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

reggie_fils_aime · May 31, 2024

Mahboi said:
If we were in 2022/2023, I'd agree. But the problem is, 2024 is the year of (the paper launch of) Lunar Lake. 2025 will be the year of Z5 LP.
The point isn't that the perf is better or worse. The point is that the companies that can do more in terms of perf, can do less for less watts. Apple can't do what Intel/AMD can, but Intel/AMD can do it, and if the latest Skymont rumours prove true, they already effectively have. Without news on Z5 LP yet, we don't know how well will AMD do, but they haven't disappointed with CPUs in a looooong time.

But that doesn't address the point that people who use/have used MacOS are, let's say, "accustomed" to a certain level of performance. The M chips are snappy, the user experience is still great, and poor relative value and performance has never once stopped them from selling a boatload of machines. And all of this is ignoring the one huge thing that an x86 windows laptop has never been able to do: Provide equivalent performance on and off AC power. Benchmarks are all fine and good, but until a windows laptop can pretend at an equivalent experience to running your Macbook Pro into the ground for 16 hours, I'm not sure people will notice or care.

Mahboi said:
If Apple sells a large area, highly efficient chip, and it works for their business, within a year or so, Intel/AMD and QC will all have M1/M2 tier stuff. If they wanted to make a really phat SoC with lower frequencies to get more perf but not too much wattage, they could get that done very well too.
Apple is now in their little corner where they have no personnel to raise their IPC, and have proven that the low power decent perf market is real. They showed everyone how big that market was, and lost the ability to compete further with the others.
Not good.

None of this is wrong, and the stagnation we've seen from Apple on ARM is something to keep an eye on, but I would like to bring up one more thing which I think will save their bacon for a long time - diminishing returns. Computers have taken a quantum leap forward in processing power since 2020, and a great number of regular users struggle to fully utilize even an M1 Macbook Air aside from wringing the hell out of their ssds with swap. I could very easily see Apple coasting on the M chips like they did with the 68k back in the early days of the Mac.

igor_kavinski · May 31, 2024

The Hardcard said:
Plus, I still have the question of whether Apple is even done with this first design. Not only does analysis from Andrei and Geekerwan (to the extent they can be relied on) indicate constant changes in Apple microarchitecture, Geekerwan’s measurements show back and forth sizing or reorder structures.

So they are optimizing/tweaking the base architecture. Coz they don't have a clue how to make a new, better one?

The Hardcard · May 31, 2024

igor_kavinski said:
E-core is like the maid. Essential but do you really care how skinny it is?

It may be the key to fanless bigger than tablet devices.

Nothingness · May 31, 2024

The Hardcard said:
Exactly, so why a discussion about a problem at Apple when we don’t know what it is or even if there is one? It is weird to worry about what remains the best.

I definitely agree.

But I still wonder about the IPC improvement slowdown. As I wrote previously, we might have reached some sort of plateau. I know teams working on very advanced branch predictors and data prefetchers; they have come to the point where each small increase in performance (read a few percents on a single benchmark and <0.5% everywhere else) costs a lot of area and power.

The Hardcard said:
Plus, I still have the question of whether Apple is even done with this first design. Not only does analysis from Andrei and Geekerwan (to the extent they can be relied on) indicate constant changes in Apple microarchitecture, Geekerwan’s measurements show back and forth sizing or reorder structures.

Question being, is the first Apple architecture really done yet? I see a possible no to that, at least as of M3.

Good points. But doesn't that also show they are still tweaking the same design again and again? After a few iterations the improvements get smaller and smaller.

I suspect they have another team working on a brand new micro architecture, Apple can afford it and they'd be crazy not doing it. Will they succeed? Only time will tell.

The Hardcard · May 31, 2024

igor_kavinski said:
So they are optimizing/tweaking the base architecture. Coz they don't have a clue how to make a new, better one?

Or they have clues, but finishing this is what’s on the agenda now. After all, they have clearly leapt into uncharted territory with a huge amount of new concepts to analyze and assess with actual algorithms flowing through silicon.

How are people not more concerned that not only does Oryon not beat Apple in single core, but the multicore scaling appears to be worse. 12 big cores trading blows with 6 performance and 6 efficiency cores. And M4 implies that a Pro version would be far ahead in multicore. I’m more concerned about Qualcomm/Nuvia’s chops than Apple’s.

The Hardcard · May 31, 2024

Accident.

Mahboi · May 31, 2024

reggie_fils_aime said:
But that doesn't address the point that people who use/have used MacOS are, let's say, "accustomed" to a certain level of performance. The M chips are snappy, the user experience is still great, and poor relative value and performance has never once stopped them from selling a boatload of machines.

