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

 

[DZero]

After seeing the C lineup. No one is going to beat the M5 series in early 2026.

CPU wise totally agree, but GPU wise seems that there is competition and is reaching to so old Notebook processors (GTX 1050M) and that's interesting.

 

[poke01]

oh no QC will be fine

Even GPU?

 

[Eug]

9to5mac has a list of the Apple Silicon M series chip release dates.

Every Apple silicon M-series chip, product, and release date

M-series Apple silicon chips tout impressive performance but unpredictable release cadence. That makes it hard to keep up with which...

9to5mac.com

Let's see what schedule they're on:

M1: November 10, 2020
M1 Pro: October 18, 2021
M1 Max: October 18, 2021
M1 Ultra: March 8, 2022

M2: June 6, 2022 (+18.9 months)
M2 Pro: January 17, 2023 (+15.0 months)
M2 Max: January 17, 2023 (+15.0 months)
M2 Ultra: June 5, 2023 (+15.0 months)

M3: October 30, 2023 (+16.8 months)
M3 Pro: October 30, 2023 (+9.4 months)
M3 Max: October 30, 2023 (+9.4 months)
M3 Ultra: March 5, 2025 (+21.0 months)

M4: May 7, 2024 (+6.2 months)
M4 Pro: October 30, 2024 (+12.0 months)
M4 Max: October 30, 2024 (+12.0 months)
M4 Ultra: --

I'll ignore Mx Pro/Max/Ultra for the time being.

Most pundits M5 Macs will come out in early 2026.

If M5 is released say March 10 (+18.1 months), then Mx chips release times would average (18.9 + 16.8 + 6.2 + 22.1) / 4 = 16.0 months.

However, Ming-Chi Kuo says iPad Pro M5 production should have started in 2025 H2, with other outlets suggesting a late October to early November 2025 launch.

If M5 is released say November 11 (+18.1 months), then Mx chips release times would average (18.9 + 16.8 + 6.2 + 18.1) / 4 = 15.0 months.

That's longer than the 12 month prediction some have made, but on the low end of my 15-18 month guess. However, one wonders if Apple still may be targeting a yearly release cycle in the future, as some here have guessed, given the short M3 to M4 cycle. Also, will there will ever be an M4 Ultra?

 

[poke01]

Even GPU?

I take the silence as Qualcomm GPUs are still bad for compute and AAA gaming on Windows.

 

[adroc_thurston]

I take the silence as Qualcomm GPUs are still bad for compute and AAA gaming on Windows.

I mean you're not doing anything useful with non-AMD/NV GPUs anywhere, really.

 

[Doug S]

I take the silence as Qualcomm GPUs are still bad for compute and AAA gaming on Windows.

Its a niche market, why should developers expend any effort optimizing games for 1-2% of the Windows market? They'd rather put that effort towards starting development of another title, or sequels/expansions of current successful ones. If they were about chasing small markets they'd port to Mac before they'd port to Qualcomm GPU on ARM Windows.

Especially since that "ARM Windows" market, to the extent it is even a thing, is soon going to be split up among other contenders like Mediatek who have their own GPUs. Further fragmenting an already niche market.

 

[adroc_thurston]

Its a niche market, why should developers expend any effort optimizing games for 1-2% of the Windows market?

Well it's not about that, QC GPU uarch and driver stack are just poo poo for running Actual Real video games.

Especially since that "ARM Windows" market,

Does not exist.

 

[mvprod123]

Good research of Apple AMX/SME

 

[mvprod123]

A19 die shot

https://twitter.com/x/status/1970533400151818256

 

[DZero]

I mean you're not doing anything useful with non-AMD/NV GPUs anywhere, really.

And even nVIDIA is way more optimized than AMD

 

[adroc_thurston]

And even nVIDIA is way more optimized than AMD

where? in video games?

 

[DZero]

where? in video games?

Videogames and AI. Specially the latter.

AMD vs NVIDIA Inference Benchmark: Who Wins? – Performance & Cost Per Million Tokens

It has been long claimed that AMD’s AI servers can achieve better inference performance per total cost of ownership (TCO) than Nvidia. We have spent the past six months investigating and validating…

semianalysis.com

 

[name99]

9to5mac has a list of the Apple Silicon M series chip release dates.

