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

 

[name99]

Maynard, you have no clue what you are talking about. Go read about SVE and SME. ZA is a set of tiles (matrices), while Z are wider registers (vectors). Learn before making silly guesses, dude.

BTW I did discuss with the guys who made the program and analysis of the Apple SME block. I asked them to redo their runs zeroing Z registers (which is not really mandated after SMSTART but that was worth trying as it's trivial); same result: SSVE is 1/64th the speed of SME and much slower than plain NEON. I'm still not 100% sure it is the definitive answer, that's why I wrote "it seems". I'm not enough full of myself, as you sometimes are, to draw conclusions.

I also read about SVE and SME, going as far as implementing a simulator. So feel free to lecture me, it's funny.

We have a second confirmation of the 250GB/s SSVE results here:

M4+ Chip Generation - Speculation Megathread [MERGED]

That is true. I still think it might work well enough even with the current register cache. Apple can do 32-bit FP+Int most of the time without problems after all. And they are the only vendor who offers four-operand instructions. I mean, Nvidia does it and it works for them, even though they do...

forums.macrumors.com

We have my thoughts on the issue here:

M4+ Chip Generation - Speculation Megathread [MERGED]

Did Corsix give performance numbers for the two and four vector variants? Presumably you have to use those to get max throughput. It wasn't clear to me from the link I gave if Eric had *measured* that he got 2000 GFLOPs or not. It's certainly possible that, even though the double and quad...

forums.macrumors.com

As I said in that post it remains unclear to me the exact limits on register input into the current (M4) FMAC block.

I'll be the first to admit that I don't yet know the SSVE and SME ISA well.
However I do have a pretty good idea of how the AMX block is wired up, how it has evolved, and what it *should* and should not be capable of. Z registers obviously map onto AMX's X and Y registers. The ZA register obviously maps onto AMX's Z register. Everything follows from that.
Further discussion of limits that may or may not be relevant in the M4 implementation here:

Real World Technologies - Forums - Thread: Apple and AI

www.realworldtech.com

 

[Mopetar]

This wouldn’t cause that much of outrage. Apple is advertising 8GB.

People spent thousands of words neighing and braying last week about Apple's iPad commercial. Do you think they'll pass by a golden opportunity to complain about something that's actually legitimate?

It makes no sense for Apple to pay for 6 GB chips that they artificially hobble when they could just have bought 4 GB chips.

 

[Glo.]

This wouldn’t cause that much of outrage. Apple is advertising 8GB.

My guess would be that if Apple locked the 4GB then it would due to the M3 MacBook Pro having 8GB.

Kinda embarrassing if a $999 tablet has more ram than a $1599 MacBook.

I wonder what will happen when people will hear and find out that their iPads have 12 GB of RAM, but Apple is only giving them 8 GB's of it?

Yeah, nothing most likely, right?

 

[poke01]

I wonder what will happen when people will hear and find out that their iPads have 12 GB of RAM, but Apple is only giving them 8 GB's of it?

Yeah, nothing most likely, right?

I mean Apple clearly states 8GB?

What’s next? Finding out the there are actually 10 cores in M4 but Apple is only offering 9 cores in M4 in the base model.

That 4th P core is not defective, Apple purposely disabled it, for segmentation. Just they like did with the RAM.

Also am not saying it’s good but you know what is being is offered to you before you buy the iPad.

 

[poke01]

People spent thousands of words neighing and braying last week about Apple's iPad commercial. Do you think they'll pass by a golden opportunity to complain about something that's actually legitimate?

It makes no sense for Apple to pay for 6 GB chips that they artificially hobble when they could just have bought 4 GB chips.

You know what those people should really complain about: Apple’s SSD and RAM price upgrades. Where’s the media/social media outrage about that?

Where’s the outrage on Apple’s stupid anti-repair tactics?

Those are evil, to put bluntly.

It’s costs Apple like $30 for a LPDRR5 module.

$200 for an 8GB RAM upgrade makes me puke.

 

[Glo.]

I mean Apple clearly states 8GB?

What’s next? Finding out the there are actually 10 cores in M4 but Apple is only offering 9 cores in M4 in the base model.

