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

lobz · Dec 6, 2020

LightningZ71 said:
I think that we all realize that the universe, assuming that it has actual bounds, has a finite number of electrons in it, this putting a ceiling on any talk of "infinity" with respect to computing. Thusly, any statement about infinity must be figurative in nature.

To quote Two and a half Men: when you're assuming, you're just a *****. 😶😶😶

TESKATLIPOKA · Dec 6, 2020

Apple MacBook Air 2020 M1 Entry Review.
Notebookcheck
1129 euro for M1, 8GB RAM and 256GB SSD.

Doug S · Dec 6, 2020

scineram said:
You tell me guys, how would Linux ever recieve support for anything in this machine, especially gpu?

It probably won't be, but you'll be able to run Linux in a VM. With some work (i.e. if enough people cared about it to put forth enough effort) they could probably get it to boot and run with a dumb framebuffer.

The question is why put in that effort, but supposedly there is an effort to run Android on the iPhone so I guess people with more free time than sense are relatively plentiful...

Heartbreaker · Dec 6, 2020

TESKATLIPOKA said:
Apple MacBook Air 2020 M1 Entry Review.
Notebookcheck
1129 euro for M1, 8GB RAM and 256GB SSD.

Does that Euro price include GST/VAT?

TESKATLIPOKA · Dec 6, 2020

guidryp said:
Does that Euro price include GST/VAT?

It should.

moinmoin · Dec 6, 2020

Doug S said:
The question is why put in that effort, but supposedly there is an effort to run Android on the iPhone so I guess people with more free time than sense are relatively plentiful...

A lot of minor driver support under Linux is still driven by volunteers/hobbyists. Obviously M1 is on a completely different scale though.

Eug · Dec 6, 2020

TESKATLIPOKA said:
Apple MacBook Air 2020 M1 Entry Review.
Notebookcheck
1129 euro for M1, 8GB RAM and 256GB SSD.

So, it drops just over 1000 points from 7409 to 6401 once throttled. IOW, the fanless Air loses a maximum of 13.6% performance in this test due to throttling, but it takes many cycles of Cinebench R23 to get to that maximum throttling.

On the third cycle of R23, it had dropped to 6682, so that's a drop of 727 or only 9.8%.

The current M1 13" Pro seems kinda pointless for 95% of users in this tier. The Air in contrast has the advantage of a design that is potentially more comfortable for your wrists, lighter weight, no noise, and proper function keys.

Heartbreaker · Dec 6, 2020

moinmoin said:
A lot of minor driver support under Linux is still driven by volunteers/hobbyists. Obviously M1 is on a completely different scale though.

They usually also have access to documentation. Apple is VERY unlikely to provide any. Extremely difficult to write useful drivers without documentation.

moinmoin · Dec 6, 2020

guidryp said:
They usually also have access to documentation.

No that's wrong, publicly available technical documentation for the vast of majority consumer electronics is "usually" terrible, incomplete or downright non-existing. In many cases you are better off looking at Linux driver code documentation (thanks to somebody having reverse engineered it already, or some company's official Linux support) than looking for official documentation.

Heartbreaker · Dec 6, 2020

moinmoin said:
No that's wrong, publicly available technical documentation for the vast majority consumer electronics is "usually" terrible, incomplete or downright non-existing. In many cases you are better off looking at Linux driver code documentation (thanks to somebody having reverse engineered it already, or some company's official Linux support) than looking for official documentation.

Terrible is better than none. Writing drivers for a complex part like a GPU without any documentation is not really practical.

For work, I updated some driver code for a Fiber optic I/O card. Just one of the chips on the board had a ~300 pages of documentation detailing every HW and SW specification. Setup was controlled and switched via both physical (control lines) and SW (control registers) which changed the meaning of other registers. It would be borderline insane to attempt write a driver without some kind of documentation.

Nearly any I/O boards will use off the shelf HW chips, and those will have documentation like this unless it's considered some kind of specific in house proprietary device (those are rare). How else would they even sell these parts if they didn't document them?

So when the Open Source community writes drives, they will sensibly find out what chips are on the board, get the manufacturer documentation for those chips, to use as a starting point.

