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

 

[Mopetar]

Seems like a nice update, but there's a massive gap between the Pro and Max CPUs. I think the Pro could have had another 5-10 billion transistors thrown at it to help round out the lineup a bit better.

The iMac continues to disappoint as there's not even an option to go with a Pro chip. Add in $200 to get to a baseline respectable amount of RAM and it's only good for someone with a light workload who will keep it for 10 years.

I don't know what the supply situation is like or how much demand they anticipate, but is there much point in not releasing updates for the Mini as well? The Studio and Pro are understandably absent as there's no M3 Ultra yet, but why leave the Mini out?

I look forward to the eventual benchmarks and deep dives on these new chips and products. I'm not really swayed to update, but could be tempted to grab an M3 Mini or Studio to replace my Intel-based iMac down the road.

 

[Tigerick]

It seems reasonable to think about successor of N3E but you guys are forgetting logic density scaling. No process can beat N3B in logic density cause of 25 EUV layers.

Another reason I did not mention about is I am expecting Apple will use N3B for M4 series only. After that, Apple will switch to N2 process in 2026, cause in 2026 LPDDR6 will start hitting market. If I am Apple, I won't miss the first move opportunity to use LPDDR6 which should have double memory bandwidth. That's mean Apple could potentially increase CPU and GPU core counts by 50%. It is going to be biggest upgrade since M1 series. Of course, doubling core counts comes at power requirement, I am not even sure M5 Max will remain monolithic. Apple might switch to chiplet with M5 Max. No matter what design Apple going to choose, they have to use next gen process to make it: N2 with nanosheet timings fit in the design:

TSMC Outlines 2nm Plans: N2P Brings Backside Power Delivery in 2026, N2X Added To Roadmap

www.anandtech.com

That's why I am expecting Apple to release M4 Pro/Max by end of next year / early 2025. Apple has to adjust the release timings between M4 Pro/Max and M5 Pro/Max. It's far fetched I know, if you guys have other opinions, feel free to speculate: -

 

[Tigerick]

That's why don't expect much upgrade in M4 Max by transistor counts. Apple will quickly settle the design of M4 Max (maybe done as we speak) and move to totally new platform for M5 Max. I am excited to see what Apple will bring to the table in 2026...

 

[trivik12]

I think next year we will see N3E and 2025 will see N3P. N2 will be a 2026 product as production is late 2025 like N3B production was late 2022(and we saw products this year).

 

[jpiniero]

I think next year we will see N3E and 2025 will see N3P. N2 will be a 2026 product as production is late 2025 like N3B production was late 2022(and we saw products this year).

Current speculation is that Apple (and most companies really) is skipping N2 actually. Seems unlikely that TSMC would be able to get N2P in time for the 2026 iPhone but I suppose it's possible.

 

[repoman27]

It seems reasonable to think about successor of N3E but you guys are forgetting logic density scaling. No process can beat N3B in logic density cause of 25 EUV layers.

That's why I keep pointing out that N3S is the most likely node for Apple to move to. It's a density optimized variant of N3E scheduled to enter volume production in H1'24, right when Apple would need to start ramping the A18 for a September launch.

Also, N3P comes with a 4% optical shrink, which applies to SRAM and analog as well as logic. It works out to around 98% of the logic density and 99% of the SRAM density of N3.

 

[Tigerick]

That's why I keep pointing out that N3S is the most likely node for Apple to move to. It's a density optimized variant of N3E scheduled to enter volume production in H1'24, right when Apple would need to start ramping the A18 for a September launch.

Also, N3P comes with a 4% optical shrink, which applies to SRAM and analog as well as logic. It works out to around 98% of the logic density and 99% of the SRAM density of N3.

No, N3S is not even officially mentioned by TSMC. They even mentioned N3A by 2026. And 1.04x is most likely done by adding one EUV layer.

As I said, Apple will move to N2 with HVM in 2025 as mentioned by CC Wei, CEO of TSMC. Why did Apple wants to change design rules just for one generation while not having much benefits?

 

[repoman27]

No, N3S is not even officially mentioned by TSMC. They even mentioned N3A by 2026. And 1.04x is most likely done by adding one EUV layer.

N3S was announced during the TSMC Technology Symposium in 2022. They stopped talking about it because it's likely going to be another Apple only node. The 4% optical shrink for N3P is also straight from TSMC's mouth. It's just like the 6% shrink they did to N5P to create N4.

TSMC Readies Five 3nm Process Technologies, Adds FinFlex For Design Flexibility

www.anandtech.com

N3E Replaces N3; Comes In Many Flavors

TSMC outlines a number of 3-nanometer class node changes and flavors.

fuse.wikichip.org

 

[Doug S]

No, N3S is not even officially mentioned by TSMC. They even mentioned N3A by 2026. And 1.04x is most likely done by adding one EUV layer.

As I said, Apple will move to N2 with HVM in 2025 as mentioned by CC Wei, CEO of TSMC. Why did Apple wants to change design rules just for one generation while not having much benefits?

I agree, we heard about N3S once and then nothing ever again. Unless TSMC mentions it again I think it is no longer going to happen. Maybe it was designed as a follow-on for N3B but once TSMC couldn't fix the yields they decided that was a dead end and N3S disappeared.

Last I recall N2 was entering HVM in "H2 2025", and that probably means Q4 like with N3 so it too late for Apple. Apple needs HVM in Q2 to fit their schedule.

 

[FlameTail]

Will there be an M3 Extreme?

4 × M3 Max.

