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

 

[Heartbreaker]

Until this hardware is in someone’s capable hands I don’t believe any of Apples claims. I’m probably wrong but I don’t trust Apple. Consummate puffing the goods.

What's not to believe? You can simply extrapolate based on increased CPU/GPU core counts over the M1, and performance results are essentially what Apple claims. No real reason to disbelieve that. Apple's not going to lie about core counts. They aren't Theranos or Nikola.

The amazing part of this, is not the the extra cores deliver on expected performance.

It's that they built something so beastly for a Laptop. Eight Memory channels in a laptop SOC! That's what highest end workstations Xeon/Epic CPUs have.

 

[eek2121]

What's not to believe? You can simply extrapolate based on increased CPU/GPU core counts over the M1, and performance results are essentially what Apple claims. No real reason to disbelieve that. Apple's not going to lie about core counts. They aren't Theranos or Nikola.

The amazing part of this, is not the the extra cores deliver on expected performance.

It's that they built something so beastly for a Laptop. Eight Memory channels in a laptop SOC! That's what highest end workstations Xeon/Epic CPUs have.

Everything about statement is incorrect. The M1 contained 8 CPU cores. 4 big and 4 small. The M1 Pro, for example, ALSO only contains 8 cores. Do I need to further elaborate?

The M1 Pro/M1 Pro Max must be evaluated on their own terms thanks to significantly different power targets and core types.

 

[Eug]

What's not to believe? You can simply extrapolate based on increased CPU/GPU core counts over the M1, and performance results are essentially what Apple claims. No real reason to disbelieve that. Apple's not going to lie about core counts. They aren't Theranos or Nikola.

Who is Nikola?

The amazing part of this, is not the the extra cores deliver on expected performance.

It's that they built something so beastly for a Laptop. Eight Memory channels in a laptop SOC! That's what highest end workstations Xeon/Epic CPUs have.

You know it's funny. It's expected, but I'm seeing people now asking which M1 Pro to get to run Chrome and MS Office. They're considering 8+2 + 32 GB RAM for this. Apple's marketing is working. Even though they specifically say this is for pro work, it's pushing even average joes and jills to go for the upsell.

Meanwhile I continue to espouse the benefits of light weight and fanless operation (16 GB MacBook Air M1) to business users and students taking notes, etc. but I have a sneaking suspicion that a lot of people will ignore the advice get a 32 GB M1 Pro in order to "future proof" their machine for their "heavy" Office usage.

 

[insertcarehere]

Everything about statement is incorrect. The M1 contained 8 CPU cores. 4 big and 4 small. The M1 Pro, for example, ALSO only contains 8 cores. Do I need to further elaborate?

The M1 Pro/M1 Pro Max must be evaluated on their own terms thanks to significantly different power targets and core types.

M1: 4 Big + 4 Small = 8 CPU Cores
M1 Pro/Max: 8 Big + 2 Small = 8 CPU Cores?

Do I need to further elaborate?

Apple has straight up listed the laptop models (down to the SKU) they're comparing performance figures against, that's more detail than what Intel or AMD disclose in their comparisons.

 

[Heartbreaker]

Who is Nikola?

Nikola admits prototype was rolling downhill in promotional video

“This thing fully functions,” Trevor Milton said in 2016. It didn’t.

arstechnica.com

You know it's funny. It's expected, but I'm seeing people now asking which M1 Pro to get to run Chrome and MS Office. They're considering 8+2 + 32 GB RAM for this. Apple's marketing is working. Even though they specifically say this is for pro work, it's pushing even average joes and jills to go for the upsell.

Meanwhile I continue to espouse the benefits of light weight and fanless operation (16 GB MacBook Air M1) to business users and students taking notes, etc. but I have a sneaking suspicion that a lot of people will ignore the advice get a 32 GB M1 Pro in order to "future proof" their machine for their "heavy" Office usage.

People always convince themselves they need the best. Remember when people were outraged that you could only get 16GB of RAM in the M1 Macs, when all they wanted to do was normal home computing (where 8GB would have been fine).

 

[Doug S]

What's not to believe? You can simply extrapolate based on increased CPU/GPU core counts over the M1, and performance results are essentially what Apple claims. No real reason to disbelieve that. Apple's not going to lie about core counts. They aren't Theranos or Nikola.

