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

 

[Glo.]

6 GPU cores for iPhone chip means that maximum GPU configuration for next gen Graphics from Apple is 12(MX), 24(MX Pro) and 48(MX Max), 96(MX Ultra) and 192 GPU(Extreme) cores.

 

[Doug S]

6 GPU cores for iPhone chip means that maximum GPU configuration for next gen Graphics from Apple is 12(MX), 24(MX Pro) and 48(MX Max), 96(MX Ultra) and 192 GPU(Extreme) cores.

Based on what?

The iPhone 15 Pro line already has a 6 core GPU, but M3 and M4 both top out at a 10 core GPU.

 

[Glo.]

Based on what?

The iPhone 15 Pro line already has a 6 core GPU, but M3 and M4 both top out at a 10 core GPU.

That is MAXIMUM core configuration possible.

M2 lineup had: 5, 10, 19, 38 core GPU counts.

Apple has the flexibility to use less GPU cores, which should be obvious.

Like in an M3 Pro and M3 Max. M3 Pro has 18 cores, and M3 Max has 40 GPU cores.

The maximum config for both of them is still 24, and 48 because they scale from 6 GPU cores on iPhone chip from the same family.

Which is also connected to WHY Apple could want yearly upgrades on their hardware.

 

[poke01]

Apple Likely to Launch M4 Macs in November

Apple will likely announce new Mac models with M4-series chips in November, confidential information obtained MacRumors claims. According to a...

www.macrumors.com

November seems likely.

 

[FlameTail]

Mark Gurman:

Apple’s smaller M4 Mac mini will have 5 USB-C ports (3 on the back and 2 on the front), an internal power brick, HDMI, Ethernet, but lose USB-A. Units are about to start shipping from overseas suppliers.

 

[soresu]

Mark Gurman:

Honestly surprised that laptop manufacturers haven't made some kind of reduced footprint connector for ethernet access by this point.

RJ45 is a solid port, but oh so chonkeh in the age of super thin ultrabooks.

I seem to remember a half width variant of the connector called RJfive or some such being touted at some point, but nothing ever came of it in consumer land.



Not a perfect solution, but an improvement at least if you lay it flat.

 

[poke01]

Mark Gurman: Apple’s smaller M4 Mac mini will have 5 USB-C ports (3 on the back and 2 on the front), an internal power brick, HDMI, Ethernet, but lose USB-A. Units are about to start shipping from overseas suppliers

I might get this, great that it has front IO. Not too sad about USB-A leaving as I was prepared for this and brought a USB-C dongle Steel series mouse last year. I just want to see the size of it now.

 

[jdubs03]

Figured I’d post this here since it involves Apple. For the A20 series and M6 series.

https://twitter.com/x/status/1830440536148509137

 

[smalM]

The maximum config for both of them is still 24, and 48 because they scale from 6 GPU cores on iPhone chip from the same family.

A17 GPU: "monolithic" logic, handles all 6 cores directly
M1 GPU: "monolithic" logic, handles all 8 cores directly
M2/M3/M4 GPU: 4 subblocks, each with its own logic part (with 2 or 3 cores each)
M3 Pro/Max GPU: 2/4 subblocks, each with its own logic part, that consists of 2 subblocks, each with its own logic part (with 4 or 5 cores each)

While it seems an A GPUs could consist of 8 cores and a M GPUs of 12, we simply do not know whether the Pro/Max subblock logic can handle more cores than 5.

 

[FlameTail]

Figured I’d post this here since it involves Apple. For the A20 series and M6 series.

https://twitter.com/x/status/1830440536148509137

A16 node's timing doesn't work out for A20 SoC. A20 will use N2 ig. A21 will A16.

 

[name99]

A17 GPU: "monolithic" logic, handles all 6 cores directly
M1 GPU: "monolithic" logic, handles all 8 cores directly
M2/M3/M4 GPU: 4 subblocks, each with its own logic part (with 2 or 3 cores each)
M3 Pro/Max GPU: 2/4 subblocks, each with its own logic part, that consists of 2 subblocks, each with its own logic part (with 4 or 5 cores each)

While it seems an A GPUs could consist of 8 cores and a M GPUs of 12, we simply do not know whether the Pro/Max subblock logic can handle more cores than 5.

Quite possibly so, but this is unclear.

The most relevant patent seems to be
(2022) https://patents.google.com/patent/US20230419585A1 Tiled Processor Communication Fabric
which *seems* to suggest a GPU made from "tiles" namely blocks of GPU core with a mesh connecting them all, and presumably the tiles each have an independent link to L2.