That 4th P core is not defective, Apple purposely disabled it, for segmentation. Just they like did with the RAM.

Also am not saying it’s good but you know what is being is offered to you before you buy the iPad.

Where does it state the spec of the RAM chips?

M4 config has 10 CPU cores, 10 GPU cores and 16 NPU cores.

It also is advertised as 8 GB, but has actually 12 GB of RAM. Thats where the issue is.

The only reason, THE ONLY REASON, why would you give, physically 12 GB of RAM but limit it to 8 GB is marketing.

Absolute disgrace.

 

[poke01]

M4 config has 10 CPU cores, 10 GPU cores and 16 NPU cores.

Yes but they also offer a M4 with 9 cores when they could easily offer the full 10 cores. Apple disables the one P core in the base on purpose. So how is disabling RAM any different?

 

[poke01]

I really don’t this outrage. How is this any different from disabling cores on purpose for market segmentation?

There are worse things that Apple does we can be angry about.

 

[Eug]

Yeah, if true they are artificially limiting 6 GB parts to 4 GB RAM, that kinda sucks, but I don't think there will be any outrage about it. In fact, I don't think it will even make it to the mainstream media because the evidence for it is too obscure.

However, I'm not sure 9-core M4 is quite the same thing as the memory thing. I think it's likely M4 is truly binned in part. There's a good chance the majority are binned for marketing reasons, but that doesn't mean all of them binned for marketing reasons, unlike what this supposed situation is with the memory.

Anyhow, last year I predicted a 2024 M3 OLED iPad Pro 8 GB RAM with landscape camera and new Magic Keyboard. Unfortunately I was right about the 8 GB RAM, but fortunately I was wrong about M3, and we got M4 instead. I think the general public will focus on that instead.

If there is a silver lining here, this does bode well for base 12 GB Macs to arrive in the coming year. I had been considering picking up a used M1 8 GB MacBook Air if I could find one cheap, to replace my wife's 2017 MacBook Air. However, that i5 8/256 GB still functions OK for her, so this is a good enough reason to wait another iteration and hope for 12 GB base. If the MacBook Air goes M4 12 GB in 2025, I'd likely buy a base model. That thing would last a decade for her, whereas the M1 is already almost 4 years old now, and I really, really don't want to pay CA$250 for 16 GB RAM with a new M3.

 

[poke01]

Yeah, if true they are artificially limiting 6 GB parts to 4 GB RAM, that kinda sucks, but I don't think there will be any outrage about it. In fact, I don't think it will even make it to the mainstream media because the evidence for it is too obscure.

However, I'm not sure 9-core M4 is quite the same thing as the memory thing. I think it's likely M4 is truly binned in part. There's a good chance the majority are binned for marketing reasons, but that doesn't mean all of them binned for marketing reasons, unlike what this supposed situation is with the memory.

Anyhow, last year I predicted a 2024 M3 OLED iPad Pro 8 GB RAM with landscape camera and new Magic Keyboard. Unfortunately I was right about the 8 GB RAM, but fortunately I was wrong about M3, and we got M4 instead. I think the general public will focus on that instead.

If there is a silver lining here, this does bode well for base 12 GB Macs to arrive in the coming year. I had been considering picking up a used M1 8 GB MacBook Air if I could find one cheap, to replace my wife's 2017 MacBook Air. However, that i5 8/256 GB still functions OK for her, so this is a good enough reason to wait another iteration and hope for 12 GB base. If the MacBook Air goes M4 12 GB in 2025, I'd likely buy a base model. That thing would last a decade for her, whereas the M1 is already almost 4 years old now, and I really, really don't want to pay CA$250 for 16 GB RAM.

It’s $300AUD for an 8 GB upgrade here. You can literally get 64GB of DDR5 RAM for that price.

If Apple doesn’t change their upgrade pricing then I will move to either an x86 or X Elite machine. I need more than 16GB of RAM and Apple’s pricing is pure theft.

It’s not 2020 anymore, Lunar Lake and X Elite will likely also have great battery life. They start at 16GB of RAM due to MS requirement.