Every Apple silicon M-series chip, product, and release date

M-series Apple silicon chips tout impressive performance but unpredictable release cadence. That makes it hard to keep up with which...

9to5mac.com

Let's see what schedule they're on:

M1: November 10, 2020
M1 Pro: October 18, 2021
M1 Max: October 18, 2021
M1 Ultra: March 8, 2022

M2: June 6, 2022 (+18.9 months)
M2 Pro: January 17, 2023 (+15.0 months)
M2 Max: January 17, 2023 (+15.0 months)
M2 Ultra: June 5, 2023 (+15.0 months)

M3: October 30, 2023 (+16.8 months)
M3 Pro: October 30, 2023 (+9.4 months)
M3 Max: October 30, 2023 (+9.4 months)
M3 Ultra: March 5, 2025 (+21.0 months)

M4: May 7, 2024 (+6.2 months)
M4 Pro: October 30, 2024 (+12.0 months)
M4 Max: October 30, 2024 (+12.0 months)
M4 Ultra: --

I'll ignore Mx Pro/Max/Ultra for the time being.

Most pundits M5 Macs will come out in early 2026.

If M5 is released say March 10 (+18.1 months), then Mx chips release times would average (18.9 + 16.8 + 6.2 + 22.1) / 4 = 16.0 months.

However, Ming-Chi Kuo says iPad Pro M5 production should have started in 2025 H2, with other outlets suggesting a late October to early November 2025 launch.

If M5 is released say November 11 (+18.1 months), then Mx chips release times would average (18.9 + 16.8 + 6.2 + 18.1) / 4 = 15.0 months.

That's longer than the 12 month prediction some have made, but on the low end of my 15-18 month guess. However, one wonders if Apple still may be targeting a yearly release cycle in the future, as some here have guessed, given the short M3 to M4 cycle. Also, will there will ever be an M4 Ultra?

Seems like the bigger determinant is N2.
N2 volume manufacturing is happening around now. IF we believe Apple wants to exploit that ASAP then the likely way this plays out is that M5 series are on N2, and being manufactured right now.

Whether that translates into iPads first or MacBooks first may well depend on other things
- Does Apple want the N1 across the whole line, or only in iOS devices for now?
- Are they waiting on LPDDR6 for the Macs?
- Is some sort of new camera or screen scheduled for this release?

Point is, I suspect Apple are not as tied to a particular rhythm as we might think; they have a set of technologies they'd like in the next generation, and create a window (which may be 6..8 months wide) to try to get all the elements lined up.

Even for the SoCs, which would seem to be something over which they have total control, they've been surprisingly willing (more so than I expected at first) to mix and match elements across generations, for example the M3 ANE (18 TOPs) came at the same sort of generation as the A17, but does not get the A17's 35 TOPs.

I could especially believe that they are willing to delay Macs this year if they have some sort of AI surprise planned. Maybe they really want LPDDR6 for higher bandwidth? Maybe they have something unexpected, like addition of an MRAM block, to hold weights and allow for much larger MoE's?

 

[adroc_thurston]

Videogames

No.

AI. Specially the latter.

Kind of? Refrain from poasting dolanslop at least.

 

[johnsonwax]

Even for the SoCs, which would seem to be something over which they have total control, they've been surprisingly willing (more so than I expected at first) to mix and match elements across generations, for example the M3 ANE (18 TOPs) came at the same sort of generation as the A17, but does not get the A17's 35 TOPs.

I think this is simply because AI was not 'the thing' when M3 and A17 were being designed. ChatGPT first launched 2 years 10 months ago (yes, the AI revolution is less than 3 years old). M3 was already designed then. So Apples AI ambition in the Mac was extremely narrow while the M3 was being designed, while the NPU in the A17 was there for doing pretty serious computational photography/FaceID, in realtime or near-realtime, so it couldn't really be modest and still work. Again, Apple shipped NPUs in A series starting in 2017 - 8 years ago - back in the A11. It's doing an ENTIRELY different thing. The biggest use case for ANE was Face ID. Still is. And I wouldn't be surprised if more TOPS-hours were spent doing FaceID than all chatbots combined.

By launch, the M3 sort of mismatched to where the market had quickly gotten to, but it's not like anyone else in the PC space was shipping an NPU in their CPU - Apple was still the only one even offering an NPU. Everyone was doing their AI on GPU. x86 didn't get there until the M4 got there. They had the same timing problem.