But the patent is not at all clear as to the relevant size of the "tile". It*could* be read as even being a discussion of a full GPU (up to the size of, say, a Max) as a single tile, and the "tiling" refers to cross-SoC connections as in an Ultra or Extreme.

I agree that it seems likely that a base GPU unit is around 5..6 cores, and this is replicated to M, and then Pro and Max. But I haven't seen any DEFINITIVE proof to that effect.

A second factor is that there are multiple levels of shared logic (in the Apple, and everyone else's) GPU. For example intermediate scheduling (at say the threadblock launch level) could be handled by a block that's shared across two or three cores.

 

[johnsonwax]

Figured I’d post this here since it involves Apple. For the A20 series and M6 series.

https://twitter.com/x/status/1830440536148509137

We'll see if OpenAI still has the cash when that day arrives.

 

[johnsonwax]

Honestly surprised that laptop manufacturers haven't made some kind of reduced footprint connector for ethernet access by this point.

RJ45 is a solid port, but oh so chonkeh in the age of super thin ultrabooks.

I seem to remember a half width variant of the connector called RJfive or some such being touted at some point, but nothing ever came of it in consumer land.

Not a perfect solution, but an improvement at least if you lay it flat.

The reduced footprint connector is WiFi.

 

[igor_kavinski]

View attachment 106661

Not a perfect solution, but an improvement at least if you lay it flat.

They should try perpendicular ports so we just plug them ON the laptop around the keyboard instead of on the sides. I propose the left side of the keyboard. It would also help to remind the user's hand that they have reached the left end of the keyboard if they are a touch typist

 

[Doug S]

A16 node's timing doesn't work out for A20 SoC. A20 will use N2 ig. A21 will A16.

That's more likely than not, but far from certain.

Maybe iPhone SoCs and Apple Silicon diverge - i.e. A19 gets N3P but M5 following a couple months later gets N2, then A20 gets N2 and M6 gets A16. The lower volumes of Mac products allow that more easily than iPhone's schedule. iPhone SoCs would become the "follower" with Apple Silicon getting the new processes (and maybe cores?) first.

It was reported N2 already hit 80% yield by end of May and officially entered risk production in early July. TSMC commits to a date for N2 "mass production" in H2 2025, but if Apple feels yields are sufficient when they haven't officially hit that mass production milestone TSMC will happily sell them wafers out of risk. If they're a few ticks below 90% (TSMC's mass production threshold) when Apple starts taking wafers, that's only raising the effective cost per chip a couple bucks. And if they enter mass production officially a month or two later then they save those couple bucks on the rest.

It all depends on how aggressive Apple chooses to be. We've all seen the reports that Apple has purchased the initial allocation of N2. If A19 is N3P, what in the heck is Apple going to do with the initial allocation of N2? Even if M5 is N2, the Mac just isn't that high volume of a product (something like 5 million a quarter, and not all of them will be the latest n greatest model) meaning N2 would have to be an on awfully slow ramp for Apple to have the entire initial production.

I guess what I'm saying is, either the reports of Apple buying up the initial N2 production are wrong, or the belief that A19/M5 will not use N2 are wrong. I don't see how both can true given the known timelines with N2 officially entering risk production at the very start of Q3 - and qualifying for it in Q2!

 

[Mopetar]

We'll see if OpenAI still has the cash when that day arrives.

Never under estimate the ability for venture capitalists to keep throwing more money into the pit, even if it's on fire. Uber has had tens of billions of dollars plowed into it over the years.

 

[DZero]

After seeing how capable is Apple... maybe is time to introduce their own way of "Galaxy Dex"? the processors deserve it.

 

[Doug S]

After seeing how capable is Apple... maybe is time to introduce their own way of "Galaxy Dex"? the processors deserve it.

I was suggesting this almost a decade ago, when it became obvious it was a matter of time before Apple would switch from x86 on Macs. If they were going to do it, they would have already done it.

The fact that Dex hasn't exactly set the world on fire, or even become a selling point people discuss when deciding which Android to buy, would seem to validate their call.

 

[Eug]

Mark Gurman:

Apple’s smaller M4 Mac mini will have 5 USB-C ports (3 on the back and 2 on the front), an internal power brick, HDMI, Ethernet, but lose USB-A. Units are about to start shipping from overseas suppliers.

