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

 

[Doug S]

Imagine if somehow Apple manages to tie with Mediatek, which are below Qualcomm but way above Exynos.

If they were going to work with Mediatek they would have had to do that a few years ago. No point now that they have their own design complete.

They don't need to be competitive with Qualcomm's 5G - nobody achieves the theoretical speeds and there's not much use for speeds over a couple hundred Mbps to a phone. When there's heavy congestion there's not much anyone can do about it. What really matters is in the marginal areas on the edge of a cell where you struggle to get 5 Mbps, and it'll make a difference between maybe Apple manages 3 Mbps or Apple has no bars.

 

[name99]

Crazy fact:

A17 Pro : 2P + 4E
M3 : 4P + 4E
M3 Pro : 6P + 6E

They all have the same L2 cache capacity : 16 MB + 4 MB

Like I have said repeatedly, the obvious response when SRAM doesn’t shrink but logic does (current situation until we get BSPD) is to use more logic, and to use logic to get more out of the SRAM…

There are many ways to make a CPU faster, it’s not that some of the are good and some are bad; it’s all about are you willing to adapt to the way the process is evolving? Or do you insist on doing what worked 5 years ago, even though the process now has very different tradeoffs?
So far I’d say Apple have shown a superb evolution of their design to match exactly the changing characteristics of the target process.

 

[Mopetar]

A18 is going into everything as the base Apple Intelligence experience. iPhone SE, base iPad, iPad Mini, Apple TV, Homepod.

What does an AppleTV need an NPU for? I can at least see the argument for something like a HomePod where it could use it for producing better surround experience or something like that. The AppleTV box really doesn't need anything more powerful than what it's currently using and even that is overkill.

A18 won't hit some of those other devices for a while either. By the time it does, Apple will be on A21 or a subsequent SoC. The AI hype/craze may well be over by then anyway.

 

[jdubs03]

If they were going to work with Mediatek they would have had to do that a few years ago. No point now that they have their own design complete.

They don't need to be competitive with Qualcomm's 5G - nobody achieves the theoretical speeds and there's not much use for speeds over a couple hundred Mbps to a phone. When there's heavy congestion there's not much anyone can do about it. What really matters is in the marginal areas on the edge of a cell where you struggle to get 5 Mbps, and it'll make a difference between maybe Apple manages 3 Mbps or Apple has no bars.

I still don’t think they’d want to cede much even in the way of theoretical speeds. They’ll want to be seen as competitive purely as a matter of prestige. They’ve shown themselves to be willing to go all in to produce top class designs.

 

[Doug S]

What does an AppleTV need an NPU for? I can at least see the argument for something like a HomePod where it could use it for producing better surround experience or something like that. The AppleTV box really doesn't need anything more powerful than what it's currently using and even that is overkill.

A18 won't hit some of those other devices for a while either. By the time it does, Apple will be on A21 or a subsequent SoC. The AI hype/craze may well be over by then anyway.

You have been able to talk to it for a while. It is just for searching (at least that's all I've used it for, maybe it can do more and I'm missing out) but it doesn't have to be. If I could say "what's that movie with the apes and Charleton Heston" and it could find it for me that's an improvement over having to know the name, or searching on his name and digging through the results hoping I recognize the title (yeah it would be obvious in this example, but that's not always the case)

 

[poke01]

It isn't like an M4 Max wouldn't make for a killer desktop. The problem is, desktops are a shrinking market

This is a myth. Every market is "shrinking". People don't upgrade desktops as frequently. IMO, the M4 Max wouldn't make a good desktop chip. The RAM is soldered, the GPU non-upgradeable, I/O not much for a premium desktop and hobbled PCIe connectivity. It's a laptop SoC down to its bones.

Apple knows what a desktop is they used to make em. 2019 Mac Pro with upgradeable CPU, SSD, RAM, GPU and the 2008-2020 iMac/iMac Pro with upgradeable CPU, SSD, RAM and external GPU support for some models. The current Mac Pro and Mac Studios are only desktops because they are not battery powered and headless.

announced they will provide full support for third party GPUs does anyone really believe Nvidia or AMD would bother porting their GPU driver?

Well, Apple controls the GPU drivers. AMD did provide drivers in the Intel era. But Apple is hyper focused on Metal and its own GPUs.

 

[FlameTail]

Apple A18 Pro hits 3600 Single Core and 9000 Multi Core in new record result;

https://twitter.com/x/status/1842421653743284530

 

[jdubs03]

Apple A18 Pro hits 3600 Single Core and 9000 Multi Core in new record result;

https://twitter.com/x/status/1842421653743284530

I see yours and raise!

Looks like they must’ve altered the CPU ramp up earlier. I was expecting that in 18.1 (as seen in beta 5).

 

[okoroezenwa]

This is a myth. Every market is "shrinking".

Just want to point out that “Every market is shrinking” doesn’t make “desktops are a shrinking market” a myth.

