Discussion Apple Silicon SoC thread

Page 154 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Eug

Lifer
Mar 11, 2000
23,583
996
126
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:

Screen-Shot-2021-10-18-at-1.20.47-PM.jpg

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:


M1 Ultra discussion here:


M2 discussion here:


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:

 
Last edited:

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
Apple showed with the M1 Ultra that they can match 80-90% of the performance of an Intel 12900K while consuming 35% of the power. Sure, the average Windows user might not be as sensitive to performance per watt as an Apple user but I don’t think Intel and AMD can match the performance per watt of ARM chips, even if they put a ton more resources into it and Windows did arhictectural changes to Windows. I think it comes down to ARM phasing out x86.
What you are missing here, is that you CAN set the wattage way down on both AND and Intel and still have 90% of the performance. You just seem to have tunnel vision for Apple.
 

richardskrad

Member
Jun 28, 2022
51
44
51
Interesting… Maybe I’m working with outdated information.I would like to educate myself. Where can I read about recent Intel and AMD chips and how they’re matching the performance per watt of the M1/M2? Also, is there a Windows laptop out there right now that has the performance and battery life in a fan-less enclosure just like the M1/M2 MacBook Air?
 
  • Like
Reactions: Lodix

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
Apple showed with the M1 Ultra that they can match 80-90% of the performance of an Intel 12900K while consuming 35% of the power. Sure, the average Windows user might not be as sensitive to performance per watt as an Apple user but I don’t think Intel and AMD can match the performance per watt of ARM chips, even if they put a ton more resources into it and Windows did arhictectural changes to Windows. I think it comes down to ARM phasing out x86.
That is the miracle of voltage and frequency scaling, and die size (both larger cores die size, and more cores via throwing multiple chips together.)

A chip at 0.65, 0.90, 1.20, and 1.35 volts can have the same exact frequency but there is a quadratic (voltage*voltage) function of volts on power usage. Thus a chip with .65 vs 1.30 volts with the same frequency would consume 4x the power.

Chips are “stable” at different volts, and different frequency curves. For example running the same chip at 4.0 ghz may require 1.20 volts, and to get 5% more performance at 4.2 ghz mays require 1.35 volts, yet the 1.35 volt chip consumes roughly 32.8% more power for only 5% more speed ([1.35*1.35]/[1.20*1.20]*[4.20/4.00])

In sum maximum performance is a suckers game if you want performance per watt. This is why mobile and server actually want the chip to be wide enough as possible to set a floor on performance and run the chip at a lowest voltage possible. Yet this technique is wasteful on die size, you are making a trade off that I will gladly choose but some markets are so cut throat on margins they do not want larger die size chips.

Note all of the math I am talking about is how AMD could easily promise a few years before Zen 1 was shown in 2016 that they were going to get massive efficency improvements in 5 years. Everyone believed this was possible even without Zen when AMD was using bulldozer derivatives for lowering the voltage itself did most of the heavy lifting.

Edit: One last thing when making chips one can have the same die and then a variance of the final result which is what volts you need to apply to make the chip stable at specific frequencies. Thus server and oem customers of AMD and Intel may get a chip that has the same total performance but AMD or Intel segmented after testing this chip and saying this chip tests fine at voltage A while a less good chip needs A+0.1 volts to reach the same ghz performance. Thus they sell these chips to different customers with different monikers in the model name.
 

Thala

Golden Member
Nov 12, 2014
1,355
653
136
Interesting… Maybe I’m working with outdated information.I would like to educate myself. Where can I read about recent Intel and AMD chips and how they’re matching the performance per watt of the M1/M2? Also, is there a Windows laptop out there right now that has the performance and battery life in a fan-less enclosure just like the M1/M2 MacBook Air?

The 8CX Gen 3 devices like the Lenovo Thinkpad X13s and the Surface Pro 9 SQ3 are Windows laptops, are fan-less with a 9W TDP similar to the M1 Air with about 16% less performance - so the distance is relatively small on the efficiency side. The efficiency loss can be explained by the fact that the 8CX Gen 3 is using a Samsung process.
 

poke01

Senior member
Mar 8, 2022
651
611
106
And M2 threw away a bit of that performance per watt excellence for 10% more performance.

Even Apple isn't immune to looking bad in that specific metric.
Well I can see why. It's still based on 5mm. Until Apple moves to a new node like 3nm Pref/w will suffer.

Apples chips are not magic. Expect AMD and Intel to do wonders when they move to 3nm.
 
  • Like
Reactions: Tlh97 and Kaluan

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,618
136
Interesting… Maybe I’m working with outdated information.I would like to educate myself. Where can I read about recent Intel and AMD chips and how they’re matching the performance per watt of the M1/M2? Also, is there a Windows laptop out there right now that has the performance and battery life in a fan-less enclosure just like the M1/M2 MacBook Air?
We have a thread about a tool that allows assessing a chips power efficiency in ST and MT workloads. M1 is unbeaten in ST, but 7950X at 88W is far ahead in MT.