Performance makes snappy, it's pretty much the same thing.

reggie_fils_aime said:
And all of this is ignoring the one huge thing that an x86 windows laptop has never been able to do: Provide equivalent performance on and off AC power. Benchmarks are all fine and good, but until a windows laptop can pretend at an equivalent experience to running your Macbook Pro into the ground for 16 hours, I'm not sure people will notice or care.

No no no no. I'm so tired of reading this. x64 has no relationship with on or off AC, set power draw does. x64 just historically made bigger and more power consuming chips. Make a 15W x64 chip and it'll run off AC just as well as on. And LNL is precisely designed to offer that kind of continuous low power. So will Z5 LP I expect, but that's pretty far away still.

reggie_fils_aime said:
None of this is wrong, and the stagnation we've seen from Apple on ARM is something to keep an eye on, but I would like to bring up one more thing which I think will save their bacon for a long time - diminishing returns. Computers have taken a quantum leap forward in processing power since 2020, and a great number of regular users struggle to fully utilize even an M1 Macbook Air aside from wringing the hell out of their ssds with swap. I could very easily see Apple coasting on the M chips like they did with the 68k back in the early days of the Mac.

That's fair, CPUs have indeed entered an era where you can more or less ignore them. The problems are more on the software side to fully demand on them right now.

reggie_fils_aime · May 31, 2024

Mahboi said:
No no no no. I'm so tired of reading this. x64 has no relationship with on or off AC, set power draw does. x64 just historically made bigger and more power consuming chips. Make a 15W x64 chip and it'll run off AC just as well as on. And LNL is precisely designed to offer that kind of continuous low power. So will Z5 LP I expect, but that's pretty far away still.

What? It's night and day for sustained loads, man. Lower wattage doesn't matter in this case, because this behavior is universal for every M1-4 chip. Every single one of them is equivalent on/off the wall, which would only ever be the case in a windows machine that is engineered to be as low TDP as possible. The second the PC had an all-core load, it would fold. This has not changed radically in the time since, not in a way that would make someone who owns a MacBook Pro say "dang, should've bought that Core Ultra 7 155H".

Mahboi · May 31, 2024

reggie_fils_aime said:
What? It's night and day for sustained loads, man. Lower wattage doesn't matter in this case

https://www.notebookcheck.net/Intel-Core-i7-12800H-Processor-Benchmarks-and-Specs.589168.0.html

Now try the same comparison with something that respects a proper lower wattage like a 8840hs:

https://www.notebookcheck.net/AMD-Ryzen-7-8840HS-Processor-Benchmarks-and-Specs.780985.0.html

And then let me know if it's even close to the same loss of performance.
Intel being a wreck doesn't mean a thing about x64.

Nothingness · May 31, 2024

-- DELETE --

igor_kavinski · May 31, 2024

reggie_fils_aime said:
View attachment 100126

That 12800H also must've really bad high latency slow RAM paired with it. All that waiting for memory accesses and data to arrive is making it fare really badly in this test while the Apple laptop enjoys massive bandwidth. Yes, that's a plus on Apple's side but this particular x86 laptop is the wrong one to compare against.

reggie_fils_aime · May 31, 2024

igor_kavinski said:
That 12800H also must've really bad high latency slow RAM paired with it. All that waiting for memory accesses and data to arrive is making it fare really badly in this test while the Apple laptop enjoys massive bandwidth. Yes, that's a plus on Apple's side but this particular x86 laptop is the wrong one to compare against.

This is quibbling around the point, everyone knows that Windows PCs cannot compare to ARM Macs when they're off AC power. If that WERE the case, we'd never hear the end of it. A Macbook Air/Pro can run through it's entire battery without dropping an ounce of performance vs. being on the wall. Show me a single windows machine that can do this in a real world workload.

igor_kavinski · May 31, 2024

reggie_fils_aime said:
Show me a single windows machine that can do this in a real world workload.

A single machine might be out there. Gonna be hard to find it, though

I personally benchmarked a Ryzen 5825U laptop. It was mostly similar in performance ON/OFF AC. Not as well as an Mx chip but it was notably better than anything I've seen an Intel chip do.

reggie_fils_aime · May 31, 2024

igor_kavinski said:
A single machine might be out there. Gonna be hard to find it, though

I personally benchmarked a Ryzen 5825U laptop. It was mostly similar in performance ON/OFF AC. Not as well as an Mx chip but it was notably better than anything I've seen an Intel chip do.

As someone who is using a 7730U, and hasn't bought an intel laptop with new money since like ~2020, ryzen is damn good

Mahboi · May 31, 2024

reggie_fils_aime said:
This is quibbling around the point, everyone knows that Windows PCs cannot compare to ARM Macs when they're off AC power.