There were a certain segment of people who were arguing that making more CPU and GPU cores work was some sort of black magic that Apple was incapable of, and that while M1 was impressive for what it was they maintained they'd be unable to scale it up to bigger systems. Mostly the same people who said that while Apple could do a good phone SoC, doing PC CPUs was a different ballgame and Apple would be unable to scale up there. They were in denial even after the M1 was announced and the first benchmarks started leaking, but once full reviews from Anandtech and other sites made it clear Apple's claims were legit they moved the goalposts to suggesting that Apple would somehow find it impossible to properly scale larger than M1.

Once they are similarly forced to accept Apple's M1 Max claims are legit they will move the goalposts again and say "well Apple may have made a good laptop chip, but there's no way it'll scale up to compete with the biggest workstation Xeon CPUs and discrete AMD GPUs". Even though the 2C / 4C rumors from months ago already showed how they will scale things up - and the 400 MB/sec LPDDR5 bandwidth per M1 Max (or M1 "Extreme" if M1 Max is itself a chop of an even larger die that has the interchip fabric I/O on the bottom) shows how they plan to keep 128 GPU cores fed.

They will always have something to poke at. They'll say "well its multithread performance still can't match a 64 core Threadripper" or whine about stupid stuff like lack of hardware support for AV1. There's always something with people like that.

 

[Heartbreaker]

They will always have something to poke at. They'll say "well its multithread performance still can't match a 64 core Threadripper" or whine about stupid stuff like lack of hardware support for AV1. There's always something with people like that.

Yeah, I see a few that are obviously carrying a massive chip on their shoulders.

I've never owned a laptop, or an Apple product, but wow, these machines just look so amazing, it makes me want to change that.

I thought the first M1 machines were impressive last time, and this is even more impressive. The only times I have really gone "wow" in that last year or so has been for the new Apple HW.

 

[DrMrLordX]

They will always have something to poke at. They'll say "well its multithread performance still can't match a 64 core Threadripper" or whine about stupid stuff like lack of hardware support for AV1. There's always something with people like that.

Or it could be the HDMI 2.0 ports:

The new MacBook Pro seems to have an HDMI 2.0 port, not 2.1

120 Hz 4K monitors aren’t the norm today, but they probably will be later.

arstechnica.com

(okay okay, it's not really related to M1X but still)

Only thing that's making me arch an eyebrow here is the ST performance. It hasn't moved much since M1. A15 didn't move the bar much either (versus A14). It's like Apple is giving Qualcomm and AMD time to play catchup.

 

[Eug]

Only thing that's making me arch an eyebrow here is the ST performance. It hasn't moved much since M1. A15 didn't move the bar much either (versus A14). It's like Apple is giving Qualcomm and AMD time to play catchup.

Who was expecting M1X/Pro/Max ST to move significantly from M1? I recall many here predicting it would be the same.

 

[DrMrLordX]

Who was expecting M1X/Pro/Max ST to move significantly from M1? I recall many here predicting it would be the same.

Nobody of whom I am aware. Apple is focusing on expanding core count and demonstrating that they can do interconnect too. Meanwhile everyone else is still focusing on, well, both. It will be interesting to see how far Apple can go when they do finally decide to improve ST performance.

 

[majord]

Well, it was not me who measured M1, but rather Andrei and @sallymander. Please read #42ff. for further details. But how I see it, it is pretty much "apples to apples". And who does really care if the advantage is 400% or 500%?

First you are falling into the trap of measuring Watt/Perf , not Perf/watt ..

Both of these are venerable to being missleading, dependent on where you set the reference point, but more so with the former.

In this case you're choosing a performance point (~1500 1T CB23 ) that thanks to a significant IPC deficit (approx 35-37% higher ST IPC for M1) is only achievable for Zen3 at much higher frequencies - Frequencies it can achieve -and beyond, but at the expense of power consumption.

IF AMD decided not to clock Zen 3 beyond say , 3.2ghz, you would simply NOT be able to make this comparison. You'd instead find , rather conveniently, that at these frequencies Zen 3 vs M1 ST Power consumption is essentially the same (by numbers quoted in your thread - i.e 3.8W for M1 @ 1C-1T 3.2Ghz) , This means the M1's advantage in perf/watt is literally it's IPC .. i.e about 35-37% . A far cry from 400-500%.

I could make similar claims for Zen 3 vs Zen 2 by making a Zen3 CPU match the ST score of a 3950x, running at maximum 1C turbo - The IPC difference means the former will only have to run at , say 3.9Ghz, to match a 4.6-4.7Ghz 3950x With orders of magnitude less power as a result (around ~8w vs ~18w) With that it would be all to easy to say Zen 3 is over 200% more efficient.. Which we all know is not the case.


