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

 

[Ajay]

actually, the next step after 48 GB on 384 bit bus is 72 GB .

So 36, 48, 72 GB of RAM on a Mac Mini with M3 Pro.

Truman has been wrong before and Apple occasionally tweaks stuff to catch leakers - so who knows for sure. I’m out on this one till there’s more info.

 

[Glo.]

Truman has been wrong before and Apple occasionally tweaks stuff to catch leakers - so who knows for sure. I’m out on this one till there’s more info.

I love how people have cognitive dissonance just because there is such huge jump in memory capacity, when the explanation for this jump may be extremely simple .

 

[Doug S]

Still, if the 384b bus theory is correct, why not just increase RAM by 50%? Not like it’s expensive right now. Why use some bizzare fractional increase. Apple could always clock down the LPDDR5 on laptops to manage power.

If Apple went to a 384 bit bus for M3 Pro it wouldn't be for capacity it would be for bandwidth. If you have more/faster CPU and GPU cores you need more memory bandwidth. Sure you get a boost assuming they are going to LPDDR5X but that's a one time 30% gain over the M2's LPDDR5-6400. JEDEC only goes to LPDDR5X-8533 and LPDDR6 won't be arriving anytime soon so if they are making decisions that impact not only M3 but also M4 and perhaps M5 before we see LPDDR6...

If they just wanted to support more capacity there is plenty of room for more RAM per package, though since the packages are custom having low volume options for high capacity would increase cost and complicate ordering and inventory management.

 

[gdansk]

Maybe they'll finally be willing to upgrade from 8GB base configuration if they can use a half step of 12GB instead of 16GB.

 

[Glo.]

Maybe they'll finally be willing to upgrade from 8GB base configuration if they can use a half step of 12GB instead of 16GB.

If 36 GB on 384 bit bus is correct it may mean that is the only option for 2024.

And if so, and M3 is half of M3 Pro - we are looking at 192 bit bus and 18 GB of RAM.

P.S. Is it only me, or 36GB looks much less weird than 18 GB of RAM...

Is it only mine OCD for such numbers...?

 

[gdansk]

If 36 GB on 384 bit bus is correct it may mean that is the only option for 2024.

And if so, and M3 is half of M3 Pro - we are looking at 192 bit bus and 18 GB of RAM.

P.S. Is it only me, or 36GB looks much less weird than 18 GB of RAM...

Is it only mine OCD for such numbers...?

Yeah I have no idea. But hopefully 8gb is gone by 2024.
And I don't like it either.

 

[Doug S]

If 36 GB on 384 bit bus is correct it may mean that is the only option for 2024.

And if so, and M3 is half of M3 Pro - we are looking at 192 bit bus and 18 GB of RAM.

P.S. Is it only me, or 36GB looks much less weird than 18 GB of RAM...

Is it only mine OCD for such numbers...?

I would bet the non M3 remains 128 bit if M3 Pro goes to 384. That goes into stuff like iPad Pro & Macbook Air that are passively cooled.

 

[Glo.]

I would bet the non M3 remains 128 bit if M3 Pro goes to 384. That goes into stuff like iPad Pro & Macbook Air that are passively cooled.

IMO, M3 is either 192 or... 256 bit.

You need all of that bandwidth for higher CPU thoughput, and especially for RT capabilities in the GPUs. Macroscalar architectures have one problem. Heavy demands for memory bandwidth.

 

[Doug S]

IMO, M3 is either 192 or... 256 bit.

You need all of that bandwidth for higher CPU thoughput, and especially for RT capabilities in the GPUs. Macroscalar architectures have one problem. Heavy demands for memory bandwidth.

You can't support enough cores to need 192 let alone 256 bit wide LPDDR5X-8533 in a passively cooled device. The base M3 is not going to compromise on its power envelope, so it isn't going to get a bunch of additional cores hence no demand for a 50% or 100% increase in memory bandwidth in addition to the 33% LPDDR5X already offers.

 

[moinmoin]

Bandwidth is for the GPU obviously.

 

[Glo.]

You can't support enough cores to need 192 let alone 256 bit wide LPDDR5X-8533 in a passively cooled device. The base M3 is not going to compromise on its power envelope, so it isn't going to get a bunch of additional cores hence no demand for a 50% or 100% increase in memory bandwidth in addition to the 33% LPDDR5X already offers.

