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

 

[Doug S]

These seem to be using the last-gen cores, which kinda lines up to the rumours the refresh was meant to come in WWCD but got delayed due to supply issues. Not that it makes much of a difference, given A15 mainly seems to focus on increasing efficiency, which doesn't matter as much on their pro laptops.
I'm still questioning why they didn't just go with 4E cores, given the space they take is negligible compared to the massive die they are making, and it helps boost MC scores.

Because Apple didn't care about "boosting MC scores" with little cores. The efficiency cores have a stated purpose, and they figured they didn't need more than two. You can always say "add one more core and increase scores" . Where do you stop, why not 16 efficiency cores, the MT numbers will be great in some benchmarks but in the real world where applications may have trouble scaling to 24 cores having so many be slower cores might not be much of an improvement at all.

They're going to be selling stuff with more than 8 big cores next year, that's what they want you to buy if you really need faster performance than what the M1 Max offers.

 

[ultimatebob]

You have to give Apple credit for bringing meaningful performance increases to the Macbook product line. Intel had basically failed to do that for the past 7 years.

 

[Eug]

Geekbench 5 scores so far:

M1 Pro: 1707 / 11030
M1 Max: 1749 / 11542
M1 Max: 1745 / 12422
M1 Max: 1738 / 12447

 

[ultimatebob]

Geekbench 5 scores so far:

M1 Pro: 1707 / 11030
M1 Max: 1749 / 11542
M1 Max: 1745 / 12422
M1 Max: 1738 / 12447

That's... not as big of an improvement as I was hoping.

I'm curious how the new M1 Max iGPU is going to do with 3DMark.

 

[biostud]

I concur. M1 Pro/Max will first be used to remove the rest of the Intel lineup. So Big iMac, and replace the 6 core Intel Mac Mini (so at least pro) .

They could theoretically put it in the small iMac, but that won't be a priority.

Then all that is left will be the Mac Pro. It's still murky what that might look like, and thus still exciting.

How important is that market segment to Apple (Mac pro) ?
Because there is a long way up to threadripper/Epyc performance and memory support/expansion.

 

[jpiniero]

How important is that market segment to Apple (Mac pro) ?
Because there is a long way up to threadripper/Epyc performance and memory support/expansion.

Apple could conceivably continue to refresh the Cheese Grater with new Xeon W processors as long as they are willing to do the development work.

 

[Eug]

How important is that market segment to Apple (Mac pro) ?
Because there is a long way up to threadripper/Epyc performance and memory support/expansion.

They currently sell a Mac Pro with 28-core Xeon W-3275M.

Geekbench 5 score 1200 / 21000.

To get to that level Apple will need to release a machine with about 16 Firestorm cores.

It should be noted that Mark Gurman states Apple is working on:

Jade 2C-Die 20-core CPU / 64-core GPU
Jade 4C-Die 40-core CPU / 128-core GPU

Judging by the names, it sounds like 2 x M1 Max and 4 x M1 Max respectively, with 16 and 32 Firestorm cores respectively.

Given that 1 x M1 Max is already 383.5 mm2, I wonder how this will be implemented on the Mac Pro.

Apple could conceivably continue to refresh the Cheese Grater with new Xeon W processors as long as they are willing to do the development work.

Maybe, but then again, I wouldn't be surprised if Apple just discontinues it at the end of 2022, and releases those monster Jade 2C-Die and Jade 4C-Die machines at that time.

If Apple discontinues Intel Mac Pros in October 2022, that means they will have full AppleCare extended warranty until October 2025, and OS support until about 2027 or something, with further security updates until 2029.

 

[blckgrffn]

They currently sell a Mac Pro with 28-core Xeon W-3275M.

Geekbench 5 score 1200 / 21000.

To get to that level Apple will need to release a machine with about 16 Firestorm cores.

It should be noted that Mark Gurman states Apple is working on:

Jade 2C-Die 20-core CPU / 64-core GPU
Jade 4C-Die 40-core CPU / 128-core GPU

Judging by the names, it sounds like 2 x M1 Max and 4 x M1 Max respectively, with 16 and 32 Firestorm cores respectively.

Given that 1 x M1 Max is already 383.5 mm2, I wonder how this will be implemented on the Mac Pro.



Maybe, but then again, I wouldn't be surprised if the just discontinue it at the end of 2022, and release those monster Jade 2C-Die and Jade 4C-Die machines at that time.