I don’t know how Apple will promote AI Mac’s with 8GB of RAM when the competition is offering 16GB.

I will be keeping an eye on the MS event tomorrow.

 

[Ajay]

Where does it state the spec of the RAM chips?

M4 config has 10 CPU cores, 10 GPU cores and 16 NPU cores.

It also is advertised as 8 GB, but has actually 12 GB of RAM. Thats where the issue is.

The only reason, THE ONLY REASON, why would you give, physically 12 GB of RAM but limit it to 8 GB is marketing.

Absolute disgrace.

You do know that consumer products companies have segmented their products by 'artificial' means for decades - right?

 

[kingsleyopara]

My guess is there is or will be a parts lottery between 4gb and 6gb modules. Perhaps micron simply couldn't supply enough 4gb modules quickly enough to meet their agreement with Apple?

 

[dr1337]

How is this any different from disabling cores on purpose for market segmentation?

Aren't cores always disabled because of power/thermal constraints? Or do base model ipads/macs get way better efficiency/watt than fully specc'd ones?

And wouldn't 12gb instead of 8gb help sell more base models anyways? Apple already paid for that 4 extra gb so why not use it to make the product look better?

 

[roger_k]

Maynard, you have no clue what you are talking about. Go read about SVE and SME. ZA is a set of tiles (matrices), while Z are wider registers (vectors). Learn before making silly guesses, dude.

It's not a silly guess and Maynard is right. SME does have vector instructions that put the results in the ZA storage and usage them give you roughly 8x improvement over base SSVE versions. It's the nature of the AMX unit — the ALU is associated with the matrix storage, not vector storage. It is likely that the instructions that target Z registers as destination incur double data movement cost.

Using FMLA that writes to ZA tile I get 250 GFLOPS.

BTW I did discuss with the guys who made the program and analysis of the Apple SME block. I asked them to redo their runs zeroing Z registers (which is not really mandated after SMSTART but that was worth trying as it's trivial); same result: SSVE is 1/64th the speed of SME and much slower than plain NEON. I'm still not 100% sure it is the definitive answer, that's why I wrote "it seems". I'm not enough full of myself, as you sometimes are, to draw conclusions.

I also read about SVE and SME, going as far as implementing a simulator. So feel free to lecture me, it's funny.

The vector ZA instructions have been added in SME2, so that's probably why you might have been not aware of them.

 

[poke01]

Aren't cores always disabled because of power/thermal constraints? Or do base model ipads/macs get way better efficiency/watt than fully specc'd ones?

Apple has been segmenting cores since A12X. They have been disabling them on purpose to either use them in the future or as a upsell tactic.

For example the A15 had a 4 core GPU in the normal iPhone 13 and a 5 core GPU in the 13 Pros. It wasn't due to low yields, this was when Apple started to segement cores on iPhones. Later this year I would expect the same on the iPhone 16 models.

 

[darkswordsman17]

You know what those people should really complain about: Apple’s SSD and RAM price upgrades. Where’s the media/social media outrage about that?

Where’s the outrage on Apple’s stupid anti-repair tactics?

Those are evil, to put bluntly.

It’s costs Apple like $30 for a LPDRR5 module.

$200 for an 8GB RAM upgrade makes me puke.

Er, where have you been? People have been complaining about those for years.

I think it'd probably fall flat as most other OEMs are as bad if not worse these days. Or they have so much quality issues that its almost a moot point (see ASUS "we totally didn't move the microSD on the ROG Ally cause of overheating!")

I'm not sure I'd say that's evil, but its certainly terrible.

People spent thousands of words neighing and braying last week about Apple's iPad commercial. Do you think they'll pass by a golden opportunity to complain about something that's actually legitimate?

It makes no sense for Apple to pay for 6 GB chips that they artificially hobble when they could just have bought 4 GB chips.

I don't agree, especially for a brand that literally built its reputation by courting artists I think its a pretty big deal when they're looking to replace them with AI (especially the version of AI we have now that is built off of those creatives' work, without compensating them).