This IS the point. Power draw is the problem.

reggie_fils_aime said:
Show me a single windows machine that can do this.

Probably any super low power, Mendocino tier machine.

https://www.pcmag.com/reviews/acer-aspire-3-a315-24p

Also, whether I find a machine or not, Windows can just wildly decide to lower power by default. A more interesting (because controllable) test would be on Linux.

reggie_fils_aime · May 31, 2024

Mahboi said:
This IS the point. Power draw is the problem.

Probably any super low power, Mendocino tier machine.

https://www.pcmag.com/reviews/acer-aspire-3-a315-24p

Also, whether I find a machine or not, Windows can just wildly decide to lower power by default. A more interesting (because controllable) test would be on Linux.

Okay, but again: Apple doesn't make you choose a gimped model to get equivalent AC/DC performance, they just do it. Using Mendocino, Lunar Lake, or any other ULV thing as a gotcha isn't really the trump card you think it is when an 800 dollar 3 year old Macbook can outclass a brand new machine with an i7 equivalent when you're actually using it as a laptop.

edit: that's a freakin' 300 dollar machine, man. come on

Mahboi · May 31, 2024

reggie_fils_aime said:
Okay, but again: Apple doesn't make you choose a gimped model to get equivalent AC/DC performance, they just do it.

I will not repeat myself.

reggie_fils_aime said:
Using Mendocino, Lunar Lake, or any other ULV thing as a gotcha isn't really the trump card

It's not a trump card, it's basic specs. Read the wattage.

reggie_fils_aime said:
you think it is when an 800 dollar 3 year old Macbook can outclass a brand new machine with an i7 equivalent when you're actually using it as a laptop.

edit: that's a freakin' 300 dollar machine, man. come on

You "come on" and start to read what's shown. I'm not going to hand feed you any more than this. You either pay attention to the specs and software or you don't.

FlameTail · May 31, 2024

Nothingness said:
I definitely agree.

But I still wonder about the IPC improvement slowdown. As I wrote previously, we might have reached some sort of plateau. I know teams working on very advanced branch predictors and data prefetchers; they have come to the point where each small increase in performance (read a few percents on a single benchmark and <0.5% everywhere else) costs a lot of area and power.

So the only way to get significant uplifts (>10%) is by brute worce widening of the core (decoders/ALUs etc..) ?

FlameTail · May 31, 2024

The Hardcard said:
How are people not more concerned that not only does Oryon not beat Apple in single core, but the multicore scaling appears to be worse. 12 big cores trading blows with 6 performance and 6 efficiency cores. And M4 implies that a Pro version would be far ahead in multicore. I’m more concerned about Qualcomm/Nuvia’s chops than Apple’s.

Good assessment. Come to the Qualcomm thread and we'll talk.

reggie_fils_aime · May 31, 2024

Mahboi said:
I will not repeat myself.

It's not a trump card, it's basic specs. Read the wattage.

You "come on" and start to read what's shown. I'm not going to hand feed you any more than this. You either pay attention to the specs and software or you don't.

Do you actually use laptops, or merely read graphs about them?

FlameTail · May 31, 2024

Mahboi said:
If we were in 2022/2023, I'd agree. But the problem is, 2024 is the year of (the paper launch of) Lunar Lake. 2025 will be the year of Z5 LP.
The point isn't that the perf is better or worse. The point is that the companies that can do more in terms of perf, can do less for less watts. Apple can't do what Intel/AMD can, but Intel/AMD can do it, and if the latest Skymont rumours prove true, they already effectively have. Without news on Z5 LP yet, we don't know how well will AMD do, but they haven't disappointed with CPUs in a looooong time.

Zen5 LP or Skymont isn't going to beat the efficiency of Apple's E-core.

reggie_fils_aime · May 31, 2024

FlameTail said:
Zen5 LP or Skymont isn't going to beat the efficiency of Apple's E-core.

No, but don't you see, it's the same kind of mediocre regardless of AC power. That means Apple is doomed!

poke01 · May 31, 2024

Mahboi said:
Every single Apple chip has taken more and more area. They're all more expensive to produce than the last. M4 is just M3 but bigger, M2 is M1 but bigger, etc. There is no strong rework, it's tweaks and bigger area (and higher frequencies), which ofc at the prices they sell, Apple can afford, BUT that's only because there is no competitor on the market.

More transistors yes but the total area depends on the node. A17 is smaller than A16.

All chips are more expensive than the last unless you use a cheaper node. Every single one from any company is more expensive.

Mahboi said:
I would expect the phone cores to not be the size of a laptop core. How about comparing M4 to Zen 4?

Apple P cores are the same from A series to M series. Only the frequency changes and memory bandwidth available.