If Apple discontinues Intel Mac Pros in October 2022, that means they will have full Apple Care until October 2025, and OS support until about 2027 or something.

It's gonna be a cylinder again, so they can just stuff a whole wafer right on in there.

 

[BorisTheBlade82]

What is not true? I've only stated that Apple has created a SoC that is more efficient than the x86 counterparts currently available on the market, and it will be interesting to compare it to x86 and next gen GPUs made on same 5nm technology. I'm not saying that it is all about process technology, but that is part of it.

You and also some other members do seem to neglect the sheer margin of advantage Apple has or are (maybe accidentally) playing it down.

For comparison's sake:

• The M1 Firestorm core and the TGL 1165G Willow Cove do have near identical CB23 ST scores. But while the TGL needs 20w avg. Power on its one core the M1 only needs 3,8w. So in other terms: At ISO performance the M1 is more than 500% more energy efficient.
• Now let's take Cezanne: Here the M1 is even 11% faster than a 5900HS. And the Cezanne needs 12w avg. power. So here M1 has a better performance AND is also almost 500% more energy efficient (because it needs less time to finish the run while also consuming less on average).

So please - to all those that think that 5nm vs. 7nm was the reason for their superiority: It is not.
The process advantage is almost negligible when looking at the numbers.

/edit:
@uzzi38 I don't know where you have your numbers from. As written above the efficiency advantage is not 15 or 30% but 500%.

 

[defferoo]

@BorisTheBlade82 it's alright, there are a lot of people still in denial. still looking for reasons to be unimpressed by Apple's chips saying it's not an Apples to apples comparison because of process node. When the others get to 5nm (or Intel's 4?) we can finally do a "real" comparison right?

 

[Doug S]

They currently sell a Mac Pro with 28-core Xeon W-3275M.

Geekbench 5 score 1200 / 21000.

To get to that level Apple will need to release a machine with about 16 Firestorm cores.

It should be noted that Mark Gurman states Apple is working on:

Jade 2C-Die 20-core CPU / 64-core GPU
Jade 4C-Die 40-core CPU / 128-core GPU

Judging by the names, it sounds like 2 x M1 Max and 4 x M1 Max respectively, with 16 and 32 Firestorm cores respectively.

Given that 1 x M1 Max is already 383.5 mm2, I wonder how this will be implemented on the Mac Pro.

What do you mean "how"? The "2C" and "4C" obviously means two chips and four chips, and there would be some sort of controller chip like AMD's I/O die to connect them.

I'd guess that it would keep copies of all the cache tags for all M1 Max chips to reduce unnecessary snoop traffic, and have DDR5 controllers to allow a few DIMM slots to allow expansion beyond 256 GB for that subset of customers who need it - and perhaps keep its own cache for that DIMM expansion memory.

At 32 big cores Apple will comfortably blow away the current Mac Pro (including the potential Ice Storm upgrade) and beat the current AMD GPU options on offer. We'll have to see how it does against Nvidia's biggest and baddest workstation GPU at the time, but with access to 1.6 TB/sec of memory bandwidth, it won't be a slouch.

 

[Eug]

What do you mean "how"? The "2C" and "4C" obviously means two chips and four chips, and there would be some sort of controller chip like AMD's I/O die to connect them.

I'd guess that it would keep copies of all the cache tags for all M1 Max chips to reduce unnecessary snoop traffic, and have DDR5 controllers to allow a few DIMM slots to allow expansion beyond 256 GB for that subset of customers who need it - and perhaps keep its own cache for that DIMM expansion memory.

At 32 big cores Apple will comfortably blow away the current Mac Pro (including the potential Ice Storm upgrade) and beat the current AMD GPU options on offer. We'll have to see how it does against Nvidia's biggest and baddest workstation GPU at the time, but with access to 1.6 TB/sec of memory bandwidth, it won't be a slouch.

BTW, regarding the Pro Macs, these MacBook Pro scores are already beating the iMac Pro with 14-core Xeon W-2170B.

iMac Pro (Late 2017) - Geekbench

Benchmark results for an iMac Pro (Late 2017) with an Intel Xeon W-2170B processor.

browser.geekbench.com

 

[BorisTheBlade82]

@Doug S
This!
Plus with 32 Firestorm and 8 Ice storm cores it will leave the assembled Threadrippers in the dust. It will even trade blows with the biggest EPYC Milan 64c/128t depending on the workload (heavily, but non-optimum parallelizable).