Guess we'll see at WWDC. I have a hunch its to support AI. Honestly, I think the outrage will actually come once the people that ponied up the extra for the 16GB end up with "losing" RAM from AI features, while the lower models get AI features that don't take away RAM because they only supposedly had 8GB.

 

[poke01]

I don't agree, especially for a brand that literally built its reputation by courting artists I think its a pretty big deal when they're looking to replace them with AI (especially the version of AI we have now that is built off of those creatives, without compensating them).

Yeah, Apple quickly reliased that ad was not good, after all artists made the brand.

Guess we'll see at WWDC. I have a hunch its to support AI. Honestly, I think the outrage will actually come once the people that ponied up the extra for the 16GB end up with "losing" RAM from AI features, while the lower models get AI features that don't take away RAM because they only supposedly had 8GB.

It would be nice to see where this leads.

 

[jpiniero]

For example the A15 had a 4 core GPU in the normal iPhone 13 and a 5 core GPU in the 13 Pros. It wasn't due to low yields, this was when Apple started to segement cores on iPhones.

No node has 100% yield though. Especially given how expensive N3E wafers are, you have to do what you can to maximize the output.

 

[Glo.]

You do know that consumer products companies have segmented their products by 'artificial' means for decades - right?

I would not have any problem with 8 GB on iPad Pro IF:

1) there would be a config that would allow you to get those 12 GB of RAM, because that is the full spec of the base version of the chip.
2) The only other spec is 16 GB, with different memory chips, which is different spec of M4 chip.

So Apple is artificially limiting full spec of the chip, without you knowing it. Yes, they sell what they advertise, and yet - they don't sell what they offer, because base, full spec for M4 is 10 CPU cores, 10 GPU cores, 16 NPU cores and 12 GB of RAM.

Getting 9 CPU cores in base M4 config - thats fine, product segmentation and you know what full spec of M4 was on the compute side.

Getting 8 GB out of 12 GB, without you knowing that there is 12, and no possibility of getting those 12 GB of RAM - thats where the issue lies.

Because the hardware is there. The ONLY REASON why you do not get it - is Apple marketing. Absolute disgrace on Apple front.

 

[Eug]

I would not have any problem with 8 GB on iPad Pro IF:

1) there would be a config that would allow you to get those 12 GB of RAM, because that is the full spec of the base version of the chip.
2) The only other spec is 16 GB, with different memory chips, which is different spec of M4 chip.

So Apple is artificially limiting full spec of the chip, without you knowing it. Yes, they sell what they advertise, and yet - they don't sell what they offer, because base, full spec for M4 is 10 CPU cores, 10 GPU cores, 16 NPU cores and 12 GB of RAM.

Getting 9 CPU cores in base M4 config - thats fine, product segmentation and you know what full spec of M4 was on the compute side.

Getting 8 GB out of 12 GB, without you knowing that there is 12, and no possibility of getting those 12 GB of RAM - thats where the issue lies.

Because the hardware is there. The ONLY REASON why you do not get it - is Apple marketing. Absolute disgrace on Apple front.

Honestly I think you're making a mountain out of a molehill with this 12 GB thing. I'm a little bit annoyed too since I'll be buying an 8 GB model, but then again, it's not really a big deal overall since Apple is only advertising 8 GB. Much more important IMO is Apple's decision to limit 16 GB to the 1-2 TB models at a US$600 price premium, regardless if the base is 8 or 12 GB, even though I can understand their reasons for that too.

The other thing we don't know yet is if ALL the base models are 12 GB. Maybe they are, but we haven't ruled out the possibility that some are 8 GB and some are 12 GB, and Apple keeps them all at 8 GB so everyone effectively gets the same product.

However, I do like having the knowledge though that 6 GB DRAM chips are likely now cheap enough that it makes sense for Apple to just use them instead of specifically sourcing 4 GB chips. That bodes well for 6 GB x 2 chips being used in M4 Macs next year... when it makes sense for Apple from a marketecture point of view.

BTW, how common are 4 GB LPDDR5X chips anyway, and what is the cost difference vs 6 GB chips?

 

[Eug]

BTW, how common are 4 GB LPDDR5X chips anyway, and what is the cost difference vs 6 GB chips?