Okay here is the area for an Apple P core. That’s some nice looking die shots.

It’s the fastest P core and it’s not big for the performance it delivers. It’s not at all a fat core. Apple has been maintaining the P core area under 3.00mm^2 for a while now. The A17 P core at around 2.20mm^2 smaller will be smaller than Intels Lion Cove. Yes it’s also smaller than Zen 4c. Let’s see Zen 5 core size on Sunday/Monday.

To recap this a core that has the highest IPC while being smaller than any P core from the rest of the industry. Not a fat core by any means. If some say I don’t trust the measurement's then take a look at the die shot above, that is non-disputable. It’s the same scale as the other Cortex cores.

FlameTail said:
Oryon V2

why not V1?

Oh and one more thing, Apple is known to suprise the industry. When I made that comment earlier I was talking about how upping frequency should never be the only goal. Apple sells a lot of fanless devices and for them efficiency is a must.

What they have achieved in the iPad Pro is only because of the M4 chip. Not even Lunar Lake will have the same level of performance as the M4 and efficiency(Those E cores are amazing). There is a reason why Intel went from having performance per watt leadership as a tagline for Lunar Lake to just most power efficient x86 SoC.

Mahboi said:
Probably any super low power, Mendocino tier machine.
https://www.pcmag.com/reviews/acer-aspire-3-a315-24p

An M3 Max has about the idle efficiency as a base M3. It’s very funny that people have compare these low power x86 CPUs to even match the idle efficiency of a M chip. There are so many ways Apple is still ahead, anyway that’s a story when Lunar Lake comes out and like I said I will compare M3/M4 in a seperate thread.

Hitman928 · May 31, 2024

poke01 said:
More transistors yes but the total area depends on the node. A17 is smaller than A16.

All chips are more expensive than the last unless you use a cheaper node. Every single one from any company is more expensive.

View attachment 100135

Apple P cores are the same from A series to M series. Only the frequency changes and memory bandwidth available.

Okay here is the area for an Apple P core. That’s some nice looking die shots.
View attachment 100136

It’s the fastest P core and it’s not big for the performance it delivers. It’s not at all a fat core. Apple has been maintaining the P core area under 3.00mm^2 for a while now. The A17 P core at around 2.20mm^2 smaller than Zen 4 and will be smaller than Intels Lion Cove. Yes it’s also smaller than Zen 4c. Let’s see Zen 5 core size on Sunday/Monday.

To recap this a core that has the highest IPC while being smaller than any P core from the rest of the industry. Not a fat core by any means. If some say I don’t trust the measurement's then take a look at the die shot above, that is non-disputable. It’s the same scale as the other Cortex cores.
View attachment 100138

why not V1?

Oh and one more thing, Apple is known to suprise the industry. When I made that comment earlier I was talking about how upping frequency should never be the only goal. Apple sells a lot of fanless devices and for them efficiency is a must.

What they have achieved in the iPad Pro is only because of the M4 chip. Not even Lunar Lake will have the same level of performance as the M4 and efficiency(Those E cores are amazing). There is a reason why Intel went from having performance per watt leadership as a tagline for Lunar Lake to just most power efficient x86 SoC.

An M3 Max has about the idle efficiency as a base M3. It’s very funny that people have compare these low power x86 CPUs to even match the idle efficiency of a M chip. There are so many ways Apple is still ahead, anyway that’s a story when Lunar Lake comes out and like I said I will compare M3/M4 in a seperate thread.

What caches are you including in your core comparison? I’m pretty sure Zen 4 is smaller if you include the L2 caches, at least Zen 4c.

poke01 · May 31, 2024

Hitman928 said:
What caches are you including in your core comparison? I’m pretty sure Zen 4 is smaller if you include the L2 caches, at least Zen 4c.

Yes you’re right the Zen 4c is smaller without any cache. Zen 4c is also weaker in 1t and delivers less performance than a Zen 4 core.

A Zen 4 (without cache)core should be about the same size as the A17 core excluding cache. I have edited the post.

I think we can dismiss Apples cores being fat. They are not at all bad area wise considering the performance.

This is not an attempt to make fun. Both AMD and Apple make cores that suit their respective markets first. Both make excellent CPU cores but in the end execution to the end user matters and Apple delivers the highest IPC for now.

Discussion Apple Silicon SoC thread

Lifer

Member

Lifer

Senior member

Diamond Member

Senior member

Senior member

Golden Member

Member

Golden Member

Diamond Member

Lifer

Member

Lifer

Member

Golden Member

Member

Golden Member

Diamond Member

Diamond Member

Member

Diamond Member

Member

Platinum Member

Diamond Member

Platinum Member