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

Eug · Monday at 1:18 PM

mvprod123 said:
I still believe the Mac Pro will be released this year based on the M5 Ultra. It would be terrible to postpone the Mac Pro upgrade for so long when the M5 MacBook Pro will be on sale at the end of the year.

Nothingness said:
Do you mean we'll likely only see M5 Ultra in the next Mac Studio (2 years from now?)?

It would be great if @mvprod123 is correct and the M5 Ultra comes out this year as Mac17,1. That would better justify the existence of the Mac Pro, but I worry that may be too optimistic.

okoroezenwa · Monday at 2:03 PM

mvprod123 said:
I still believe the Mac Pro will be released this year based on the M5 Ultra. It would be terrible to postpone the Mac Pro upgrade for so long when the M5 MacBook Pro will be on sale at the end of the year.

It indeed would be terrible. To reconcile both the ‘M5 Mac Pro will exist this year’ and ‘M5 Mac Pro will be released at least next year’ I propose: M5 Mac Pro will be announced at WWDC (or even the October laptop refresh mini-event) and set to release early next year.

Doug S · Monday at 4:34 PM

mvprod123 said:
I still believe the Mac Pro will be released this year based on the M5 Ultra. It would be terrible to postpone the Mac Pro upgrade for so long when the M5 MacBook Pro will be on sale at the end of the year.

The Ultra has always followed the Max launch by months, it would be a big change to launch them alongside one another. If they ship a new Ultra this year I would bet on it being M4 Ultra not M5.

Maybe they can address whatever the reasons (yield/testing, packaging, verification?) that cause such significant delays with the Ultra launches, who knows. But the recent launch of an M3 Ultra months after M4 Max appeared doesn't seem to indicate they have yet done so. Maybe if/when they go to more of a chiplet based design an Ultra type part will be just as easy to make as a Max type part if both are built similarly from chiplets with the only real difference that the Ultra part contains more chiplets.

I agree it is bad that the Mac Pro lags. Granted the sales volumes probably aren't that large but that's their halo product and it is a bad look having it out of date compared to cheaper offerings.

I don't know what sort of volume TSMC has available for the special packaging used for Ultras, if I had to guess I'd say that's the most likely constraint since we know TSMC is packaging limited in other domains (it limited Nvidia's sales and no doubt Nvidia would have paid whatever TSMC asked to increase that capacity but it took quite some time to do so) and it is perhaps also affected by Apple's rumored use of M2 Ultras (and perhaps newer Ultras) for their AI servers. Apple might decide that's a more important corporate priority than updating Mac Pro - maybe M4 Ultras become available and they started using those for AI servers and they had some M3 Ultras left over and released the Studio to consume them. You never know what's going on behind the scenes, at least not for sure there are only rumors of varying veracity (heck my speculation in this paragraph might appear elsewhere as a "rumor" later this week lol)

Eug · Monday at 5:15 PM

Thermal throttling of M4 in fanless MacBook Air, vs. fan-endowed MacBook Pro.

MB Pro is 35% faster.

The MB Air does better if you put it on a stand with a built-in fan.

It does even better if you add a thermal pad inside.

Not sure why he has three different scores for the MacBook Pro.

I'm going to get one of these for my wife. She is only using about 150 GB on her 2017 Broadwell dual-core i5 MacBook Air, so a 256 GB model would work fine for the next couple of years, but I may consider a 512 GB model since the hope is to use it until the battery dies in 8 years or whatever.

mvprod123 · Monday at 6:08 PM

okoroezenwa said:
It indeed would be terrible. To reconcile both the ‘M5 Mac Pro will exist this year’ and ‘M5 Mac Pro will be released at least next year’ I propose: M5 Mac Pro will be announced at WWDC (or even the October laptop refresh mini-event) and set to release early next year.

Doug S said:
The Ultra has always followed the Max launch by months, it would be a big change to launch them alongside one another. If they ship a new Ultra this year I would bet on it being M4 Ultra not M5.

Maybe they can address whatever the reasons (yield/testing, packaging, verification?) that cause such significant delays with the Ultra launches, who knows. But the recent launch of an M3 Ultra months after M4 Max appeared doesn't seem to indicate they have yet done so. Maybe if/when they go to more of a chiplet based design an Ultra type part will be just as easy to make as a Max type part if both are built similarly from chiplets with the only real difference that the Ultra part contains more chiplets.

I agree it is bad that the Mac Pro lags. Granted the sales volumes probably aren't that large but that's their halo product and it is a bad look having it out of date compared to cheaper offerings.