Secondly, this doesn't account for Throughput.. The M1 core is HEAVILY geared towards maximum 1T IPC, and low frequencies. Once you put two threads down a Zen3 core with SMT , the efficiency delta drops even more. since core power does not increase in line with perf uplift.

Some examples below: Different SKU's sustaining different frequencies..








Anyway. this will all become more than apparent if the M1X/pro scores as predicted - i.e around 12000 @ 30w - but with the help of 2x efficiency cores. Vs Cezzane @ 10500 or so @ 35w 8 core.

So no - there's no magic happening with the M1. It's just a very focused, high IPC, low clock design.

 

[Abwx]

Wrong! M1 cpu consumes only 12-15w at full load while 5980HS is far higher than 35w at full load.

Wrong.??.
I posted the Cinebench score at 35W of the 5980HS, there s different score at different TDP settings.

Cinebench R23: CPU-Ergebnisse mit AMD und Intel

Maxon hat Cinebench R23 veröffentlicht. Wie schnell welcher Prozessor im CPU-Benchmark ist, kann die Community im Schwarm ermitteln.

www.computerbase.de

13943 pts@95W
12854 pts@80W
10617 pts@35W

As for the M1 here the claimed CPU TDP :

30W according to Apple.
So this is hard numbers contrary to your statement that is raised from nowhere and is just wishfull thoughts.

@Abwx The M1 uses ~15W for ~7500 pts.

But even if it were 15W that s not that extraordinary, at this score the 5980HS use barely 15W, and that s with a 7nm process.

 

[Eug]

Wrong.??.
I posted the Cinebench score at 35W of the 5980HS, there s different score at different TDP settings.

Cinebench R23: CPU-Ergebnisse mit AMD und Intel

Maxon hat Cinebench R23 veröffentlicht. Wie schnell welcher Prozessor im CPU-Benchmark ist, kann die Community im Schwarm ermitteln.

www.computerbase.de

13943 pts@95W
12854 pts@80W
10617 pts@35W

As for the M1 here the claimed CPU TDP :

30W according to Apple.
So this is hard numbers contrary to your statement that is raised from nowhere and is just wishfull thoughts.



But even if it were 15W that s not that extraordinary, at this score the 5980HS use barely 15W, and that s with a 7nm process.

M1 ≠ M1 Pro

That graph you posted is for M1 Pro. The Cinebench score you posted is for M1, which is a much slower SoC.

Actually M1 is on the graph too, but at 15 W. M1 is the red curve on the left.

M1
15 W
Cinebench R23: 7508

M1 Pro
30 W
Cinebench R23: ???

There is no score for M1 Pro available yet but I would not be surprised if it hit 12000 or more at 30 W. However, we probably won’t know M1 Pro’s Cinebench score until next week.

 

[Leo9]

Wrong.??.
I posted the Cinebench score at 35W of the 5980HS, there s different score at different TDP settings.

Cinebench R23: CPU-Ergebnisse mit AMD und Intel

Maxon hat Cinebench R23 veröffentlicht. Wie schnell welcher Prozessor im CPU-Benchmark ist, kann die Community im Schwarm ermitteln.

www.computerbase.de

13943 pts@95W
12854 pts@80W
10617 pts@35W

As for the M1 here the claimed CPU TDP :

30W according to Apple.
So this is hard numbers contrary to your statement that is raised from nowhere and is just wishfull thoughts.



But even if it were 15W that s not that extraordinary, at this score the 5980HS use barely 15W, and that s with a 7nm process.

?
M1 is at 15W.
M1 Pro/Max at 30W and the score will be much higher than 10k (5980HS).
Also Cinebench is just one benchmark.

 

[insertcarehere]

First you are falling into the trap of measuring Watt/Perf , not Perf/watt ..

Both of these are venerable to being missleading, dependent on where you set the reference point, but more so with the former.

In this case you're choosing a performance point (~1500 1T CB23 ) that thanks to a significant IPC deficit (approx 35-37% higher ST IPC for M1) is only achievable for Zen3 at much higher frequencies - Frequencies it can achieve -and beyond, but at the expense of power consumption.

IF AMD decided not to clock Zen 3 beyond say , 3.2ghz, you would simply NOT be able to make this comparison. You'd instead find , rather conveniently, that at these frequencies Zen 3 vs M1 ST Power consumption is essentially the same (by numbers quoted in your thread - i.e 3.8W for M1 @ 1C-1T 3.2Ghz) , This means the M1's advantage in perf/watt is literally it's IPC .. i.e about 35-37% . A far cry from 400-500%.