Its not the question of how many cores the SOC has, but how much memory bandwidth starved they are.

Putting new architectures, that M3 will involve, for both CPU and GPU may require you to bring much more memory bandwidth. And the only way possible without increasing costs, is by increasing the bus width. IMO, the LPDDR5 still will be 6400 MHz. But we get everywhere 50% wider bus.



P.S. M3 Max chip has 20 CPU cores: 12P/8E and 48 GPU cores, and M3 Ultra: 40 CPU cores, 24P/18E and 96 GPU cores

 

[igor_kavinski]

P.S. M3 Max chip has 20 CPU cores: 12P/8E and 48 GPU cores, and M3 Ultra: 40 CPU cores, 24P/18E and 96 GPU cores

Apple's E-cores must be missing something important, for them to rely more on their P-cores, unlike Intel who seems to want to pack more E-cores into their designs.

 

[LightningZ71]

Apple chose to make their "P" cores more balanced with respect to power/performance and their "E" cores as power efficient as possible in an effort to maximize power efficiency in low power devices. It's largely a legacy of the mobile origins of the whole platform.

As for the suspected increase in memory bus width, perhaps it isn't meant to feed the CPUs alone, nor is it meant to feed the GPU cores alone, but is instead designed to address the fact that all of the above, and including any focused functional units such as the neural block, the decoding block, and any other blocks, are ALL going to have greater memory demands and are expected to be operating in parallel. If all of those units are improved, that could represent a significant increase in aggregate memory bandwidth demand.

 

[Ajay]

If Apple went to a 384 bit bus for M3 Pro it wouldn't be for capacity it would be for bandwidth. If you have more/faster CPU and GPU cores you need more memory bandwidth. Sure you get a boost assuming they are going to LPDDR5X but that's a one time 30% gain over the M2's LPDDR5-6400. JEDEC only goes to LPDDR5X-8533 and LPDDR6 won't be arriving anytime soon so if they are making decisions that impact not only M3 but also M4 and perhaps M5 before we see LPDDR6...

If they just wanted to support more capacity there is plenty of room for more RAM per package, though since the packages are custom having low volume options for high capacity would increase cost and complicate ordering and inventory management.

I wasn’t saying that Apple was increasing the bus width to increase memory capacity. I’m dumb, not stupid. That said, I shouldn’t be surprised that Apple would do what they can to save a few bucks on memory.

 

[Glo.]

What if that 36 GB of RAM is actually 32 GB of RAM on 256 bit LPDDR5 and 4 GB of HBM2/3?

Apple Patent May Hint at the Future of Its Chips: A Multi-Level Hybrid Memory Subsystem

Apple patents hybrid memory system comprising of cache DRAM and main DRAM

www.tomshardware.com

New Memory Patent Could Give Future Apple M1 Chip More Power | Digital Trends

Future versions of Apple's MacBook notebooks could benefit from a new memory patent that will deliver more performance and even longer battery life.

www.digitaltrends.com

 

[Doug S]

That would complicate the operating system's job trying to keep stuff that needs high bandwidth in HBM and the rest in LPDDR5. Apple derives a lot of benefit from its unified memory architecture, which doesn't have to worry about moving data into VRAM for the GPU to operate on it etc.

Obviously Apple has put some work into this if they bothered to patent it, but not everything they patent shows up in a product. Still, it is an interesting possibility, and if they used multiple 4 GB HBM stacks in higher end products (2 for Max, 4 for Ultra, 8 for "Extreme") they'd have some truly insano GPU bandwidth available (up to 6.5 TB/s if they used HBM3 at 819 GB/s)

Maybe if someone pointed guys like Gurman and Kuo at my suggestion of a third memory controller on the Pro die to account for the 36 GB and this suggestion of HBM they might be able to tease something out of their sources to confirm or deny it.

 

[Glo.]

Thats why I don't believe that is the case, and we are looking at 384 bit LPDDR5. Plain and simple.

 

[A///]

the chatter in here is a lot more exciting than the intel or amd threads. if even a fourth of what's been discussed turns out true for the m3 pro that will be my next daily computer outside desktop for several years. I expect the ensuing 4 ed. of the M to not be impressive as the third if going by the current scheme timeline so far. I'm excited but my checkbook is not!