Apparently in his newsletter, he said the 5 USB-C ports would be on the M4 Pro model, and furthermore he had earlier said that Apple was testing new Mac minis with at least 3 USB-C ports, presumably referring to the non-Pro M4 model.

This is less than my previous hope of 4 USB-C for the M4, and 6 USB-C for the M4 Pro. Dropping USB-A was fully expected though.

Hmmm... If that is accurate, that gives me a bit of hope for the M4 Mac mini. However, I'm afraid that if Apple does adopt 4 Thunderbolt ports on the M4 Mac mini, they'll just eliminate the 2 USB-A ports.

Currently, the M2 Mac mini has 2 Thunderbolt 4 ports and 2 USB-A ports.
The M2 Pro Mac mini has 4 Thunderbolt 4 ports and keeps the 2 USB-A ports.

View attachment 101757

I’d be overjoyed if it’s actually 5 USB-C ports for the non-Pro M4 (and more USB-C ports for the M4 Pro), but that seems too optimistic.

 

[poke01]

I’d be overjoyed if it’s actually 5 USB-C ports for the non-Pro M4 (and more USB-C ports for the M4 Pro), but that seems too optimistic.

It looks like its 3 USB-C for M4 and 5 USB-C for M4 Pro. What will be more interesting is the RAM configs, it seems like base M4 will be up to 32GB now and M4 Pro Mac mini might go up to 48GB RAM.

 

[mvprod123]

Apparently in his newsletter, he said the 5 USB-C ports would be on the M4 Pro model, and furthermore he had earlier said that Apple was testing new Mac minis with at least 3 USB-C ports, presumably referring to the non-Pro M4 model.

This is less than my previous hope of 4 USB-C for the M4, and 6 USB-C for the M4 Pro. Dropping USB-A was fully expected though.

I’d be overjoyed if it’s actually 5 USB-C ports for the non-Pro M4 (and more USB-C ports for the M4 Pro), but that seems too optimistic.

Considering that the M4 has +2 additional TB controllers, 5 USB-C ports in a regular Mac mini may be true.

https://twitter.com/x/status/1804125571670036831

 

[Doug S]

It looks like its 3 USB-C for M4 and 5 USB-C for M4 Pro. What will be more interesting is the RAM configs, it seems like base M4 will be up to 32GB now and M4 Pro Mac mini might go up to 48GB RAM.

That makes a lot of sense when you consider M4 has 4 TB interfaces. You split one between HDMI, ethernet and wifi/BT, and the other three are the external TB/USB-C ports. Maybe they can't split one TB between those things (how fast does their HDMI need to be...?) so if they dedicated two to them maybe they have one full speed TB/USB-C and split one TB between two 20 Gbps USB-C ports.

You sure about base M4 at 32 GB? That seems like a giant step up from the previous 8/12 GB. I would lay heavy money on the base being 16 GB, but you can probably get 32 GB in the "step up" option where you get the M4 without disabled cores and more NAND for an extra $200 or whatever.

 

[poke01]

You sure about base M4 at 32 GB? That seems like a giant step up from the previous 8/12 GB. I would lay heavy money on the base being 16 GB, but you can probably get 32 GB in the "step up" option where you get the M4 without disabled cores and more NAND for an extra $200 or whatever.

Yeah up to 32GB on M4 with maybe a new base of 16GB.

 

[Eug]

I would have bet money on a 12 GB base, but the pundits believe it will be 16 GB base. It should be noted that currently there is a minimum requirement of 16 GB for AI based predictive code completion support in Xcode, but I had been wondering if that was because no 12 GB Apple Silicon Macs existed yet. However, perhaps it really does need 16 GB minimum.

What I really want to see is an inexpensive option for 24 GB in the Mac minis, but I suspect even if they do offer that, it will only be for the non-Pro Mac mini M4. For the Mac mini M4 Pro, they may offer 16 GB and 32 GB options only again, which if not priced reasonably may make that 32 GB option kinda pointless to most, since it's so close to the Mac Studio (unless they decide to go upmarket with Mac Studio). One can hope though that if they do restrict it to 16 --> 32 GB with no 24 GB option, they'll actually price the upgrade to 32 GB reasonably. I can't see them pricing it at $200 though. Perhaps $300?

What I'm also wondering about is what they will do with the HDMI port. Right now the M2 Mac mini maxes out at 4K 60 Hz over HDMI. To get >4K over HDMI, you need to upgrade to the M2 Pro.

 

[okoroezenwa]

It should be noted that currently there is a minimum requirement of 16 GB for on-device AI support

What are you talking about, Apple intelligence?