Even Linus Torvalds thinks this is the case, but what would he know.

"The main problem with the M1 for me is the GPU and other devices around it, because that's likely what would hold me off using it because it wouldn't have any Linux support unless Apple opens up."

amrnuke · Dec 6, 2020

Eug said:
View attachment 35186

So, it drops just over 1000 points from 7409 to 6401 once throttled. IOW, the fanless Air loses a maximum of 13.6% performance in this test due to throttling, but it takes many cycles of Cinebench R23 to get to that maximum throttling.

On the third cycle of R23, it had dropped to 6682, so that's a drop of 727 or only 9.8%.

The current M1 13" Pro seems kinda pointless for 95% of users in this tier. The Air in contrast has the advantage of a design that is potentially more comfortable for your wrists, lighter weight, no noise, and proper function keys.

Did they happen to run single-thread on a loop, or GB5 on a loop?

LightningZ71 · Dec 7, 2020

As a more accessible example of a similar issue (Linux on a SOC with poor driver support), Intel partnered with PowerVR for the cedar view atoms. The sgx545 gma units barely got one working driver release out of Intel for Windows 7. After that, it was all but abandoned. PowerVR released no documentation on it, same for Intel. We're now most of a decade later and Linux has a barely functioning driver with almost no acceleration. This system is nowhere near as locked down as Apple's gear is.

Linux is going to be very difficult to make work WELL without apple supporting it.

moinmoin · Dec 7, 2020

nouveau is another example.

LightningZ71 said:
Linux is going to be very difficult to make work WELL without apple supporting it.

That much is obvious.

Heartbreaker · Dec 7, 2020

LightningZ71 said:
As a more accessible example of a similar issue (Linux on a SOC with poor driver support), Intel partnered with PowerVR for the cedar view atoms. The sgx545 gma units barely got one working driver release out of Intel for Windows 7. After that, it was all but abandoned. PowerVR released no documentation on it, same for Intel. We're now most of a decade later and Linux has a barely functioning driver with almost no acceleration. This system is nowhere near as locked down as Apple's gear is.

Linux is going to be very difficult to make work WELL without apple supporting it.

No one is as closed as Apple. NVidia has released extensive GPU documentation many years ago, and Imagination released full ISA documentation in 2014:

Imagination Releases Full ISA Documentation For PowerVR Rogue GPUs

From Apple there will be the sound of crickets. Apple has ZERO interest in making Third party HW drivers work on Apple Silicon. There will be ZERO documentation or help from Apple and thus an Apple Silicon GPU driver will go nowhere.

Eug · Dec 7, 2020

Bloomberg - Are you a robot?

Up to 16 performance CPU cores for iMac and MacBook Pro.

Up to 32 performance CPU cores for Mac Pro.

Up to 32 GPU cores for iMac.

Up to 128 GPU cores for Mac Pro.

However, not guaranteed all of these will make it to market immediately if ever. eg. May start with 8-12 performance CPU cores for MacBook Pro and iMac.

Also working on Mac Pro mini.

Heartbreaker · Dec 7, 2020

Eug said:
https://www.bloomberg.com/news/arti...ac-chips-with-aim-to-outclass-highest-end-pcs
Up to 16 performance CPU cores for iMac and MacBook Pro.

Up to 32 performance CPU cores for Mac Pro.

Up to 32 GPU cores for iMac.

Up to 128 GPU cores for Mac Pro.

However, not guaranteed all of these will make it to market immediately if ever. eg. May start with 8-12 performance CPU cores for MacBook Pro and iMac.

Also working on Mac Pro mini.

Need huge grains of Salt with that rumor. They also cover all the options so aspects will be true no matter what.

They say the next iMac Chips will have either 8, 12, or 16 high performance CPU cores. Way to cover all the bases...

For the Mac Pro. The only way they get to 128 GPU cores, is with multiple discrete chips. A 64 Core part, would be similar sized to NVidia/AMD largest parts, so possible. But I would expect they would go for something like 32 core part, and use 1-4 of them if they really aim to achieve a 128 core top end part, that would be 1 tape out they could share in multiple configs for multiple models.

Where does it say anything about Mac Pro Mini?