I could make similar claims for Zen 3 vs Zen 2 by making a Zen3 CPU match the ST score of a 3950x, running at maximum 1C turbo - The IPC difference means the former will only have to run at , say 3.9Ghz, to match a 4.6-4.7Ghz 3950x With orders of magnitude less power as a result (around ~8w vs ~18w) With that it would be all to easy to say Zen 3 is over 200% more efficient.. Which we all know is not the case.


Secondly, this doesn't account for Throughput.. The M1 core is HEAVILY geared towards maximum 1T IPC, and low frequencies. Once you put two threads down a Zen3 core with SMT , the efficiency delta drops even more. since core power does not increase in line with perf uplift.

Some examples below: Different SKU's sustaining different frequencies..


View attachment 51635


View attachment 51637


Anyway. this will all become more than apparent if the M1X/pro scores as predicted - i.e around 12000 @ 30w - but with the help of 2x efficiency cores. Vs Cezzane @ 10500 or so @ 35w 8 core.

Everyone here should know that it takes exponentially more power to achieve higher clocks, that's how silicon works. I am sure AMD knows how their cores power efficiency varies by clock, and yet they decided to sell zero consumer Zen 3 SKUs at 3.2ghz max clocks because that wouldn't be competitive at all in performance, while the M1 is more than competitive at these clocks.

So despite operating at a far more efficient V/F point for Zen 3 vs M1 in your comparison, equating the M1's package power vs Zen 3's core power, the M1 still comes out with nearly 40% (more depending on benchmark) more performance at the same/similar power.

To put that into perspective - TSMC N5 only claims 15% more performance, iso-power vs TSMC N7, so it's clear that there's much more than just process advantages at play here.

 

[leoneazzurro]

I mean,

C23 score was evaluated already on M1 by Anandtech

The 2020 Mac Mini Unleashed: Putting Apple Silicon M1 To The Test

www.anandtech.com

M1 is on nominal 15W but in MT workload it seems to draw 18-27W depending on the workloads (estimates of 20W-24W real), 4800U (which is not even the latest Zen architecture) in MT workload beat it by a wide margin and loses in ST by a wide margin: why ignore this data and point to mental Gymnastic (by the way, Cezanne ST is almost on par with the M1 in the same power envelope).

 

[tempestglen]

GPU score is coming, just for reference only, not sure about the correctness....

 

[tempestglen]

Laptop 3080's GFX

 

[igor_kavinski]

Very respectable showing by the M1 Max. The Car Chase scores are lacking because apparently, Apple's GPU cannot do Tessellation (either in hardware or there is no available API call for that in Metal).

Also, look at the offscreen scores. Apple can easily upgrade the display to 240Hz or 360Hz and it would destroy AMD/nVidia. Do high fps gaming monitors work with Apple hardware?

 

[DrMrLordX]

Just an FYI, for those of you who don't know GFXBench, Manhattan and T-Rex are old by feature set. Manhattan is a unicorn DX10 (?!?!?) benchmark, while T-Rex is DX9 I think? Anyway MacOS obviously doesn't use DX anything, so I'm guessing GFXBench had to port those benchmarks over to Metal.

Aztec Ruins is DX12/DX11/Vulkan on PC.

 

[sallymander]

Also, weirdly, why are the M1 Max's offscreen scores so high? What is preventing it from flexing its muscles onscreen?

The screen is running at 120Hz so I guess it's basically v-sync?

 

[igor_kavinski]

The screen is running at 120Hz so I guess it's basically v-sync?

Yeah sorry. Took a few seconds for that to register in my organic silicon

 

[insertcarehere]

Also, weirdly, why are the M1 Max's offscreen scores so high? What is preventing it from flexing its muscles onscreen?

Onscreen means running the benchmarks at the Mac's full screen resolution, which is nearly 4k. Most likely the RTX 3080 laptop's screen isn't nearly as high resolution.

 

[igor_kavinski]

Aztec Ruins is DX12/DX11/Vulkan on PC.

Thanks for the heads up on the DX level of that test. Apparently, it doesn't run on RTX or AMD DX12 hardware, which is weird. Probably GFX Bench's compatibility issue?

194 fps in High Tier offscreen. That must be a stressful test for the Apple GPU, compared to the others.

 

[Accord99]

Thanks for the heads up on the DX level of that test. Apparently, it doesn't run on RTX or AMD DX12 hardware, which is weird. Probably GFX Bench's compatibility issue?

Could be nobody bothered to run those tests with a 3070 Ti.