I don't know what sort of volume TSMC has available for the special packaging used for Ultras, if I had to guess I'd say that's the most likely constraint since we know TSMC is packaging limited in other domains (it limited Nvidia's sales and no doubt Nvidia would have paid whatever TSMC asked to increase that capacity but it took quite some time to do so) and it is perhaps also affected by Apple's rumored use of M2 Ultras (and perhaps newer Ultras) for their AI servers. Apple might decide that's a more important corporate priority than updating Mac Pro - maybe M4 Ultras become available and they started using those for AI servers and they had some M3 Ultras left over and released the Studio to consume them. You never know what's going on behind the scenes, at least not for sure there are only rumors of varying veracity (heck my speculation in this paragraph might appear elsewhere as a "rumor" later this week lol)

Perhaps Apple will present the entire line of M5 chips, including the new MacBook Pro and the updated Mac Pro, in late fall. At the same time, Mac Studio will wait for the next update in the M6 generation.

Mopetar · Monday at 6:17 PM

Perhaps M3 Ultra had some issues that took longer than anticipated to iron out and they decided to stick with it rather than canceling it.

Keep in mind that M4 launched about a little under 7 months after M3, which is practically unheard of in terms of chip cadence. If M3 Ultra was delayed at the same time that M4 turned out to be ready ahead of schedule it's not surprising that it looks messy.

In a more typical world we would have expected M4 to launch late last year or early this year. The M3 Ultra launching alongside new M4 chips wouldn't seem abnormal at all. M5 launching sometime later this year puts us back around 15 months between launches which fits within the typical 12 - 18 months between launches that are more typical.

Thibsie · Monday at 6:25 PM

Mmmm M3 is N3B and M4 is N3E ?
If so, explanation is obvious IMO.

poke01 · Monday at 6:28 PM

WIPO - Search International and National Patent Collections

GPU matrix multiplication units?

johnsonwax · Tuesday at 3:45 AM

Apple generally releases their highest revenue products first. The Mac Pro is always going to be the end of the line. It's such an awkward product for Apple and has really low unit sales. My guess is WWDC next year earliest, and only if they can announce some additional benefit to it - discrete GPU or RAM, etc. Otherwise it'll be a relatively quiet release.

johnsonwax · Tuesday at 4:13 AM

poke01 said:
WIPO - Search International and National Patent Collections

GPU matrix multiplication units?

No, I think it's a compiler/scheduler optimization by identifying dependent mmults and locally caching the result of the first to speed up the second as lots of compute does this kind of chaining of mmults (lots of compiler optimization around it as well). Looks like the goal is to reduce the hit to memory - remove some of that cost from situations when you're really cranking on chained mmults. Per the patent, looks like the compute can be in any unit (GPU, NPU, SME, etc.)

I do love how stupidly pedantic patents need to be that they have to specify you might want to do this in a computer that has a battery.

mvprod123 · Tuesday at 6:45 AM

poke01 said:
WIPO - Search International and National Patent Collections

GPU matrix multiplication units?

This is what Apple's GPU lacks.
David Huang wrote about this in his blog.

Apple统一内存适合运行LLM？理想很丰满，现实很骨感 | David Huang's Blog

Doug S · Tuesday at 3:20 PM

mvprod123 said:
This is what Apple's GPU lacks.
David Huang wrote about this in his blog.

Apple统一内存适合运行LLM？理想很丰满，现实很骨感 | David Huang's Blog

This suggests what their direction is for their server chips. I posted recently about how using Ultras for that role is a stopgap solution only and Apple will have designed some sort of custom solution (since using Ultras as a stand in is a compromise that wastes a lot of silicon area on irrelevant functionality) I wondered whether Apple would build that solution of their GPU, their NPU, or a custom solution based that exists only in their servers.

This patent suggests that the answer may be that they're going to build it out of their GPU. If there was no market for Macs able to run LLMs it would seem to make more sense to go custom but so long as they see that as a viable (if niche) market it makes sense to beef up their iGPU. And there may be enough of a market, especially if as David indicates Apple could be quite competitive with Nvidia and AMD dGPU based alternatives if they fill in any functionality gaps with stuff like matmal in their iGPU. That would be an interesting way to get their foot in the door in markets that have never seen Apple as a potential option.

Doug S · 2025-04-02T15:54:50-0400

So its been a month now since the 16e with the C1 modem was released, and it seems to have been a slam dunk for Apple. Other than people nitpicking about the lack of mmwave, I haven't heard any real criticism - nothing like the articles/reviews detailing the shortcomings of Intel modems vs Qualcomm modems ~7 years ago. I honestly expected to see something similar, and/or some issues on certain carriers where people see a lot of dropped calls or connection failures or the like. Maybe its out there but if it is it isn't widespread enough to have reached the tech press.

You'd have to think Qualcomm would have bought a few and tried to find an Achilles heel where it looks really bad but if they have they must not have found anything or they would have leaked it to a friendly journalist. Looks like Apple is on a glide path to replace Qualcomm in every iPhone next year without any complaints.

Something interesting to note - I'm seeing more and more "5Guw" (using Verizon's Visible+) around town - though still the same 2 bars of LTE at my actual house lol - but every place I've checked it is band 77. Yep, Verizon is apparently claiming "uw" for 3.7 GHz - which a C1 modem works just fine for. So I'm not sure how you can even tell if you really are on mmwave unless you check the band you're connected to, which in the US I believe are bands in the 250-260 range.