48P + 16E CPU
160c GPU
1600 GB/s memory bandwidth.

It has been established that the Ultra chip (which doubled the Max) isn't strong enough to rival the best in PC (RTX 4090, Threadripper/Xeon). The Extreme chip will be strong enough to rival them and also give a purpose for the Mac Pro- which at the moment seems to be a glorified Mac Studio.

 

[FlameTail]

Also some food for thought: if Apple upgraded the RAM to LPDDR5X, then the M3 Pro could have maintained it's predecessor's 200 GB/s bandwidth while shrinking the memory bus to 192-bit.

Because LPDDR5X-8533 has 33% more bandwidth than LPDDR5-6400.

 

[Ajay]

Well, aside from the Max Soc, this launch is disappointing. My next PC upgrade cycling, excluding minor improvements, will be in 2026. I'm getting to the point were having a PC mainly because I play some AA and AAA titles will be waning. I'm hoping the progress over the next couple of generation will be a bit better. I'd like to buy a Mac Mx Pro based Mac mini. If things continue as they are, I'm likely looking at a more entry level Mac Studio with an Mx Max SoC (32-48GB of DRAM and a 2-4 TB SSD). Of course, I could change my mind again and still stick with a Windows x86 PC. I've tried using Linux as my daily driver and find it too annoying to use on the desktop. We run everything else in the house on some form of IOS - 2 iPads, two iPhones and AppleTV. Having to buy some new packages of software won't we awesome, but that's the price of admission.

 

[Ajay]

Also some food for thought: if Apple upgraded the RAM to LPDDR5X, then the M3 Pro could have maintained it's predecessor's 200 GB/s bandwidth while shrinking the memory bus to 192-bit.

Because LPDDR5X-8533 has 33% more bandwidth than LPDDR5-6400.

Yeah, I noticed that as well. Seems like it had to be a cost cutting measure - or, less likely, a supply issue. Too bad, it would maximize the benefit of these new SoCs, but Apple choose to shoot a bit lower - just as many of the large Win/x86 OEM do as well.

 

[Henry swagger]

Geekbench Search - Geekbench

Apple m3 geekbench single core

 

[Gideon]

Geekbench Search - Geekbench

Apple m3 geekbench single core

@ 4.05 Ghz

So most of the uplift is essentially down to clock-speed?

vs M2 @ 3.48Ghz:

MacBook Air (15-inch, 2023) vs Mac15,3 - Geekbench

TBF, that still means the ST geekbench score is ever so slightly higher than the Snapdragon X Elite (even when considering the slight clock speed regression)

 

[jpiniero]

@ 4.05 Ghz

So most of the uplift is essentially down to clock-speed?

vs M2 @ 3.48Ghz

MacBook Air (15-inch, 2023) vs Mac15,3 - Geekbench

That was pretty much the case with A17. So no surprise there.

 

[FlameTail]

TBF, that still means the ST geekbench score is ever so slightly higher than the Snapdragon X Elite (even when considering the slight clock speed regression)

Compared to the X Elite running Windows?

X Elite running Linux does 3200 ST

 

[igor_kavinski]

Of course, I could change my mind again and still stick with a Windows x86 PC.

Forgetting about Windows on ARM?

 

[igor_kavinski]

X Elite running Linux does 3200 ST

That's not silent. You need hefty cooling capacity or very noisy fans to manage that.

 

[igor_kavinski]

Geekbench Search - Geekbench

Apple m3 geekbench single core

A big YAWN from the 7800X3D users

Gigabyte Technology Co., Ltd. B650 GAMING X AX vs Mac15,3 - Geekbench

 

[Gideon]

A big YAWN from the 7800X3D users

Gigabyte Technology Co., Ltd. B650 GAMING X AX vs Mac15,3 - Geekbench

I don't know, this is the cheapest design Apple makes also going to fanless laptops. A more appropriate comparison (including power) would be Ryzen 7840U:

https://browser.geekbench.com/v6/cpu/compare/3343681?baseline=3340381

EDIT, sorry (and screw AMDs naming scheme) i accidentally linked to a 7735U instead of a 7840U. Added a linux score (as it's overall closer to macos historically on intel cpus)

 

[poke01]

A big YAWN from the 7800X3D users

Gigabyte Technology Co., Ltd. B650 GAMING X AX vs Mac15,3 - Geekbench

While that's a great CPU. The 7800X3D is not a laptop SoC.

 

[igor_kavinski]

https://browser.geekbench.com/v6/cpu/compare/3343681?baseline=3340381

Different versions. Same version of GB6 here: https://browser.geekbench.com/v6/cpu/compare/3332440?baseline=3342825

 

[igor_kavinski]

While that's a great CPU. The 7800X3D is not a laptop SoC.

True but the 7800X3D users could easily limit the TDP and still get higher multi-core performance in a mini-ITX form factor with up to 96GB RAM, for a lot less than they have to pay for an M3 Mac Mini or Studio. M3 doesn't look that appealing.

 

[Henry swagger]

@ 4.05 Ghz

So most of the uplift is essentially down to clock-speed?

vs M2 @ 3.48Ghz:

View attachment 88196

MacBook Air (15-inch, 2023) vs Mac15,3 - Geekbench

TBF, that still means the ST geekbench score is ever so slightly higher than the Snapdragon X Elite (even when considering the slight clock speed regression)

Oryon has a ipc advantage compared to M3.. if fabbed on N3 oryon will do 4.7ghz because they can hit 4.3ghz on N4p