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

 

[igor_kavinski]

2) You're annoyed because M1 Macs retain their value too well.

I'm more annoyed because it's artificial, just like the GPU price gouging was. People holding onto stuff for their dear life and insisting on selling it for more than it's worth.

 

[igor_kavinski]

LOL. All this disdain for  and it’s user base, and you are still pining away for a M1 Mac

I would like to benchmark an M1 laptop but Apple and their zombies have refused me that privilege with their deep pockets. You guys here shouldn't be offended since you actually use your Macs for getting actual work done and not only for browsing/watching movies and showing off your Macbook at cafes, right? RIGHT????

 

[Eug]

I'm more annoyed because it's artificial, just like the GPU price gouging was. People holding onto stuff for their dear life and insisting on selling it for more than it's worth.

But that's just it. You can get good money for selling used Macs, and this is doubly true for the Apple Silicon models. People who don't sell for bargain basement prices are probably the smart ones.

If you look hard enough you sometimes can find good deals on used Macs, but it takes some luck and lots of perseverance, and currently that only really applies to the Intel Macs.

BTW, remember that some of the Apple Silicon Macs look exactly the same as the Intel Macs, but the Intel Macs seem to be dropping resale values quicker than in previous years. That negates your buying-to-be-seen argument.

 

[Ajay]

He's just trolling all of us at this point. If one is going to derail a thread, they might as well go big

 

[DrMrLordX]

"Apple users" are fanboys

Careful. That term has been bannable in several subforums here.

Windows users are the way bigger and more worrying "zombiebase" out there, most of them aren't even fans!

It's safe to say that most people who use Windows have a love/hate relationship with it. Mostly it's a hate/hate relationship.

 

[ElFenix]

gais apel users aer all fanbois cuz nun of htem wil seel me anh mcbook air 4 33% off!

 

[amosliu137]

3.5G avalanche

Mac14,7 - Geekbench

Benchmark results for a Mac14,7 with an Apple M2 processor.

browser.geekbench.com

 

[Eug]

3.5G avalanche

Mac14,7 - Geekbench

Benchmark results for a Mac14,7 with an Apple M2 processor.

browser.geekbench.com

Another one: 1927 / 8805.

 

[Eug]

As expected, Mac14,7 = 13" MacBook Pro.

List of Mac BoardID, DeviceID, Model Identifiers & Machine Models

Updated list of Mac BoardIDs, Model Identifiers & Machine Models from all intel Macs 2006-2019. MBPro, MBAir, MacBook, Mac Pro, iMac Mac Mini

mrmacintosh.com

IOW, Apple has changed the naming scheme, starting with Mac Studio. Going forward it appears everything is just going to be MacXX,X.

 

[Eug]

M2 gets around 30000 in Geekbench Metal.

Mac14,7 - Geekbench

Benchmark results for a Mac14,7 with an Apple M2 processor.

browser.geekbench.com

This represents about a 35% increase over M1's GPU (8-core).

 

[JasonLD]

Looking like it is similar to A14 to A15 jump, which was also based on N5 and N5P respectively.

 

[Eug]

Another Metal score: 30627

Mac14,7 - Geekbench

Benchmark results for a Mac14,7 with an Apple M2 processor.

browser.geekbench.com

 

[trivik12]

So M2 single core score went up tad below 12% with 9% clock speed. I am curious how A16 would do. Is that expected to have Armv9?

 

[Doug S]

So M2 single core score went up tad below 12% with 9% clock speed. I am curious how A16 would do. Is that expected to have Armv9?

I don't think there are any rumors about A16 performance yet. Whether it is ARMv9 or not would make zero difference to its performance. This is not like ARMv8 which cleaned up a lot of problems with AArch32 and allowed for performance increases despite the negative effect of larger pointers.

 

[Heartbreaker]

I don't think there are any rumors about A16 performance yet. Whether it is ARMv9 or not would make zero difference to its performance. This is not like ARMv8 which cleaned up a lot of problems with AArch32 and allowed for performance increases despite the negative effect of larger pointers.