 

[Ajay]

the chatter in here is a lot more exciting than the intel or amd threads. if even a fourth of what's been discussed turns out true for the m3 pro that will be my next daily computer outside desktop for several years. I expect the ensuing 4 ed. of the M to not be impressive as the third if going by the current scheme timeline so far. I'm excited but my checkbook is not!

Yep. About twice a year I threaten to buy a Mac. Either some Windows update weirdness or some hardware/driver problems drives me crazy. Then I look up a decent Mac mini with everything I want (since they are not upgradeable) and it’s a wallet busting $2600. Huh - only $200 more for a Mac Studio with the better Max SoC 👍 I’d miss PC gaming though.

 

[A///]

Yep. About twice a year I threaten to buy a Mac. Either some Windows update weirdness or some hardware/driver problems drives me crazy. Then I look up a decent Mac mini with everything I want (since they are not upgradeable) and it’s a wallet busting $2600. Huh - only $200 more for a Mac Studio with the better Max SoC 👍 I’d miss PC gaming though.

I keep windows and linux desktops for working on but a general use computer and putting around a mac wins. they have their flaws but I've been a buyer of the brand for a very long time. the mac studio is neat but way overboard for my needs. a mbp is fine and will last me 7-8 years and still be useable by the time I need a new one with a battery change in the interim time period. I justify the price with my sanity being kept at bay compared to buying a windows laptop which I do own but have quirks depending on the oem.

 

[Glo.]

Guys, I would like to point out one thing.

That rumored spec is still possible even for plain M3 .

192 bit bus and 12 GB RAM chips. M2 Mac Mini, and M2 MacBook Air already used those 12 GB RAM chips in CTO configuration of 24 GB RAM.

So if M3 has 192 bit bus - we get 36 GB in top end spec .

And it actually makes sense that its M3 that appears in logs, since this is actively in development and iMac, Macbook Pro and MBA with M3 are actively in development, in final stages of it.

Im not going to say that what Gurman leaked is M3. But based on when it appeared in logs, and what could be visible in them, based on development cycle - I am leaning towards that this 6P/6E, 36 GB RAM, and 18 GPU core chip is actually - plain M3, but in highest possible spec.

 

[A///]

your gonna make people wet their trousers out excitement making promises like that.

 

[Doug S]

Im not going to say that what Gurman leaked is M3. But based on when it appeared in logs, and what could be visible in them, based on development cycle - I am leaning towards that this 6P/6E, 36 GB RAM, and 18 GPU core chip is actually - plain M3, but in highest possible spec.

TSMC's N3 process would have to be a lot more impressive than advertised to make something like that possible in the devices M3 is intended for - i.e. iPad Pro and Macbook Air. Sure, you can say "what about the Mini it is also used there" but that's a fraction of the M1 & M2's installed base and it already has the ability to use the 'Pro' variant for a higher end offering.

There simply is no market segment for something as aggressive as you describe in the passively cooled devices M3 is built for - you're talking about increasing the CPU core count by 50%, the GPU core count by 90%, and the memory bandwidth by 200% (or more if LPDDR5X is used)

Sure, it is nice to dream but TSMC's specs for N3 just don't appear to allow for that without dramatically increasing the power envelope - which means passive cooling is no longer possible and the long battery life everyone loves is severely compromised. That's what people are buying those devices for, not for what you can do with 50% more cores, 90% more GPU etc. - if you want that you buy a Macbook Pro using the 'Pro' variant of Apple Silicon.

 

[Geven]

You've made some interesting points about the potential challenges and strategies for increasing the processing power in Apple's M series chips, @Doug S. It's indeed a delicate balance between power consumption, processing power, and thermal management. The shift to N3 might reduce power use per core, as you mentioned, but you're right that any gains could be offset by more complex or higher-clocked cores. The introduction of more proprietary accelerators could be a smart move to enhance performance without overly increasing power draw. The cooling design in the Mac Studio does suggest Apple might be planning for future versions with higher power draw or more cores. It's also fascinating to think about the possibility of cores that are only activated in specific circumstances, or in devices without stringent cooling constraints. All in all, the evolution of Apple Silicon is definitely something to keep an eye on!

 

[Glo.]

your gonna make people wet their trousers out excitement making promises like that.

Its a possibility. Not a promise.