Eug · Dec 7, 2020

guidryp said:
Need huge grains of Salt with that rumor.

Mark Gurman has an 89% accuracy rate, the same as L0vetodream. Both of them are significantly more reliable than the Wall Street Journal.

They also cover all the options so aspects will be true no matter what.

This is the first I've heard of them working on a 32 performance core part. It makes sense, and some of here predicted in this thread that that was likely in the lab, but no big media outlet has ever said it before in print AFAIK.

Where does it say anything about Mac Pro Mini?

"For higher-end desktop computers, planned for later in 2021 and a new half-sized Mac Pro planned to launch by 2022, Apple is testing a chip design with as many as 32 high-performance cores."

I don't know if he's suggesting it means all new Mac Pros will be smaller, or if there will be a second Mac Pro line that's smaller than their main line. But it will be "mini" compared to the current Mac Pro line.

Anyhow, I'm good for now with my lowly quad-core Kaby Lake i5 iMac, but in the future I might get a Mac mini with 16 performance cores, 32 GB RAM, and 10GigE.

Heartbreaker · Dec 7, 2020

Eug said:
Mark Gurman has an 89% accuracy rate.

Easy to be accurate when you say things like 8, 12, or 16 cores. Because one of those is sure to be correct.

Basically they have covered everything from doubling to quadrupling the M1 for the iMacs next year.

Doubling up on a 16 Billion transistor part is really the likely limit for next year. Quadrupling it is unlikely in the extreme. Unless they going to separate chips on the iMac next year.

For higher-end desktop computers, planned for later in 2021 and a new half-sized Mac Pro planned to launch by 2022, Apple is testing a chip design with as many as 32 high-performance cores.

Ok. It's a smaller Mac Pro. Not a Pro version of Mini. Even half the size of the current Mac Pro will be a sizable computer, and it won't be called Mini.

Not an in between model, but a the newer, smaller Mac Pro, likely with less expansion capability. But more expansion than the Trashcan Macs.

Eug · Dec 7, 2020

guidryp said:
Easy to be accurate when you say things like 8, 12, or 16 cores. Because one of those is sure to be correct.

You haven't followed Mark Gurman very much I take it. He's definitely not perfect, but as far as rumours go, he has a lot of street cred.

And it's one thing to publish a blog, and another to put it in print for a site like Bloomberg.

And yeah, his article does cover all bases, but it is nice to see the big media outlets printing the same thing we in this thread have been predicting.

Ok. It's a smaller Mac Pro. Not a Pro Mini. Even half the size of the current Mac Pro will be a sizable computer, and it won't be called Mini. It will just be the newer, smaller Mac Pro.

Of course it won't be called a mini. That was a joke. However, I was just emphasizing it wouldn't be a behemoth like the current Mac Pro.

Heartbreaker · Dec 7, 2020

I know Gurman has an above average record, but when you publish something that just covers all the bases, it's kind of meaningless.

Eug · Dec 7, 2020

guidryp said:
I know Gurman has an above average record, but when you publish something that just covers all the bases, it's kind of meaningless.

Actually, my outta-my-ass prediction for an actual 2022 Mac Pro was that they might release a 24-performance-core model, for maybe a total of 28 CPU cores with 4 efficiency cores, or maybe no efficiency cores. I wasn't sure if that would be the top end or not, but I was not restricting the possibilities to 16 and 32. It would be 16, 24, and (possibly) 32.

LightningZ71 · Dec 7, 2020

Having gobs of gpu cores us one thing... But how are they feeding that beast? If they maintain their current setup, those chips will require packages surrounded by 8 or more LPDDRX stacks, or a couple of HBM stacks. They're going to look more like Consoles and less like traditional desktops...

amrnuke · Dec 7, 2020

LightningZ71 said:
Having gobs of gpu cores us one thing... But how are they feeding that beast? If they maintain their current setup, those chips will require packages surrounded by 8 or more LPDDRX stacks, or a couple of HBM stacks. They're going to look more like Consoles and less like traditional desktops...