Yep, it's mostly extensions like SVE2, and AFAIK, Apple still hasn't implemented SVE1.

 

[Doug S]

Yep, it's mostly extensions like SVE2, and AFAIK, Apple still hasn't implemented SVE1.

SVE2 is optional in both ARMv8 and ARMv9. If they wanted they could add it without going to ARMv9.

In fact, pretty much everything ARMv9 brings is optional, so the benefit of going to it is questionable. Apple may do anyway as they haven't been lagging on ARMv8.x updates too much, but people who expecting that to mean anything if it happens will be disappointed.

 

[NostaSeronx]

SVE2 is optional in both ARMv8 and ARMv9.

ARMv9.0A has SVE/SVE2 default enabled. So, it is mandatory.
Crypto SVE2 and Bitmanip SVE2 is optional.

ARMv9.0A without SVE2 must be labeled as ARMv8.5A. NO SVE = ARMv8, there is thus no case of an ARMv9 core not having SVE2. Unless, they break compatibility with all current ARMv9 compilers/utils/etc. Which state SVE/SVE2 is mandatory/compulsory with auto-enabled-optimization.

ARMv9.0A can not be labeled such if it doesn't have SVE2. It will return as a ARMv8.5A core instead.
ARMv9.1A can not be labeled such if it doesn't have SVE2. It will return as a ARMv8.6A core instead.
ARMv9.2A can not be labeled such if it doesn't have SVE2. It will return as a ARMv8.7A core instead.
ARMv9.3A can not be labeled such if it doesn't have SVE2. It will return as a ARMv8.8A core instead.

Cortex®‑X2 core features;

• Implementation of the Scalable Vector Extension (SVE) with a 128-bit vector length and Scalable Vector Extension 2 (SVE2)

Cortex®‑A710 core features;

• Implementation of the Scalable Vector Extension (SVE) with a 128-bit vector length and Scalable Vector Extension 2 (SVE2)

Cortex®‑A510 core features;

• Scalable Vector Extension (SVE) and SVE2 SIMD instruction set, offering Advanced SIMD and floating-point architecture support

Thus, SVE&SVE2 is mandatory. All ARMv9A compliant cores MUST have SVE&SVE2. If they don't then they aren't ARMv9A but ARMv8A.

Rumors state 4nm/3nm Apple cores will be ARMv9 and support the same above instructions because it is mandatory.

 

[DrMrLordX]

Apple still hasn't implemented SVE1.

SVE/SVE1 is pretty niche. I don't think it was ever meant for consumer parts. Who besides Fujitsu has actually implemented it in production silicon?

 

[Nothingness]

SVE/SVE1 is pretty niche. I don't think it was ever meant for consumer parts. Who besides Fujitsu has actually implemented it in production silicon?

IIRC Arm Neoverse V1 has the first variant of SVE. But I guess all subsequent will only have SVE2.

 

[Mopetar]

SVE/SVE1 is pretty niche. I don't think it was ever meant for consumer parts. Who besides Fujitsu has actually implemented it in production silicon?

Apple owns the software stack for their own devices. There's likely something that could make use of those instructions on the OS side, even if most consumers applications wouldn't touch them.

You'd think they'd certainly want to build those in to any chip that would go into a MacPro though. SIMD is definitely something that crowd can or does make use of.

 

[LightningZ71]

We have been told on these very forums, repeatedly, that there's nothing in SIMD instruction that can't be done FAR better on gpu type circuits. Since Apple has a healthy iGPU in every SoC that they make, I don't see them having much use for SIMD instructions, aside for some crypto ones, in their CPU ISA.

 

[Heartbreaker]

ARMv9.0A has SVE/SVE2 default enabled. So, it is mandatory.
Crypto SVE2 and Bitmanip SVE2 is optional.

