Discussion Apple Silicon SoC thread

Page 70 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Eug

Lifer
Mar 11, 2000
23,825
1,396
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:


M4 Family discussion here:

 
Last edited:

Mopetar

Diamond Member
Jan 31, 2011
8,113
6,768
136
Yeah I'm still putting my money the Mac Pro using the chips that go in the lower end Macs able to work alone or as chiplets. You get up to 32 big cores, 4x the GPU, and 4x the memory channels. If they really wanted to go insane they could even release an 8 chip version of that later.

I've been wondering what they'd do for the bigger chip. Adding in some additional cores and possibly targeting a higher clock speed seems more than reasonable. They definitely have a lot of thermal headroom to play with, but the GPU is what gives me the most pause. Apple has never cared overly much about having the most powerful graphics, but it seems like it would be hard to compete with the high-end Nvidia/AMD GPUs that professionals would want to use for their work. I have to think that some of the type end models will still have dedicated graphics.

Note: I've been a big fan of the mini. I had the 2012 quad core i7 model previously. Upgrading to the 2018 model. never gave me a "wow, that was fast moment". It was an incremental speedup. The upgrade to the M1 definitely changed that. There were several times I thought, "wow, that was fast!".

I think that's just general Intel stagnation over that time period. Intel released mainstream 4C/8T parts all the way back in 2008 with Core 2 Quads and Nehalem. It took them a decade until the started releasing 6C/12T consumer parts with Coffee Lake. There were some good IPC gains and the clock speeds that the chips could hit slowly crept up, but if you had a Sandy Bridge CPU you were still generally fine and didn't have a lot of reason to upgrade unless you were a professional user that could get some benefits and didn't already have an HEDT part.
 

Mopetar

Diamond Member
Jan 31, 2011
8,113
6,768
136
Cope. They just have 50% more IPC than Skylake, or so.

What are you even talking about with this "cope" nonsense? Yeah, the M1 has a higher IPC, but the reason it feels like such a big upgrade is that until Zen came out Intel was under no pressure to make big improvements. Even their process tech fell behind.

If Intel were as aggressive with their CPU designs as Apple had been over the years, I'd wager that the M1 Mac Mini would have get more like just a usual upgrade instead of such a big jump.
 

Doug S

Platinum Member
Feb 8, 2020
2,784
4,744
136
N4 will enter risk production in the second half of this year and HVM next year (and I'm pretty sure it will be after Chinese New Year).
Apple does not use risk production.

If N4 is only risk production in H2 of this year there's no way Apple will use it for anything, I thought I had seen something indicating it would enter mass production before the end of this year but that's clearly not the case. In fact, when I googled to check on this I saw an article suggesting N4 wouldn't begin risk production until Q4.

N3 will be entering risk production pretty soon, well before N4. TSMC seems to have a risk production window of about 9-12 months before entering HVM, and TSMC targets a window of Q2 for the start of HVM to match up with the iPhone launch. A tweaked process like N4 would undoubtedly have a shorter risk production window, but if it starts in Q4 it is unlikely to beat N3 to HVM by much so it would seem unlikely Apple would have any reason to choose N4 over N3 if best case they only gain a month or two.
 

Doug S

Platinum Member
Feb 8, 2020
2,784
4,744
136
I've been wondering what they'd do for the bigger chip. Adding in some additional cores and possibly targeting a higher clock speed seems more than reasonable. They definitely have a lot of thermal headroom to play with, but the GPU is what gives me the most pause. Apple has never cared overly much about having the most powerful graphics, but it seems like it would be hard to compete with the high-end Nvidia/AMD GPUs that professionals would want to use for their work. I have to think that some of the type end models will still have dedicated graphics.