Mopetar · 2025-04-02T19:25:48-0400

Doug S said:
This patent suggests that the answer may be that they're going to build it out of their GPU. . . . Apple could be quite competitive with Nvidia and AMD dGPU based alternatives if they fill in any functionality gaps with stuff like matmal in their iGPU.

I'm sure everyone in the GPU forums would love Apple if they managed to get everyone to switch over to scalping Mac minis and leave their discrete cards alone.

I'd also find it kind of funny to see people buying up thousands of Macs to stick into racks to train LLMs on as opposed to paying $$$$$ for Nvidia accelerator cards.

mvprod123 · 2025-04-02T21:11:25-0400

Microsoft Surface looks terrible and shameful against the M4 MacBook Air.

SpudLobby · 2025-04-02T21:50:58-0400

Doug S said:
So its been a month now since the 16e with the C1 modem was released, and it seems to have been a slam dunk for Apple. Other than people nitpicking about the lack of mmwave, I haven't heard any real criticism - nothing like the articles/reviews detailing the shortcomings of Intel modems vs Qualcomm modems ~7 years ago. I honestly expected to see something similar, and/or some issues on certain carriers where people see a lot of dropped calls or connection failures or the like. Maybe its out there but if it is it isn't widespread enough to have reached the tech press.

You'd have to think Qualcomm would have bought a few and tried to find an Achilles heel where it looks really bad but if they have they must not have found anything or they would have leaked it to a friendly journalist. Looks like Apple is on a glide path to replace Qualcomm in every iPhone next year without any complaints.

Something interesting to note - I'm seeing more and more "5Guw" (using Verizon's Visible+) around town - though still the same 2 bars of LTE at my actual house lol - but every place I've checked it is band 77. Yep, Verizon is apparently claiming "uw" for 3.7 GHz - which a C1 modem works just fine for. So I'm not sure how you can even tell if you really are on mmwave unless you check the band you're connected to, which in the US I believe are bands in the 250-260 range.

T-Mobile does this as well with “UC”. It’s not mmWave, but man is it fast.

mmWave is pretty much a waste in most cases imho; C1 isn’t missing much without it.

SpudLobby · 2025-04-02T21:52:55-0400

Doug S said:
So its been a month now since the 16e with the C1 modem was released, and it seems to have been a slam dunk for Apple. Other than people nitpicking about the lack of mmwave, I haven't heard any real criticism - nothing like the articles/reviews detailing the shortcomings of Intel modems vs Qualcomm modems ~7 years ago. I honestly expected to see something similar, and/or some issues on certain carriers where people see a lot of dropped calls or connection failures or the like. Maybe its out there but if it is it isn't widespread enough to have reached the tech press.

You'd have to think Qualcomm would have bought a few and tried to find an Achilles heel where it looks really bad but if they have they must not have found anything or they would have leaked it to a friendly journalist. Looks like Apple is on a glide path to replace Qualcomm in every iPhone next year without any complaints.

Something interesting to note - I'm seeing more and more "5Guw" (using Verizon's Visible+) around town - though still the same 2 bars of LTE at my actual house lol - but every place I've checked it is band 77. Yep, Verizon is apparently claiming "uw" for 3.7 GHz - which a C1 modem works just fine for. So I'm not sure how you can even tell if you really are on mmwave unless you check the band you're connected to, which in the US I believe are bands in the 250-260 range.

Agree also RE: Intel vs Qualcomm modems.

And even more notable about that is that Apple actually *used* Intel modem and while they sucked it still wasn’t a dealbreaker. Here Apple should have even less hesitancy if their modems are indistinguishable or 80-90% as good, at lower cost.

LightningZ71 · 2025-04-02T23:05:44-0400

The problem that I see in most mmWave implementations is that while the radio can certainly meet the spec and transfer data at crazy speeds, it hits a wall on back haul out of the micro cell. Yes, I can have a bunch of devices hit the same cell and establish crazy connections, but when you actually stress the throughput, it gets you nowhere faster than other well provisioned 5G cells can. It does handle saturation on the radio side better, and lots of small transactions from lots of devices works swimmingly.

johnsonwax · 2025-04-03T02:45:34-0400

LightningZ71 said:
The problem that I see in most mmWave implementations is that while the radio can certainly meet the spec and transfer data at crazy speeds, it hits a wall on back haul out of the micro cell. Yes, I can have a bunch of devices hit the same cell and establish crazy connections, but when you actually stress the throughput, it gets you nowhere faster than other well provisioned 5G cells can. It does handle saturation on the radio side better, and lots of small transactions from lots of devices works swimmingly.

mmWave was designed for enterprise installations where they could throw down a ton of backhaul and deal with the range/interference issues. Not really useful for consumers.

Discussion Apple Silicon SoC thread

Lifer

Lifer

Member

Diamond Member

Lifer

Senior member

Diamond Member

Golden Member

Diamond Member

Member

Member

Senior member

Diamond Member

Diamond Member

Diamond Member

Senior member

Golden Member

Golden Member

Platinum Member

Member