ARMv9.0A without SVE2 must be labeled as ARMv8.5A. NO SVE = ARMv8, there is thus no case of an ARMv9 core not having SVE2. Unless, they break compatibility with all current ARMv9 compilers/utils/etc. Which state SVE/SVE2 is mandatory/compulsory with auto-enabled-optimization.

ARMv9.0A can not be labeled such if it doesn't have SVE2. It will return as a ARMv8.5A core instead.

Apple doesn't seem all that concerned with following ARM designs going forward, or with needing ARMv9 labelling.

It's not like they are selling ARM SoC to a third party, and need to worry about having ARMV9 designation for marketing, and most of ARMv9 seems to be extensions aimed at things Apple already has it's own solutions for like Security, or AI, or SVE2 when Apple hasn't even bothered with SVE1.

 

[Nothingness]

Apple doesn't seem all that concerned with following ARM designs going forward, or with needing ARMv9 labelling.

It's not like they are selling ARM SoC to a third party, and need to worry about having ARMV9 designation for marketing, and most of ARMv9 seems to be extensions aimed at things Apple already has it's own solutions for like Security, or AI, or SVE2 when Apple hasn't even bothered with SVE1.

There's a lot of benefit in using the official Arm architecture. Following the architecture brings them one huge benefit: they get official validation suites from Arm. Anything outside architecture and they are on their own. And this applies to all tools and libraries. Of course they could develop everything in-house, but taking advantage of an existing ecosystem is much wiser.

BTW I don't understand why people insist on SVE1. This has never been a feature for client CPU, it's an HPC thing.

 

[Eug]

M2 is around 8900 for Geekbench multi-core, but the value that would go on this graph would likely be a fair bit lower after averaging out some scores. Perhaps 8400ish?

 

[Doug S]

ARMv9.0A has SVE/SVE2 default enabled. So, it is mandatory.
Crypto SVE2 and Bitmanip SVE2 is optional.

ARMv9.0A without SVE2 must be labeled as ARMv8.5A. NO SVE = ARMv8, there is thus no case of an ARMv9 core not having SVE2. Unless, they break compatibility with all current ARMv9 compilers/utils/etc. Which state SVE/SVE2 is mandatory/compulsory with auto-enabled-optimization.

ARMv9.0A can not be labeled such if it doesn't have SVE2. It will return as a ARMv8.5A core instead.
ARMv9.1A can not be labeled such if it doesn't have SVE2. It will return as a ARMv8.6A core instead.
ARMv9.2A can not be labeled such if it doesn't have SVE2. It will return as a ARMv8.7A core instead.
ARMv9.3A can not be labeled such if it doesn't have SVE2. It will return as a ARMv8.8A core instead.

Cortex®‑X2 core features;

• Implementation of the Scalable Vector Extension (SVE) with a 128-bit vector length and Scalable Vector Extension 2 (SVE2)

Cortex®‑A710 core features;

• Implementation of the Scalable Vector Extension (SVE) with a 128-bit vector length and Scalable Vector Extension 2 (SVE2)

Cortex®‑A510 core features;

• Scalable Vector Extension (SVE) and SVE2 SIMD instruction set, offering Advanced SIMD and floating-point architecture support

Thus, SVE&SVE2 is mandatory. All ARMv9A compliant cores MUST have SVE&SVE2. If they don't then they aren't ARMv9A but ARMv8A.

Rumors state 4nm/3nm Apple cores will be ARMv9 and support the same above instructions because it is mandatory.

I know it was widely reported SVE2 was mandatory, but that's 100% wrong. It is definitely optional, and says nothing about having to report as ARMv8 if it lacks SVE2.

Now I'm not saying that if Apple goes ARMv9 they won't do SVE2, but they don't have to and might prefer to trod their own path with their AMX stuff instead.

Here's a link to the ARMv9-A architecture reference manual. See page 26, where it says:

B1.1.2 FEAT_SVE2, Scalable Vector Extension version 2
The Scalable Vector Extension version 2 (SVE2) is a superset of SVE that incorporates functionality similar to Advanced SIMD, and other enhancements.
FEAT_SVE2 is OPTIONAL.