Interesting. I just did a search for Micron LPDDR5X at DigiKey, and 4 GB chips are completely unavailable. These are the sizes they have:

3 GB
6 GB
12 GB
16 GB

Yes, I know 4 GB and 8 GB SKUs exist (and Apple is using 8 GB chips in the higher end model), but it may be that the pricing and availability for 4 GB at high volume didn't make sense for Micron and Apple. Or else they didn't have enough for the first batch of iPad Pros.

 

[dr1337]

Given that above info, it could be that all are 12 GB.

That would be even worse and means there is no real binning at all going on here.

A company that has been upselling itself as green and eco friendly since the iPhone 7 now literally selling people good silicon but arbitrarily disabled so they could make the gap between the base and top ipad look bigger?

I think anyone with a healthy consumer conscious sees why this practice is bad. Consumers are paying the price for a 12gb model (since that is physically whats there and ram isn't free to apple either), and only getting 8gb for marketing reasons.

They have been disabling them on purpose to either use them in the future or as a upsell tactic.

You're just assuming that though? My point was that if they do artificially segment CPUs and cut away perfectly good silicon, there would be evidence in the fact that base model chips would innately consume less power and put off less heat.

I am assuming phones have extremely tight margins for their power envelopes and that most of the binning apple does is entirely for that purpose. Maybe they have perfect yields or something, but again there should be evidence of that out in the world. And surely they'd use epic yields to raise ASPs instead cutting down a chip that's worth more in a more expensive product.

 

[Glo.]

Honestly I think you're making a mountain out of a molehill with this 12 GB thing. I'm a little bit annoyed too since I'll be buying an 8 GB model, but then again, it's not really a big deal overall since Apple is only advertising 8 GB. Much more important IMO is Apple's decision to limit 16 GB to the 1-2 TB models at a US$600 price premium, regardless if the base is 8 or 12 GB, even though I can understand their reasons for that too.

The other thing we don't know yet is if ALL the base models are 12 GB. Maybe they are, but we haven't ruled out the possibility that some are 8 GB and some are 12 GB, and Apple keeps them all at 8 GB so everyone effectively gets the same product.

However, I do like having the knowledge though that 6 GB DRAM chips are likely now cheap enough that it makes sense for Apple to just use them instead of specifically sourcing 4 GB chips. That bodes well for 6 GB x 2 chips being used in M4 Macs next year... when it makes sense for Apple from a marketecture point of view.

BTW, how common are 4 GB LPDDR5X chips anyway, and what is the cost difference vs 6 GB chips?

Yes, one possibility is that there is a mixture of 4 and 6 GB chips. And that is the only possible redeeming solution. If all are 12 GB - its absolute disgrace.

Funnily enough - I don't have a problem with 16 GB of RAM being only on highest SSD capacity models. Thats still marketing, but... reasonable. If it is caused by the 8 GB memory chips. However...

Interesting. I just did a search for Micron LPDDR5X at DigiKey, and 4 GB chips are completely unavailable. These are the sizes they have:

3 GB
6 GB
12 GB
16 GB

Yes, I know 4 GB and 8 GB SKUs exist (and Apple is using 8 GB chips in the higher end model), but it may be that the pricing and availability for 4 GB at high volume didn't make sense for Micron and Apple. Or else they didn't have enough for the first batch of iPad Pros.

There is one, another possibility.

Apple is actually, in 16 GB models, using 12 GB chips, and limiting them to only 16 GB.

So there is physically 12, and 24 GB but you get only 8 and 16.

 

[Eug]

There is one, another possibility.

Apple is actually, in 16 GB models, using 12 GB chips, and limiting them to only 16 GB.

So there is physically 12, and 24 GB but you get only 8 and 16.

No, it's only the 8 GB iPad Pro model's chip SKU that is odd. According to the teardowns:

8 GB iPad Pro = 2 x 6 GB
16 GB iPad Pro = 2 x 8 GB

 

[Mopetar]

If they're paying for 6 GB modules and not using any 4 GB modules, just give people 12 GB of RAM. No one who wants the top model is going to decide not to purchase it because the base model has 12 GB of RAM.