 

[Eug]

I see yours and raise!
View attachment 108703
Looks like they must’ve altered the CPU ramp up earlier. I was expecting that in 18.1 (as seen in beta 5).

Looking good for 4000/10000 for my A19 Pro next year.

 

[jdubs03]

Looking good for 4000/10000 for my A19 Pro next year.

It’s gonna be close I think. Might just reach those.

3642/9463 is a ridiculous score, they must’ve got an absolute top binned chip. I wouldn't* be surprised if they are using a freezer/testing in a cold place cause that MT score in particular is quite high compared to the average.

 

[FlameTail]

Looking good for 4000/10000 for my A19 Pro next year.

Only 10% gain over A18 Pro? I would expect more. ~15%

 

[name99]

What does an AppleTV need an NPU for? I can at least see the argument for something like a HomePod where it could use it for producing better surround experience or something like that. The AppleTV box really doesn't need anything more powerful than what it's currently using and even that is overkill.

A18 won't hit some of those other devices for a while either. By the time it does, Apple will be on A21 or a subsequent SoC. The AI hype/craze may well be over by then anyway.

Consider source material that’s of low quality (eg DVD or NTSC source). When you display it on a 4K TV you want to upsample it (and, ideally, also contrast-stretch it to HDR).
Right now I think Apple TV relies on the TV to do that, but the Apple TV box should do it, and be able to do it better than the TV.
This is basically an NPU task - sure classical upscaling was just filtering, but things like VSR and FSR are way beyond that…

Done correctly there are other versions of this idea.
How about temporal rescaling (handle cruddy 2:3 pull-up and 50->59.97 conversions; remove incompetently handled field artifacts)?
In the most extreme case, what about Apple TV on the fly colorizing BW movies?
It’s all, in theory, possible…

Yes, yes, we all know that some (very very loud and obnoxious people) don’t want anyone ever touching their precious pixels, and will tell us so till the cows come home.
Whatever.
Just switch off the feature in your aTV and let the rest of us enjoy it.

 

[Eug]

Consider source material that’s of low quality (eg DVD or NTSC source). When you display it on a 4K TV you want to upsample it (and, ideally, also contrast-stretch it to HDR).
Right now I think Apple TV relies on the TV to do that, but the Apple TV box should do it, and be able to do it better than the TV.
This is basically an NPU task - sure classical upscaling was just filtering, but things like VSR and FSR are way beyond that…

Apple TV 4K does 4K upscaling. (It would be very strange if it didn't.)

However, LG now has a so-called "AI" upscaler (whatever that means), built into their SoC. I don't know if in the future Apple will adopt a similar technology but currently Apple TV 4K supposedly doesn't support similar AI upscaling.

LG OLED TV | α9 Gen4 AI Processor | LG Levant

See what extraordinary looks like thanks to the a9 Gen4 AI Processor. LG OLED TV automatically analyzes and enhances picture and sound quality.

www.lg.com

I don't know how well this AI upscaling actually works though, since my older LG OLED TV predates this feature. My TV has LG's α9, but a much older version, I believe gen 1.

 

[ikjadoon]

What does an AppleTV need an NPU for? I can at least see the argument for something like a HomePod where it could use it for producing better surround experience or something like that. The AppleTV box really doesn't need anything more powerful than what it's currently using and even that is overkill.

A18 won't hit some of those other devices for a while either. By the time it does, Apple will be on A21 or a subsequent SoC. The AI hype/craze may well be over by then anyway.

It's maybe also needed for feature-parity for ported iOS features / apps like improved Siri (allegedly lol), captions / translation of live content, portrait screensavers, enhanced dialogue, etc.

Though, the CPU, GPU, and NPU do work together on intensive tasks, so it may make sense to upgrade them all simultaneously (e.g., as they do anyways by re-using older SoCs).

//

In my opinion, it's good the Apple TV SoC is seen as relatively overkill: they still feel speedy years later after apps have grown, tvOS has expanded, etc. Either features, or bloat. Some data on whether a streaming box's hardware perf is relevant for general UI tasks, from CNET's tests.

Speed is what people are here for, and I wanted to see if it was 30% faster than before. Using Endgame as my reference I used search from the main screen and pressed go. As you would expect, the 2022 model was the slowest, as it also has no caching capabilities, and it took an average of 17 seconds to load. The 2024 Ultra was quicker, and after an initial 19-second load, it opened every subsequent time in 14 seconds. Is the processor quicker, or is the cache doing its job? Only Roku can say, and the company did not reply to CNET's request for clarification. It's worth noting that the one-generation-behind Apple TV 4K (2021) was still noticeably faster at 8 seconds to load this movie.

From the home page, open the movie Endgame: when does playback start? I think the A55s are clocked at ~1.4 GHz, while the A12 Bionic is ~2.5 GHz.

I don't think people only buy Apple TVs for their speed, but it does let them feel "new" even in the long-term.

