Discussion Apple Silicon SoC thread

Eug · Nov 10, 2020

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

igor_kavinski · Mar 13, 2025

Yes, yes but can it run x86 Crysis?

Nothingness · Mar 14, 2025

igor_kavinski said:
Yes, yes but can it run x86 Crysis?

Yes:

https://www.reddit.com/r/macgaming/comments/1j6yqjb/crysis_3_remastered_m4_pro_16c_gpu_dxmt_60_fps_4k

name99 · Mar 14, 2025

mvprod123 said:
M3 Ultra Blender CPU
View attachment 119765

Is Blender CPU of any interest other than as a dick-measuring contest?
What I mean is: does anyone actually do real Blender work using the CPU (I don't know, maybe because they're using some weird material that can more easily be represented using CPU code, or they're testing new algorithms, for water or sub-surface scattering or whatever)?

Obviously for Team Apple it's nice to get a respectable showing, but does it actually MATTER? (As opposed to eg the GPU result.)

Nothingness · Mar 14, 2025

name99 said:
Is Blender CPU of any interest other than as a dick-measuring contest?
What I mean is: does anyone actually do real Blender work using the CPU (I don't know, maybe because they're using some weird material that can more easily be represented using CPU code, or they're testing new algorithms, for water or sub-surface scattering or whatever)?

Obviously for Team Apple it's nice to get a respectable showing, but does it actually MATTER? (As opposed to eg the GPU result.)

I wonder the same. The same remark applies to Cinebench.

poke01 · Mar 14, 2025

name99 said:
Is Blender CPU of any interest other than as a dick-measuring contest?

Yes it’s actually part of SPEC CPU 2017 for a reason.

You can measure generational improvement for 3D applications

The Hardcard · Mar 14, 2025

name99 said:
Is Blender CPU of any interest other than as a dick-measuring contest?
What I mean is: does anyone actually do real Blender work using the CPU (I don't know, maybe because they're using some weird material that can more easily be represented using CPU code, or they're testing new algorithms, for water or sub-surface scattering or whatever)?

Obviously for Team Apple it's nice to get a respectable showing, but does it actually MATTER? (As opposed to eg the GPU result.)

I don’t know if current GPUs have changed the game, but for many years while GPUs were appreciated for time constraints, if you had the time top quality renders were done on CPU. For both video and 3D work. People wanted powerful GPUs for workstations for humans to design the scenes as smoothly as possible, but then the render farms would be a sea of CPUs grunting it out.

But that was many years ago. I haven’t been following recently.

Nothingness · Mar 14, 2025

The Hardcard said:
I don’t know if current GPUs have changed the game, but for many years while GPUs were appreciated for time constraints, if you had the time top quality renders were done on CPU. For both video and 3D work. People wanted powerful GPUs for workstations for humans to design the scenes as smoothly as possible, but then the render farms would be a sea of CPUs grunting it out.

But that was many years ago. I haven’t been following recently.

It looks like Avatar 2 was rendered on CPUs only: https://www.datacenterdynamics.com/...of-water-was-rendered-in-amazon-web-services/

But I'm not sure what's needed for film rendering is the same as what the typical end-user needs on his PC for making 3d stuff. I guess even early scene design for films are done with GPU acceleration before going to the render farms.

Mopetar · Mar 14, 2025

CPU render will still produce the absolute best quality, but unless it's the final render or close enough to it, using GPUs is going to be a lot faster, which is a lot better when you want a fast turn around to try different things out or test a proof of concept.

Maybe Cameron is the kind of perfectionist that would insist on only using CPU renders, but few can afford to be that picky. The Oscar winner for best animated film this year was made in Blender. I don't know what they used to work on or produce the final render, but they didn't have a large budget.

MS_AT · Mar 14, 2025

Nothingness said:
But I'm not sure what's needed for film rendering is the same as what the typical end-user needs on his PC for making 3d stuff.

What is cheaper, having 32 GB of RAM or 32GB of VRAM, especially if you are a hobbyist that does not mind running the render during the night as long as you can run it at all?

This is just my guess, I don't have any render experience, but usually the amount of memory is the reason people prefer CPU to GPU for things that GPU is generaly faster at.

poke01 · Mar 15, 2025

If you’re not gaming and this is the laptop to get.
$849 at Micro centre, for the M4 MacBook Air is a excellent deal. No other laptop comes close especially for web applications.

