 years ago - putting Apple ARM servers in the cloud to act as a way of supplementing computation for apps beyond what they can provide locally (which could be useful for something like Mathematica, i.e. apps where you need extra CPU grunt but aren't needing to push many gigabytes of data to the cloud) So iCloud wouldn't just be something that offers data storage, it would now be like AWS/Azure and offer CPU but since they could perfectly mirror the user's hardware and software environment it would be a standard iOS/macOS library call for devs to support in their apps.

I pretty much assumed there was zero chance Apple was pursuing that when it was reported GW III and co left Apple because they wanted to design servers. If Apple is doing this now, I guess they decided after GW III left? Or maybe he wanted to design chips ground up for servers and not repurpose Mac Pro chips? Or maybe the claim he was leaving because he wanted to work on servers was never the truth...

But if they are offering computation and not just AI in the cloud it makes sense they'd use Ultra chips instead of designing custom chips for AI servers. Might also explain the reason they've added support for nested virtualization?

 

[SpudLobby]

Apparently they expire this year.

I’m skeptical it’s about this at all. Patents also last a long time.

Maynard mentioned something about this and I did a quick skim of the transcript but didn't see anything about servers. Do you have a link?

He was implying one of the things Apple was doing is something I had suggested in RWT 7-8 years ago - putting Apple ARM servers in the cloud to act as a way of supplementing computation for apps beyond what they can provide locally (which could be useful for something like Mathematica, i.e. apps where you need extra CPU grunt but aren't needing to push many gigabytes of data to the cloud) So iCloud wouldn't just be something that offers data storage, it would now be like AWS/Azure and offer CPU but since they could perfectly mirror the user's hardware and software environment it would be a standard iOS/macOS library call for devs to support in their apps.

Well…. In a way yea lol exactly

“The root of trust for Private Cloud Compute is our compute node: custom-built server hardware that brings the power and security of Apple silicon to the data center, with the same hardware security technologies used in iPhone, including the Secure Enclave and Secure Boot. We paired this hardware with a new operating system: a hardened subset of the foundations of iOS and macOS tailored to support Large Language Model (LLM) inference workloads while presenting an extremely narrow attack surface. This allows us to take advantage of iOS security technologies such as Code Signing and sandboxing.”

I pretty much assumed there was zero chance Apple was pursuing that when it was reported GW III and co left Apple because they wanted to design servers. If Apple is doing this now, I guess they decided after GW III left? Or maybe he wanted to design chips ground up for servers and not repurpose Mac Pro chips? Or maybe the claim he was leaving because he wanted to work on servers was never the truth...

But if they are offering computation and not just AI in the cloud it makes sense they'd use Ultra chips instead of designing custom chips for AI servers. Might also explain the reason they've added support for nested virtualization?

So yeah that’s what they’re doing roughly but not with API access like a traditional cloud provider yet or extending for traditional Apple compute which, AFAICT their services business runs on Google Cloud. So just for LLMs and Apple’s AI cloud… *for now* presumably.


There’s also this session later today just for the Swift ecosystem, interestingly timed.

Explore the Swift on Server ecosystem - WWDC24 - Videos - Apple Developer

Swift is a great language for writing your server applications, and powers critical services across Apple's cloud products. We'll explore...

developer.apple.com

 

[SpudLobby]

I strongly suspect it’s M2 Ultras. They have the scale for it and the orders for packaging in the rumor mill suggested it too. It’s also “good enough”.

I am curious if at some point they’ll just throw other services on stuff like this. With enough scale it can make sense and they could get serious TCO gains, but with an M2 Ultra the core counts aren’t really high and you have a lot of GPU die space, so stuff like Apple Music or iCloud would be just silly.

But at some point maybe they’ll tape out their own chip. I’m skeptical for now, but it could last them a really long time and be fairly generic.

Here it looks extremely specialized and yes, in the direction Maynard thought, which is terrible for his ego but you have to hand it to him.

 

[poke01]

Returnal which is a game that requires AVX now runs on Mac.

Method: CrossOver 24 + GPTK 2.0 (D3DMetal ON - MSYNC ON)
Graphics: 1080pGraphics: Low

1080p at 30fps
Tested on M1 Pro with 16GB RAM.

 

[Doug S]

I’m skeptical it’s about this at all. Patents also last a long time.

According to wikipedia AVX instructions were first proposed by Intel in 2008, and at the time patents were for 17 years (though who knows when they would have patented it, since "proposed" is a fairly nebulous term) So it could be about the right time, at least theoretically.