They may get away with just enlarging the SoC -- at an estimated 120mm2, you can pretty easily double it and you're still under 250mm2... but once you start getting up to 16 +8 layouts, assuming they keep the cache sizes, NE, ML, and all that the same, almost 500mm2 on N5 EUV... that'll probably make them consider a chiplet design, which should be child's play for them.

awesomedeluxe · Dec 7, 2020

Eug said:
Bloomberg - Are you a robot?
Up to 16 performance CPU cores for iMac and MacBook Pro.

Up to 32 performance CPU cores for Mac Pro.

Up to 32 GPU cores for iMac.

Up to 128 GPU cores for Mac Pro.

However, not guaranteed all of these will make it to market immediately if ever. eg. May start with 8-12 performance CPU cores for MacBook Pro and iMac.

Also working on Mac Pro mini.

Wow. This is written by Gurman, but it reads like a Macrumors thread. The claim that Apple is making a 16x firestorm core part for laptops is kind of hard to believe on its face.

It does survive a baseline viability test. Clocked at iPhone speeds, 16 perf cores would use about 40W. Clocked at Mac Mini speeds, 32 GPU cores would also use about 40W. I have serious heat/area concerns about putting this in a 16", but total power consumption is lower than the ~90W of the current Intel+AMD solution.

If I had to think about how this chip actually looks... I guess there's no IGP? If "unified memory" is still a thing, I guess it's M1X - 4x LPDDR5 - GPU?

guidryp said:
Easy to be accurate when you say things like 8, 12, or 16 cores. Because one of those is sure to be correct.

Basically they have covered everything from doubling to quadrupling the M1 for the iMacs next year.

Doubling up on a 16 Billion transistor part is really the likely limit for next year. Quadrupling it is unlikely in the extreme. Unless they going to separate chips on the iMac next year.

This is a bolder prediction than that, though. They're saying it's a 16 core part with cores disabled. Which is weird, but they are staking themselves out to the idea a 16-core part exists and is intended for the MBP16.

Agree that the ballooning size of the chips in these predictions is pretty sus but it's not impossible. The M1 is only 120mm2 and less than a quarter of that is from the four firestorm cores and their memory. Remove the GPU cores, move to N5P, and you can probably make a 16 firestorm core part that is 150mm2 or less. Yield on this is probably kind of sucky at this size on a new process, but if you're planning on disabling as many as half the CPU cores anyway and your die is mostly comprised of CPU cores maybe that doesn't matter.

Eug · Dec 7, 2020

awesomedeluxe said:
Wow. This is written by Gurman, but it reads like a Macrumors thread. The claim that Apple is making a 16x firestorm core part for laptops is kind of hard to believe on its face.

It does survive a baseline viability test. Clocked at iPhone speeds, 16 perf cores would use about 40W. Clocked at Mac Mini speeds, 32 GPU cores would also use about 40W. I have serious heat/area concerns about putting this in a 16", but total power consumption is lower than the ~90W of the current Intel+AMD solution.

If I had to think about how this chip actually looks... I guess there's no IGP? If "unified memory" is still a thing, I guess it's M1X - 4x LPDDR5 - GPU?

This is a bolder prediction than that, though. They're saying it's a 16 core part with cores disabled. Which is weird, but they are staking themselves out to the idea a 16-core part exists and is intended for the MBP16.

Agree that the ballooning size of the chips in these predictions is pretty sus but it's not impossible. The M1 is only 120mm2 and less than a quarter of that is from the four firestorm cores and their memory. Remove the GPU cores, move to N5P, and you can probably make a 16 firestorm core part that is 150mm2 or less. Yield on this is probably kind of sucky at this size on a new process, but if you're planning on disabling as many as half the CPU cores anyway and your die is mostly comprised of CPU cores maybe that doesn't matter.

Well, 16+4 would be fine in an iMac. The other thing is that you're assuming a 16+4 part would be running at the same clock speed as their 4+4 part.

Discussion Apple Silicon SoC thread

Lifer

Platinum Member

Platinum Member

Platinum Member

Diamond Member

Platinum Member

Diamond Member

Lifer

Diamond Member

Diamond Member

Diamond Member

Golden Member

Golden Member

Diamond Member

Diamond Member

Lifer

Diamond Member

Lifer

Diamond Member

Lifer

Diamond Member

Lifer

Golden Member

Golden Member

Member

Lifer