Apple MacBook Air 13" MW123LL/A (Early 2025) 13.6" Laptop Computer - Midnight; Apple M4 10-Core CPU; 16GB Unified - Micro Center

Get it now! MacBook Air is the worlds most popular laptop for a reason. Actually, for a lot of reasons. It delivers up to 18 hours of battery life. The M4 chip unlocks a whole new level of performance for work and play. Apple Intelligence is built in to help you get things done effortlessly.

www.microcenter.com

johnsonwax · Mar 15, 2025

Mopetar said:
CPU render will still produce the absolute best quality, but unless it's the final render or close enough to it, using GPUs is going to be a lot faster, which is a lot better when you want a fast turn around to try different things out or test a proof of concept.

Maybe Cameron is the kind of perfectionist that would insist on only using CPU renders, but few can afford to be that picky. The Oscar winner for best animated film this year was made in Blender. I don't know what they used to work on or produce the final render, but they didn't have a large budget.

So, friend of ours is in this industry - has an Oscar on his mantle, gave the speech and all that. We were talking about the tech side of things a few years ago and he said that they didn't rely on GPUs for final work (they would also do selective high quality renders this way to see how scenes will look in the final result). Main reason he cited was that the high resolution, max effects render was too RAM intensive for GPUs available at the time, so you wanted a lot of parallel compute hanging off a huge bank of fast RAM with pretty fast interconnects to move scene data in and out. At that point in production you can throw money at renting that kind of stuff, but you can't really afford delays, and GPUs really complicated the process because they aren't predictable in their output without nerfing the drivers so much that they stop being very fast. And they're much harder to rent. But for the equivalent of dailies, that was just cranked out in house using GPUs. Artifacts were fine there since you weren't usually looking at a complete rendered scene anyway. (Hopefully I didn't misunderstand this stuff too much).

Now, he wasn't an animation studio, so maybe they're different - he was ILM and Digital Domain doing VFX work for live action, so your source content is probably 8K RAW.

He described a similar reasoning to what a different friend of mine who was in charge of Battlenet when WoW first launched (we didn't see him for like 2 years - he rarely left the office) on how they were building out capacity and weren't focused on the fastest hardware or cutting edge. They could get capacity up fastest by focusing on the simplest, most reliable hardware to source. Running twice as many servers was an easier problem to solve than half as many cutting edge ones. They had more vendors they could source from, they were building out multiple data centers around the country (and world, Wow launched in Korea during that period), and his engineers were basically waiting at the factories to take delivery of hardware so the bottleneck wasn't engineer time, it was compute/manufacturing hour. What could HP throw off the assembly line fastest, and that was last gen hardware because there were no supply chain bottlenecks for it. This was before there was an AWS to go rent.

jdubs03 · Mar 17, 2025

A20 Chip for iPhones Said to Remain 3nm, But With Key Upgrade for Apple Intelligence

While the iPhone 17 series is still around six months away, a rumor pertaining to next year's iPhone 18 series has already surfaced. In a...

www.macrumors.com

I’m not really buying this rumor. Two years on N3P. That would be a huge fail on the part of TSMC.

poke01 · Mar 17, 2025

jdubs03 said:
A20 Chip for iPhones Said to Remain 3nm, But With Key Upgrade for Apple Intelligence

While the iPhone 17 series is still around six months away, a rumor pertaining to next year's iPhone 18 series has already surfaced. In a...

www.macrumors.com

I’m not really buying this rumor. Two years on N3P. That would be a huge fail part of TSMC.

Maybe N2 is for the M series. Or Apple is going straight to N2P with A21. Would interesting to see what Qualcomm does.

jdubs03 · Mar 18, 2025

poke01 said:
Maybe N2 is for the M series. Or Apple is going straight to N2P with A21. Would interesting to see what Qualcomm does.