I wouldn't be shocked to see the next step up from the M1 (8+4 or whatever it ends up being) having 4x the GPU cores instead of only doubling. If they use a chiplet solution for the Mac Pro they'll need bigger GPU building blocks than 2x M1 performance to have any shot of beating AMD/NVidia GPUs - and yes I do think beating their best mobile GPU will be the goal for the MBP 16" high end SKU and beating their best discrete GPU will be the goal for the Mac Pro (though that will obviously be a LOT harder to accomplish)

From eyeballing the die photos the GPU looks to be around 20% of the M1, let's call it 25 mm^2. If they added 75 mm^2 for quadrupling the CPU, added let's call it 20 mm^2 for four more big cores, and another 10 mm^2 or so for more memory controller/system cache resources. That gets it up in the 230 mm^2 range, or just under twice the size of the M1 die. Unfortunately N5P won't get them any scaling benefit over N5, but at least there's a decent 7% performance or 15% power improvement to play with. This all assumes the GPU cores are roughly the same size in the A15 generation - if they are significantly bigger/better they might not need to go a full 4x.

Yield shouldn't be a concern at that die size, especially since they could use dies with defects in CPU or GPU areas that are still otherwise functional (which should be well over half the defective chips based on area) as 6+4 chips with fewer GPU cores to create a mid level between the M1 and the 8+4.

Plugging a few numbers into a wafer calculator (using 12x20 for a 240 mm^2) shows 242 dies per wafer. If you conservatively assume you get 200 good dies (as either 8+4 or 6+4) then the cost per die based on reported TSMC wafer costs (i.e. not including testing, packaging etc.) is still comfortably under $100. So I don't think that's an unreasonable die size even though it would be over 50% larger than anything Apple has done with previous A-series SoCs.
 

Mopetar

Diamond Member
Jan 31, 2011
8,113
6,768
136
I don't think the number of dies or the cost is a big concern to Apple. Obviously the want to make money as much as the next corporation, but it's a question of performance and what they can achieve with whatever amount of die space they do end up allocating to graphics.

There's been a good deal of comparisons between what their CPU cores are capable of compared to what both AMD and Intel have, but I don't think I've seen nearly as much on what their graphics cores are capable of or how they stack up relative to anything outside of those found in other ARM-based designs that feature in phones or tablets.

Another problem is whether there's native support for their GPU because even if it's really good, if they have to use Rosetta for most applications it's going to hurt performance. Some of the benchmarks results that have been posted showed that you could get roughly 1650 performance out of the M1 GPU in synthetic tests, but once you moved into an actual game and had to rely on translation the M1 was only 60% - 75% of the performance.
 

LightningZ71

Golden Member
Mar 10, 2017
1,798
2,156
136
Well, 60-75% of 1650 performance from an iGPU with rather conservatively specced dual channel ram is nothing to sneeze at. If they effectively doubled the iGPU size and doubled the RAM bandwidth, they would be beyond 1650 super performance. If they quadrupled the iGPU and switched to soldered, "quad" channel DDR5 (for up to 200+GB/sec), which should be available in quantity, they could be approaching 1660 performance or better. I know that that's all very optimistic scaling, but, this is apple...
 

Doug S

Platinum Member
Feb 8, 2020
2,784
4,744
136
I don't think they will worry too much about performance under Rosetta 2. Every application of consequence will be ported by the end of this year. Some old applications that are no longer actively maintained won't be, but an ARM Mac will run them faster than the x86 Macs available at the time of the application's release even if it gets "only" 65-70% of native performance.
 

B-Riz

Golden Member
Feb 15, 2011
1,595
761
136
You expected them to upgrade the entire line to ARM at once, dropping all x86 models the same day? Of course it is bifurcated for a while, there has to be a transition period.

All they've done so far is stick the M1 in entry level systems, they haven't even dropped the "higher end" x86 SKU (higher end in quotes since it is slower) for their cheapest products like the Air and Mini.

They will support larger memory configurations in non-entry level systems that will get whatever follows M1 - we'll probably see something later this year based on the A15 cores. Some will likely still have soldered RAM depending on the model, but they won't be able to use LPDDR4/5 outside the entry level due to its inherent capacity limit.

No, I didn't expect a full upgrade. The thing about this product, is, it is fully integrated, dropping BOM and licensing vs Intel for the Mini. That leads to cheaper production costs and increased profits and it is a cheap way to get hardware in developers hands.

So in 2 years, I'm guessing, after devs have learned and worked with the Mx series, there will be a huge hardware push by Apple.

But I feel that it would be a huge mistake if Apple pushes to end x86 software support after the 2+ years of Mx being out in the wild for consumers.