 
  • Like
Reactions: Tlh97 and Markfw

poke01

Senior member
Mar 8, 2022
651
611
106
We have a thread about a tool that allows assessing a chips power efficiency in ST and MT workloads. M1 is unbeaten in ST, but 7950X at 88W is far ahead in MT.

Well a 16C part should be ahead in MT.
 
  • Like
Reactions: Nothingness

Eug

Lifer
Mar 11, 2000
23,583
996
126
Rare to see a two-year-old chip still be (arguably) the best mobile part available.
Well, M2 has been out for quite some time.

Or are you suggesting M1 still wins because of its efficiency?

 

Thala

Golden Member
Nov 12, 2014
1,355
653
136
Well, M2 has been out for quite some time.

Or are you suggesting M1 still wins because of its efficiency?


The problem of course is, that notebookcheck does not really do a SoC analysis. They compare the 2 SoCs at different voltages - which is totally pointless when you want to reason about the efficiency of the SoC.
That having said, the M2 could (and most likely is) more efficient than the M1 at iso voltage.
 
  • Like
Reactions: Tlh97 and Eug

moinmoin

Diamond Member
Jun 1, 2017
4,933
7,618
136
Well a 16C part should be ahead in MT.
These charts are about performance per watt, the amount of cores doesn't matter.

Actually it does matter, but not for the reason you may be thinking of:
For excellent ST power efficiency IO and uncore need to be very frugal to let the overall chip shine in power efficiency. That's something where Apple indeed has no equals among the competition. But if your actual cores are very efficient, the more of them you got the more the IO's and uncore's power consumption can be split among cores. 7950X at 88W performing as well as it does in the MT chart where M1 doesn't do too well shows that M1's special part actually isn't its cores but its IO and uncore.

Or are you suggesting M1 still wins because of its efficiency?
From what I've seen M1 does appear to run at a more efficient frequency than M2.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
Yes the M1 is still the best. Better than M2 and AMD and MUCH better than any of Intel's offerings in the ultrabook space.

Given that M2's cores have little change from M1, and the process is tweaked for more performance (N5 vs N5P) which they are obtaining with 10% higher frequency and possibly a bit more voltage (i.e. see Roland's post above for why that matters) this isn't surprising.

They'll need a new core or new process family (so M3 using A17 core not A16, or built on N3) to change that equation.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
I hope that in a couple generations the iPhone SOC can match the 5800X. Sounds a bit ridiculous, sure, but single core is already up there, multicore trails just behind the Ryzen 5 3600: https://browser.geekbench.com/v5/cpu/18596014


Why? What's the use case for doubling its multiprocessing power? What smartphone apps are being limited by available multithread throughput? No one is doing NRT rendering or number crunching on their phone. Nobody sane, at least.

Single thread CPU, GPU, and perhaps NPU performance (as well as CPU power efficiency of course) are all more important for an iPhone.
 

eek2121

Platinum Member
Aug 2, 2005
2,904
3,905
136
Why? What's the use case for doubling its multiprocessing power? What smartphone apps are being limited by available multithread throughput? No one is doing NRT rendering or number crunching on their phone. Nobody sane, at least.

Single thread CPU, GPU, and perhaps NPU performance (as well as CPU power efficiency of course) are all more important for an iPhone.

It may surprise you at what the iPhone is capable of and what scenarios it, along with the iPad, are being used for. Small business owners, building architects, interior designers, etc. Not all the functionality they use can be powered via the GPU/NPU. Some of it needs raw INT performance. Also, you can split screen apps and stuff.

Side note: Someone released a stable diffusion app for the iPhone.
 

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
It may surprise you at what the iPhone is capable of and what scenarios it, along with the iPad, are being used for. Small business owners, building architects, interior designers, etc. Not all the functionality they use can be powered via the GPU/NPU. Some of it needs raw INT performance. Also, you can split screen apps and stuff.

Side note: Someone released a stable diffusion app for the iPhone.
But do we need more CPU power with the iPhone and not other forms of power like GPU power, Neural Processing Unit / Neural Engine, Image Signal Processor, etc?

iPhone chips of the last few generations are 2 performance and 4 efficiency cores. Do we really need a M1Max which is 8+2 or a M1Ultra which is 2x(8+2) in the tdp of an iPhone?

=====

For example the $125 Apple TV that just came out last week is an A15 (2+4) cores. Same soc as a $429 iPhone SE (and more expensive models if you want a larger screen), and the $499 iPad Mini. Do we need a faster Apple TV or is the low hanging fruit that I truly want is to run a real OS and real software on that $125 miracle device? (give me macOS, Linux, or windows please)
 
Last edited:

Eug

Lifer
Mar 11, 2000
23,583
996
126
Little news tidbit: Apparently the A15 in the new Apple TV 4K is a binned version.


A15 in Apple TV is actually a binned version with only five CPU cores as opposed to six in other A15-based products, as confirmed by the TV Info app.

I wonder how much volume Apple sells of these but I assume the amount is tiny, and there are no other devices using CPU core binned variants, so the amount of CPU core binned A15s out there overall will also be tiny.