 

[biostud]

@BorisTheBlade82 it's alright, there are a lot of people still in denial. still looking for reasons to be unimpressed by Apple's chips saying it's not an Apples to apples comparison because of process node. When the others get to 5nm (or Intel's 4?) we can finally do a "real" comparison right?

I'm impressed with what Apple has done, but just not really interested in laptops/phones. They are definitely ahead of competition in these markets from a hardware standpoint. Are there any gaming benchmarks done on th M1 SoCs?

 

[jpiniero]

At 32 big cores Apple will comfortably blow away the current Mac Pro (including the potential Ice Storm upgrade) and beat the current AMD GPU options on offer. We'll have to see how it does against Nvidia's biggest and baddest workstation GPU at the time, but with access to 1.6 TB/sec of memory bandwidth, it won't be a slouch.

Speaking of workstation, does anyone know if the M1 GPU supports DP?

 

[Heartbreaker]

Apple could conceivably continue to refresh the Cheese Grater with new Xeon W processors as long as they are willing to do the development work.

I expect Apple will stick to the transition plan, all models will have ARM cores next year. The Mac Pro may linger in the lineup, but I wouldn't expect it to get an upgrade after next year.

After this years "too good to be true" 32 core GPU in a laptop, turned out to actually be true, it looks like Gurman really had excellently placed inside sources at Apple.

So the Mac Pro really looks like it will somehow be coming with either dual or Quad M1-Max chips(or varation of). How they achieve that remains to be seen. But once you get into core counts that high, then some kind of NUMA arrangement really isn't an issue.

Since Apple controls every bit of HW/Firmware/Driver/OS SW, they can probably work the routing and scheduling efficiently.

Though you might be limited to "only" 256 GB of RAM, so it might need some work in this area.

 

[uzzi38]

You and also some other members do seem to neglect the sheer margin of advantage Apple has or are (maybe accidentally) playing it down.

For comparison's sake:

• The M1 Firestorm core and the TGL 1165G Willow Cove do have near identical CB23 ST scores. But while the TGL needs 20w avg. Power on its one core the M1 only needs 3,8w. So in other terms: At ISO performance the M1 is more than 500% more energy efficient.
• Now let's take Cezanne: Here the M1 is even 11% faster than a 5900HS. And the Cezanne needs 12w avg. power. So here M1 has a better performance AND is also almost 500% more energy efficient (because it needs less time to finish the run while also consuming less on average).

So please - to all those that think that 5nm vs. 7nm was the reason for their superiority: It is not.
The process advantage is almost negligible when looking at the numbers.

/edit:
@uzzi38 I don't know where you have your numbers from. As written above the efficiency advantage is not 15 or 30% but 500%.

Like I said in that post, drop the clocks on Zen 3 and that efficiency gap falls significantly.

Comparing efficiency at max clocks is pointless. Comparing performance at the same power is a much more sensible metric.

 

[BorisTheBlade82]

@uzzi38
Really? So Cezanne is already slower than M1. How much performance would you like to sacrifice in order to improve the efficiency?
And by the way: At less than 12w cTDP Renoir's efficiency actually decreases. I tested that. There is a clear peak. And Cezanne will behave much in the same way at least as far as I think of.

 

[igor_kavinski]

M1 Max should break the 150fps barrier in this game at these settings. Any idea if Ian has placed an order?

 

[jeanlain]

Like I said in that post, drop the clocks on Zen 3 and that efficiency gap falls significantly.

Comparing efficiency at max clocks is pointless. Comparing performance at the same power is a much more sensible metric.

That's why Apple showed performance vs power curves. Sure, we do not have the raw data, but I expect these graphs to reflect reality. At least this time Apple specified power consumption values on the X axis and the PC laptop used.

 

[Hitman928]

You and also some other members do seem to neglect the sheer margin of advantage Apple has or are (maybe accidentally) playing it down.

For comparison's sake:

• The M1 Firestorm core and the TGL 1165G Willow Cove do have near identical CB23 ST scores. But while the TGL needs 20w avg. Power on its one core the M1 only needs 3,8w. So in other terms: At ISO performance the M1 is more than 500% more energy efficient.
• Now let's take Cezanne: Here the M1 is even 11% faster than a 5900HS. And the Cezanne needs 12w avg. power. So here M1 has a better performance AND is also almost 500% more energy efficient (because it needs less time to finish the run while also consuming less on average).