Gurman was saying last week that next years iPhone Pro series are supposed to be a design change (even more than this camera bar for this year). And it’s the 20th year of iPhone. I’d be super surprised if they held off 2nm for a milestone product release such as the iPhone 18s. Particularly because it would’ve been the second year of holding off on 2nm. Plus the iPhone fold will probably be coming out as well. I’d think they would want a very best for both products. On the other hand, if this does happen, it’ll force Apple to improve on the architecture.

poke01 · Mar 18, 2025

jdubs03 said:
Gurman was saying last week that next years iPhone Pro series are supposed to be a design change (even more than this camera bar for this year). And it’s the 20th year of iPhone. I’d be super surprised if they held off 2nm for a milestone product release such as the iPhone 18s. Particularly because it would’ve been the second year of holding off on 2nm. Plus the iPhone fold will probably be coming out as well. I’d think they would want a very best for both products. On the other hand, if this does happen, it’ll force Apple to improve on the architecture.

Yeah Jeff Pu was very accurate, he knew about A20 getting advanced packaging, so it’s likely it’s on N3P.

Unless some other credible leakers say otherwise I would assume it’s on N3P for A20.

jdubs03 · Mar 18, 2025

poke01 said:
Yeah Jeff Pu was very accurate, he knew about A20 getting advanced packaging, so it’s likely it’s on N3P.

Unless some other credible leakers say otherwise I would assume it’s on N3P for A20.

That would definitely be a delay and perhaps open up some daylight for Intel with achieving process leadership. I see a lane opening there if Intel doesn’t own goal themselves.

Though like you said with Qualcomm, it’ll be interesting to see what they do with a Gen 3 Elite, in 22 months from now.

mikegg · Mar 18, 2025

jdubs03 said:
A20 Chip for iPhones Said to Remain 3nm, But With Key Upgrade for Apple Intelligence

While the iPhone 17 series is still around six months away, a rumor pertaining to next year's iPhone 18 series has already surfaced. In a...

www.macrumors.com

I’m not really buying this rumor. Two years on N3P. That would be a huge fail on the part of TSMC.

What? I thought everyone knew that A19 will be on N3 family.

Apple stayed on N5 family for 3 years as well. TSMC's cadence seems to be 3 years now.

N2 starts volume production in the second half of 2025. In order to make the schedule for the next iPhone, it needs to start production basically now. It's not in high volume production now.

okoroezenwa · Mar 18, 2025

mikegg said:
What? I thought everyone knew that A19 will be on N3 family.

Rumour is about A20 however. That was expected to be N2.

johnsonwax · Mar 18, 2025

I wouldn't be surprised if Apple split the base and pro SOCs again, where we get a minor rev on the base and a whole new SOC for the pro. So much of this stuff comes down to what kind of volume TSMC can handle. At 200M+ units per device cycle, taking ~100M off of TSMCs newest process appears to be necessary depending on the timing and their ability to ramp.

mikegg · Mar 18, 2025

okoroezenwa said:
Rumour is about A20 however. That was expected to be N2.

Oh ok. I was confused because the person I quoted said N3 for 3 years so I assumed we were talking about the iPhone 17 since we've already had N3 for 2 years.

I think iPhone 18 will use N2.

mvprod123 · Mar 18, 2025

A Japanese tech site has published an analysis of the M4 Pro/Max Macs. There may be mistakes in the translation.

2025”N‚à¸—Í“I‚ÉV»•i‚ð”•\‚µ‚Ä‚¢‚éAppleB¡‰ñ‚ÍA2024”N‚©‚ç2025”N‚É‚©‚¯‚Ä””„‚³‚ê‚½uMac miniv‚âuMac Studiov‚ðŽæ‚èã‚°A‚»‚ê‚ç‚É“‹Ú‚³‚ê‚Ä‚¢‚éƒvƒƒZƒbƒTuM4 ProvuM4 Maxv‚ð•ñ‚·‚éB

eetimes.itmedia.co.jp

Apple is not only focused on developing digital processors but also actively designs peripheral chips—such as power management ICs and interface components—to maximize processor performance. By integrating these elements, the company creates complete chipsets as total systems. It's important to recognize that Apple is not just a processor manufacturer but a comprehensive semiconductor developer.
While future Apple products from 2025 will be covered in upcoming reports, this article focuses on the "Mac mini" released in November 2024, along with the new "M4 Pro" and "M4 Max" processors found in the MacBook Pro.
Figure 1 shows the Mac mini, which was launched on November 8, 2024. Compared to its predecessor, the device's maximum volume has been reduced from 1389mm³ to 807mm³, and its weight has decreased from 1.28kg to 0.73kg—a reduction of approximately 40%. The device is now compact enough to fit in the palm of a hand.
The previous Mac mini, released in February 2023, featured the "M2 Pro" chip manufactured using TSMC’s 5nm process. In contrast, the 2024 version is powered by the M4 Pro, produced using a 3nm process, enabling a smaller form factor. The teardown difficulty level is not particularly high.
Figure 1 also depicts the main logic board of the Mac mini (separate from the power board) and the connected SSD. The SSD has a unique form factor, differing from standard M.2 SSDs. Additionally, the SSD features a dual-sided design, with memory chips mounted on both sides.