Though honestly it may never have been about patents, despite what rumors claimed. Apple probably supported SSE 4.1 or whatever it was they supported with the initial Rosetta 2 because that was sufficient for getting all the x86 Mac apps they cared about running. None of those apps required AVX/AVX2, they had codepaths for SSE that would work.

Since then perhaps due to customer demand or maybe engineers with a little too much spare time, they have caught it up to something more current.

 

[Eug]

"8 GB ought to be enough for anybody"

Xcode 16 Beta Release Notes | Apple Developer Documentation

Update your apps to use new features, and test your apps against API changes.

developer.apple.com

Xcode 16 includes predictive code completion, powered by a machine learning model specifically trained for Swift and Apple SDKs. Predictive code completion requires a Mac with Apple silicon and 16GB of unified memory, running macOS 15.

 

[poke01]

https://twitter.com/x/status/1801323681370804509

MLX is going places

 

[reggie_fils_aime]

https://twitter.com/x/status/1801323681370804509

MLX is going places

This is (unironically, we love clusters) really cool, and will absolutely lead to people buying thousands of Walmart iphones the second a qualified device goes on sale. Innovation!

 

[FlameTail]

Apple planning redesigned iPhone, MacBook Pro, and Apple Watch that are significantly thinner - 9to5Mac

Last month, Apple unveiled its thinnest product ever with the new M4 iPad Pro. Measuring in at 5.1mm, the 13-inch...

9to5mac.com

They are going to make the Macbook Pro thinner.

Which means the TDP is going to be reduced, while the Apple Silicon team continues to raise the power consumption...

Unless, they are cooking up a new microarchitecture that significantly reduces the power consumption?

Or they are going to make the Macbook Pro thinner while maintaining the TDP?

PS: A thinner Macbook Pro will also have worse speaker quality.

 

[igor_kavinski]

Yeah, like Doug, they figured, people don't need their battery past their bedtime. Let's make the laptop thinner and raise the prices!

 

[FlameTail]

Let's make the laptop thinner and raise the prices!

Utter blasphemy! I am bringing the guillotine.../s

 

[TwistedAndy]

Apple planning redesigned iPhone, MacBook Pro, and Apple Watch that are significantly thinner - 9to5Mac

Last month, Apple unveiled its thinnest product ever with the new M4 iPad Pro. Measuring in at 5.1mm, the 13-inch...

9to5mac.com

They are going to make the Macbook Pro thinner.

Which means the TDP is going to be reduced, while the Apple Silicon team continues to raise the power consumption...

Unless, they are cooking up a new microarchitecture that significantly reduces the power consumption?

Or they are going to make the Macbook Pro thinner while maintaining the TDP?

PS: A thinner Macbook Pro will also have worse speaker quality.

The M3 Max in 14" chassis now has the 80-90W peak SoC package limit, while the same chip in 16" chassis has a 120-130W package limit.

I think Apple will reduce the package power a bit, introduce a more aggressive frequency boost policy, and allow CPU cores to consume all the SoC power under full load.

Apple M4 Max will be fine with the 100-110W package in a thinner 16" chassis for the CPU-only workload (7-9W * 12 = 84 - 108W + some power for E-cores).

As for the smaller 14" chassis, there will be some kind of throttling to 60-70W (3.6-3.8GHz) when all the cores are loaded. Also, we can expect some sort of Turbo Boost for a minute to pass the Geekbench 6 test without throttling.

 

[CouncilorIrissa]

Yeah, like Doug, they figured, people don't need their battery past their bedtime. Let's make the laptop thinner and raise the prices!

When was the last time you held a 16 inch MacBook Pro? The thing could definitely use some dieting (to being as thick as its 14 inch counterpart) as it's unironically very beefy.

 

[igor_kavinski]

When was the last time you held a 16 inch MacBook Pro?

I have held fatter laptops. I don't carry them around unnecessarily but I can understand that Apple users need to flaunt their Apple devices. I mean, you know, they spent a lot of money on those things. Might as well show them off as much as possible to tell the world how well off they are.

 

[CouncilorIrissa]

I have held fatter laptops. I don't carry them around unnecessarily but I can understand that Apple users need to flaunt their Apple devices.

It's not a desktop replacement, you're supposed to carry it. The thing is not very comfortable to use even when working from home and moving one room to another. It's not about bragging rights.

But sure, keep treating those using Apple products as different species.

 

[igor_kavinski]

But sure, keep treating those using Apple products as different species.

Who? Homo Apple-us?