So please - to all those that think that 5nm vs. 7nm was the reason for their superiority: It is not.
The process advantage is almost negligible when looking at the numbers.

/edit:
@uzzi38 I don't know where you have your numbers from. As written above the efficiency advantage is not 15 or 30% but 500%.

I'm not sure you're comparing apples to apples in the single thread efficiency score. How do the chips compare in multi-threaded? Comparing the power consumption, especially single thread, between different architectures (especially Apple silicon) can be tricky because they report things in different ways. There's also significantly more IO support on the x86 side compared to M1 which will show the largest impact on idle and single threaded power consumption if you are taking package power on x86 as your measurement.

 

[Eug]

Heh. Linus Tech Tips just bragged they spent $20k on several variations of the MacBook Pros to review.

It seems some of the high end config-to-order machines are shipping later though, so it may take some time before we get all the reviews of the full slate of machines.

 

[repoman27]

What do you mean "how"? The "2C" and "4C" obviously means two chips and four chips, and there would be some sort of controller chip like AMD's I/O die to connect them.

I'd guess that it would keep copies of all the cache tags for all M1 Max chips to reduce unnecessary snoop traffic, and have DDR5 controllers to allow a few DIMM slots to allow expansion beyond 256 GB for that subset of customers who need it - and perhaps keep its own cache for that DIMM expansion memory.

At 32 big cores Apple will comfortably blow away the current Mac Pro (including the potential Ice Storm upgrade) and beat the current AMD GPU options on offer. We'll have to see how it does against Nvidia's biggest and baddest workstation GPU at the time, but with access to 1.6 TB/sec of memory bandwidth, it won't be a slouch.

I can almost guarantee you that there will not be an additional I/O chip.

After studying the die shot, and noting the complete lack of I/O on the bottom edge, I'm thinking that there's still more to this floor plan that we haven't seen yet. In other words, M1 Max is itself a chop, and Jade 2C and 4C will use a slightly longer version. Apple should be able to comfortably fit three PCIe Gen5 x16 interfaces for CXL interconnect links along the bottom edge of the die. Since Apple stuck with a conventional organic substrate and (slightly crazy) LPDDR memory, there is no issue with scaling up the package to accommodate 2 SoC dies and 8 memory packages, or even 4 SoC dies and 16 memory packages.

I'm now far less convinced that any additional DDR controllers will be forthcoming though. For a high-end iMac, the 2C maxing out at 128GB of LPDDR5 is probably fine, and depending on the design, they might even be able to jam the 4C in there. On the other hand, a Mac Pro with the 4C would max out at 256GB, which is a lot less than the current Intel Mac Pro. Perhaps this will go in the rumored smaller Mac Pro, which can be smaller because it sheds the DIMM slots and most of its PCIe slots. And maybe the big Mac Pro gets an Icelake refresh to keep that market segment satisfied for a couple years until Apple does their next round of pro chips. I guess we'll know in another 8 months.

One other thing to note about this strategy is that although there will be reduced bandwidth / increased latency between dies, Apple probably won't have to reduce clock speeds compared to the M1 Max at all. TDP will be around 340W, but whatever, that's still less than a single GA102 in an NVIDIA RTX 3090.

 

[StinkyPinky]

Heh. Linus Tech Tips just bragged they spent $20k on several variations of the MacBook Pros to review.

It seems some of the high end config-to-order machines are shipping later though, so it may take some time before we get all the reviews of the full slate of machines.

Would think they only need to order four types? The base 14" cpu/gpu with some cut cores, the M1 Pro maxed out, the 24 core max, and the 32 core m1 max. All with 32GB ram I guess.

 

[The Hardcard]

Whereas,

The accuracy of previous rumors about the dies were almost spot on,

and

given the rumors implying that Apple will lash 2 and 4 M1 Max dies together

Let it be stated on this day

With full awareness of the unlikely nature

and acceptance of the additional costs to be passed on to me and then some

I claim desire of a Jade 2C Chop for Mac Minis and 16 inch laptops.

Even a cut down double M1 Pro 12 P + 4 E + 28 GPU and I’d be happier than a 5-year-old in a bounce house.