Figure 2 shows the chip teardown of the Mac mini's SSD. In the 512GB model, four memory chips (dual-sided) are installed, with each chip holding 128GB. Inside one package, there are three silicon components: two 64GB NAND flash memory chips from Kioxia and one SSD controller chip developed by Apple. The controller chip provides SSD control functions and the interface.
In typical SSDs, the entire SSD is managed by a single controller chip. However, Apple uses a controller for each package. This reflects Apple's unique approach to memory control (details omitted here).
In Apple products, even when memory chips from manufacturers such as Samsung Electronics, SK hynix, or Micron Technology are used, Apple always pairs them with Apple-designed controllers.

Figure 3 shows the Apple chipset of the latest Mac mini (a few additional chips are omitted). Surrounding the M4 Pro processor, there are a total of six power ICs. These ICs manage the power voltage and other related functions.

Apple acquired the power IC business of the British company Dialog Semiconductor (now Renesas Electronics) in 2018, and since then, has been integrating power ICs with processors in all of its products (iPhone, Apple Watch, AirPods). By implementing finer power control, Apple has improved both power efficiency and computational performance. Transistors increase speed with higher voltage, but excessive voltage can lead to setup issues. By centralizing processor and power development, Apple achieves highly consistent power management. Qualcomm and MediaTek are also engaging in similar developments.
Furthermore, Apple introduced the Thunderbolt 5 interface in its products released in November 2024. Thunderbolt 5 offers transmission speeds more than twice as fast as the previous Thunderbolt 4. Since Thunderbolt 4, Apple has been using its custom-developed interface chips, and in the November 2024 Mac mini and MacBook Pro, Apple-designed Thunderbolt 5 is integrated with more ports. From the input/output interfaces to power, memory control, and processors, everything is Apple-designed.
Figure 4 shows the circuit board of the previous generation Mac mini from February 2023 with the M2 Pro processor and the new 2024 Mac mini with the M4 Pro. The board area is about two-thirds the size of the previous version, though the processor package size remains the same. The M2 Pro used a 5nm process, while the M4 Pro is manufactured using a 3nm process, so the processor is smaller, but that does not mean the board shrank due to the smaller processor. The reduction in size of the Mac mini was achieved by redesigning the entire board. It is clear that semiconductor evolution, such as optimization of power ICs and lower processor voltages, enabled the overall reduction of the board size. Apple has always been adjusting its board designs based on semiconductor characteristics.

Figure 5 shows the MacBook Pro equipped with the M4 Max, released in November 2024. It is the successor to the MacBook Pro with the M3 Max, which was released in November 2023. The internal structure, design, and board size are almost identical to the 2023 model. The major changes are the replacement of the M3 Max processor with the M4 Max and the upgrade of the interface from Thunderbolt 4 to Thunderbolt 5.

Figure 6 shows the back side of the processor on the MacBook Pro's circuit board. Five ceramic capacitors and power ICs are installed to stabilize the power supply (cross-sectional analysis has also been conducted). The chip's silicon is engraved with the model name of the Apple chip. This is custom-designed silicon by Apple. Apple not only develops power control ICs but also develops silicon capacitors used in devices like the iPhone. Apple’s chip development integrates both functionality and performance.

Figure 7 compares the heat sinks of the 2023 M3 Max-equipped MacBook Pro and the 2024 M4 Max-equipped MacBook Pro. In the M3 Max, the heat sink only covered the processor area, but in the M4 Max, the heat sink has been redesigned to also cover the unified memory. This change reflects an improvement in heat management that includes the memory. Apple products are continuously undergoing minor design changes.