Streaming Box	Time (button → movie playback)	OS & CPU
Roku Ultra (2024) - cold start	19 seconds	Roku OS & ??
Roku Ultra (2024) - cached start	14 seconds	Roku OS & ??
Roku Ultra (2022)	17 seconds	Roku OS & Cortex-A55
Apple TV 4K (2021)	8 seconds	Apple tvOS & A12 Bionic

 

[Eug]

It's maybe also needed for feature-parity for ported iOS features / apps like improved Siri (allegedly lol), captions / translation of live content, portrait screensavers, enhanced dialogue, etc.

Though, the CPU, GPU, and NPU do work together on intensive tasks, so it may make sense to upgrade them all simultaneously (e.g., as they do anyways by re-using older SoCs).

//

In my opinion, it's good the Apple TV SoC is seen as relatively overkill: they still feel speedy years later after apps have grown, tvOS has expanded, etc. Either features, or bloat. Some data on whether a streaming box's hardware perf is relevant for general UI tasks, from CNET's tests.


From the home page, open the movie Endgame: when does playback start? I think the A55s are clocked at ~1.4 GHz, while the A12 Bionic is ~2.5 GHz.

I don't think people only buy Apple TVs for their speed, but it does let them feel "new" even in the long-term.

Streaming Box	Time (button → movie playback)	OS & CPU
Roku Ultra (2024) - cold start	19 seconds	Roku OS & ??
Roku Ultra (2024) - cached start	14 seconds	Roku OS & ??
Roku Ultra (2022)	17 seconds	Roku OS & Cortex-A55
Apple TV 4K (2021)	8 seconds	Apple tvOS & A12 Bionic

I don't know about Roku, but historically OS navigation on Amazon Fire TV and Chromecast and most of the TV OSes have sucked in comparison to Apple TV. The latest non-ATV device have gotten better, but slow down with time. The Apple TV as you say still feels reasonably speedy after some OS updates. That's the way it should be IMO considering that OS navigation is such an important metric with these devices.

 

[Eug]

New Macs Nov. 1:

First, I expect Apple to announce the following products around the end of October. The company is currently aiming to then release at least some of them on Friday, Nov. 1:

- Low-end 14-inch MacBook Pro with M4 chip (code-named J604).
- High-end 14-inch and 16-inch MacBook Pros with M4 chips (J614 and J616).
- Revamped Mac mini in M4 and M4 Pro chip configurations (J773).
- Refreshed iMac with an M4 chip (J623).
- Refreshed iPad mini (J410).

I expect Apple to kick off 2025 with several other new devices in the first half of the year:

- 13-inch and 15-inch MacBook Air models with M4 (J613 and J615).
- Revamped iPhone SE (V59).
- Refreshed 11-inch and 13-inch iPad Air models (J607 and J637).
- New Magic Keyboards for the updated iPad Air line (R307 and R308).
- An upgraded AirTag item finder accessory (B589).

That’s a lot of stuff. New versions of the Mac Studio and Mac Pro with M4 chips are also in development, but they may take longer to arrive. I expect the Mac Studio launch to be closer to the middle of the year, with the Mac Pro going on sale sometime in the second half of 2025.

 

[poke01]

The keynote should be on October 22 or thereabouts

 

[poke01]

First Geekbench score with M4 on a Mac.
https://browser.geekbench.com/v6/cpu/8171874

 

[poke01]

This video confirms a lot of things. Space black comes to base pro and 16GB is standard and it now has 3 thunderbolt ports on the base

 

[Eug]

First Geekbench score with M4 on a Mac.
https://browser.geekbench.com/v6/cpu/8171874

And it has 16 GB RAM. Hopefully that is the base RAM configuration. If it is the base RAM configuration, is the next step up 24 GB or 32 GB?

 

[poke01]

And it has 16 GB RAM. Hopefully that is the base RAM configuration. If it is the base RAM configuration, is the next step up 24 GB or 32 GB?

Mark did say 16GB and 32GB was tested, so I’m leaning 32GB

 

[Eug]

This video confirms a lot of things. Space black comes to base pro and 16GB is standard and it now has 3 thunderbolt ports on the base

Hmmm... That video suggests he is the one that uploaded that M4 MacBook Pro Mac16,1 Geekbench 6.3 score of 3864/15288.

How legit is this guy?

 

[poke01]

Hmmm... That video suggests he is the one that uploaded that M4 MacBook Pro Mac16,1 Geekbench 6.3 score of 3864/15288.

How legit is this guy?

Well it is a legit GB entry. He’s probably based in Russia where Apple can’t do much there. Everything matches, so we got only a couple more weeks till Apple confirms.

 

[mvprod123]

First Geekbench score with M4 on a Mac.
https://browser.geekbench.com/v6/cpu/8171874

https://browser.geekbench.com/v6/compute/2893034 OpenCL