 

[Mopetar]

It's a place to put A19 chips that don't hit parametric yield for power/thermal that can still turn a profit.

Apple can adjust the specs to hit whatever targets they want as far as parametric yields are concerned. If they need to have 20 million chips for the iPhone release they'll adjust the clock and power targets until they will get 20 million chips that clear those hurdles.

The AppleTV needs updates so seldomly that they could start binning chips for it years in advance. They have enough historical sales data that they could plan something that far out.

Its a niche market, why should developers expend any effort optimizing games for 1-2% of the Windows market?

Some developers don't even optimize for the other 98% (looking at you Gearbox) so expecting them to do anything for Qualcomm is right out.

 

[Doug S]

Seems like the bigger determinant is N2.
N2 volume manufacturing is happening around now. IF we believe Apple wants to exploit that ASAP then the likely way this plays out is that M5 series are on N2, and being manufactured right now.

Whether that translates into iPads first or MacBooks first may well depend on other things
- Does Apple want the N1 across the whole line, or only in iOS devices for now?
- Are they waiting on LPDDR6 for the Macs?
- Is some sort of new camera or screen scheduled for this release?

Point is, I suspect Apple are not as tied to a particular rhythm as we might think; they have a set of technologies they'd like in the next generation, and create a window (which may be 6..8 months wide) to try to get all the elements lined up.

Even for the SoCs, which would seem to be something over which they have total control, they've been surprisingly willing (more so than I expected at first) to mix and match elements across generations, for example the M3 ANE (18 TOPs) came at the same sort of generation as the A17, but does not get the A17's 35 TOPs.

I could especially believe that they are willing to delay Macs this year if they have some sort of AI surprise planned. Maybe they really want LPDDR6 for higher bandwidth? Maybe they have something unexpected, like addition of an MRAM block, to hold weights and allow for much larger MoE's?

When I heard the rumors about M5 Macs not launching until next year I immediately posted wondering if that was about the timing of N2. I can't really see any other reason why Apple would wait, since an N3P based M5 cannot be the source of the delay. At least there's nothing chip related I can see, it would have to something else like switching to OLED displays that affected the launch timing - but if they were waiting until next year for that they might as well do N2 as well.

LPDDR6 is not going to be the reason for delay, nor is it going to deliver more bandwidth out of the gate. The initial JEDEC speed is -10666, and you can already get LPDDR5X at that speed. I don't know about vendor plans, maybe they will skip forward to -12800 at launch but regardless it isn't even in mass production yet. Do you really believe Apple would introduce that level of third party schedule risk just to get a (maybe) small bump in bandwidth?

 

[poke01]

A19 die shot

https://twitter.com/x/status/1970533400151818256

So it’s smaller than A18 while being faster than A18 Pro. Wow, maybe nvidia could learn a thing or two.

 

[adroc_thurston]

So it’s smaller than A18 while being faster than A18 Pro. Wow, maybe nvidia could learn a thing or two.

You do understand that client Blackwell dies are both smaller and faster?

 

[Mopetar]

If they wanted to launch a new A-series-based inexpensive Mac I could see Apple holding off on the M-series products to give that new product some time to breathe.

Minis don't have to wait for an OLED or anything else really. There's always a possibility they could launch an A-series laptop this fall alongside an M-series Mini while holding off the Pro chips and products until Spring.

I don't know how much I like the idea of them doing that, but it's certainly something they could do or is at least conceivable for them to do. Apple is always fairly predictable until they aren't.

 

[poke01]

You do understand that client Blackwell dies are both smaller and faster?

“Faster” they use more power…

 

[adroc_thurston]

“Faster” they use more power…

Laptop ones are ISO power (or lower in some cases) and they do passable perf bumps.

 

[DZero]

You do understand that client Blackwell dies are both smaller and faster?

Compared to AMD, is not that smaller, but indeed is faster.

 

[Eug]

There is a battery tray that spans the width of the iPhone, and that is where the vapour chamber is located, spreading heat evenly throughout the aluminum shell.

In this test the 16 Pro Max shell gets to just under 38C and the SoC throttles, whereas the 17 Pro Max shell gets to just under 35C and the SoC does not throttle.

 

[jpiniero]

So it’s smaller than A18 while being faster than A18 Pro. Wow, maybe nvidia could learn a thing or two.

N3P vs N3E tho.