Figure 8 shows the chip disassembly of the M4 Pro and M4 Max (detailed internal photos of the wire layer peeling are omitted). The M4 Pro is announced to have 10 Performance CPU (PCPU) cores and 4 Efficiency CPU (ECPU) cores, but the actual silicon has 12 PCPU cores, the same number as the M4 Max. Two-thirds of the silicon is identical between the two. The M4 Pro + 20-core GPU equals the M4 Max. By commonizing the design assets, development efficiency is improved. By commonizing the CPU and NPU, tests can also be reused. The upper structure of the M4 Max is not shown, but it will be analyzed along with the M3 Ultra in the near future and covered later. Apple consistently develops three types of high-middle-entry silicon in the M series, and by maximizing the reuse of internal IP (Intellectual Property), they create a scalable architecture.

Table 1 compares the appearance of the PCPU (Performance CPU) of Apple chips from 2024 to 2025, which utilize the second-generation 3nm TSMC N3E process. From the A18 to the M4 Max, the PCPU is the same, with only the number of cores varying. The GPU, NPU, and ECPU are also similar. Each core is carefully designed, and scalability is achieved through the number of cores.

igor_kavinski · Mar 18, 2025

This just in. Apple HQ is a glorified torture chamber for hw engineers: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...a2-architectures-thread.2589999/post-41418454

name99 · Mar 18, 2025

Is the SSD design they describe new?
We always talk about the SSD "controller" being part of the SoC.

But it seems like Apple actually has kinda a split SSD controller, perhaps a part that handles overall transaction scheduling and data movement (maybe even crypto) within the SoC, and then attached to each flash package a lower level controller that handles timing and hand-holding of those particular flash chips.

Or is the standard way all SSDs are done now: a larger "master" controller and a bunch of small controllers that handle each "stick" (about 512GB or so) of flash chips?

I'm guessing the lack of DRAM is because the Apple (on SoC? maybe the per flash package?) SSD controller(s) use system DRAM for the relevant purposes (FTL, holding pages while erasing blocks, that sort of thing)?

Doug S · Mar 18, 2025

name99 said:
Is the SSD design they describe new?
We always talk about the SSD "controller" being part of the SoC.

But it seems like Apple actually has kinda a split SSD controller, perhaps a part that handles overall transaction scheduling and data movement (maybe even crypto) within the SoC, and then attached to each flash package a lower level controller that handles timing and hand-holding of those particular flash chips.

Or is the standard way all SSDs are done now: a larger "master" controller and a bunch of small controllers that handle each "stick" (about 512GB or so) of flash chips?

I'm guessing the lack of DRAM is because the Apple (on SoC? maybe the per flash package?) SSD controller(s) use system DRAM for the relevant purposes (FTL, holding pages while erasing blocks, that sort of thing)?

My understanding is that the chips in the NAND package are PCIe/flash bridges. They'd need a lot of wires for the full NAND interface that's typically required for a controller to talk to flash chips, so they are essentially tunneling that interface via PCIe.

I expect these guys are just confused by the custom Apple chip they see and for whatever reason are assuming it does more than it actually does.

Doug S · Mar 18, 2025

jdubs03 said:
A20 Chip for iPhones Said to Remain 3nm, But With Key Upgrade for Apple Intelligence

While the iPhone 17 series is still around six months away, a rumor pertaining to next year's iPhone 18 series has already surfaced. In a...

www.macrumors.com

I’m not really buying this rumor. Two years on N3P. That would be a huge fail on the part of TSMC.

A20 on N3P, A20P on N2. N2 will cost more, so the base iPhone won't get it right away.

The article even mentions the A19/A19P distinction but they don't mention A20P at all. I think they know what I'm saying here is true but know a headline like that will drive more clicks, then they can post something else next week saying "iPhone 18 will get N2 after all, but only the Pro models" so they can drive the next round of clicks.

Really sucks what the engagement based advertising model has done to journalism.

Discussion Apple Silicon SoC thread

Lifer

Lifer

Diamond Member

Senior member

Diamond Member

Diamond Member

Senior member

Diamond Member

Diamond Member

Senior member

Diamond Member

Member

Golden Member

Diamond Member

Golden Member

Diamond Member

Golden Member

Golden Member

Member

Member

Golden Member

Senior member

Lifer

Senior member

Diamond Member

Diamond Member