Personally, I think the M1 Mini base is just a luxury Chromebox right now, no up-gradable-ness, just a disposable item.
 

B-Riz

Golden Member
Feb 15, 2011
1,595
761
136
It was compelling for me. I had a mac mini i5 6 core (2018) with 32GB RAM (installed it myself - it was easy). The M1 with 16GB runs circle around the Intel. General usage is significantly faster (browsing, app launch, scrolling). Photo editing is a lot faster. A large photomerge in photoshop (15 42MP RAW files) was almost 2X faster. Even Lightroom CC classic running under Rosetta was about 20% faster. I also dabble in video editing - but my cursory usage reveals it is at least 3X faster or more.

Note: I've been a big fan of the mini. I had the 2012 quad core i7 model previously. Upgrading to the 2018 model. never gave me a "wow, that was fast moment". It was an incremental speedup. The upgrade to the M1 definitely changed that. There were several times I thought, "wow, that was fast!".

I don't regret the upgrade to M1 at all. Unless you really need to run x86/x64 virtualization - upgrading to an M1 IMO is a no-brainer.

I do think it is really great how a workflow like this is significantly faster on the M1, but I see it as a lot of custom hardware that is accelerating the work. So Apple was able to have this massive performance increase because they designed for it, while the old x86 is just a simple workhorse.

Right now, I am waiting for an Mx product from Apple that I can upgrade memory and storage on.
 
  • Like
Reactions: Burpo

Doug S

Platinum Member
Feb 8, 2020
2,784
4,744
136
But I feel that it would be a huge mistake if Apple pushes to end x86 software support after the 2+ years of Mx being out in the wild for consumers.

Why do you think they'd push to end x86 support after only two years? They support old iPhones a lot longer than that, I imagine they wouldn't support old Macs for a shorter time.
 

jpiniero

Lifer
Oct 1, 2010
15,223
5,768
136
Why do you think they'd push to end x86 support after only two years? They support old iPhones a lot longer than that, I imagine they wouldn't support old Macs for a shorter time.

x86 support isn't going to be dropped for awhile. Especially when it sounds like the Mac Pro is getting an Intel refresh.
 

smalM

Member
Sep 9, 2019
72
77
91
N3 will be entering risk production pretty soon, well before N4. TSMC seems to have a risk production window of about 9-12 months before entering HVM, and TSMC targets a window of Q2 for the start of HVM to match up with the iPhone launch.
N3 will enter HVM in the second half of next year.
 

Mopetar

Diamond Member
Jan 31, 2011
8,113
6,768
136
x86 support isn't going to be dropped for awhile. Especially when it sounds like the Mac Pro is getting an Intel refresh.

Really, that's the first I've heard. Assuming they did want another x86 Mac, which does make a certain amount of sense, why go with Intel? AMD has far and away the better HEDT and sever CPUs right now.

Unless it's contractual obligations, Apple has no reason to stick with Intel.
 
  • Like
Reactions: Tlh97 and MangoX

Doug S

Platinum Member
Feb 8, 2020
2,784
4,744
136
N3 will enter HVM in the second half of next year.

Well there's a difference between "when a wafer enters production" and "when a wafer exits production" - something like two months with modern processes, possibly up to three (I understand it was close to three in 7nm DUV due to all the multiple patterning steps, but hopefully EUV has reduced the number of wafer processing steps somewhat)

N3 will be used for next year's iPhone SoCs, so the first HVM wafers may begin exiting in H2 but they will definitely be entering it in Q2, otherwise TSMC would miss the launch date for their biggest customer's biggest product.
 
Last edited:
  • Like
Reactions: smalM

B-Riz

Golden Member
Feb 15, 2011
1,595
761
136
Really, that's the first I've heard. Assuming they did want another x86 Mac, which does make a certain amount of sense, why go with Intel? AMD has far and away the better HEDT and sever CPUs right now.

Unless it's contractual obligations, Apple has no reason to stick with Intel.

I too wonder why Apple did not go to a Zen2 Threadripper setup; my only guess is that their terms were too in favor for Apple and AMD is happier supplying the Zen lineup to the non-Apple ecosystem.

Also, a whole lot of testing for UNIX certification on AMD.

 

Commodus

Diamond Member
Oct 9, 2004
9,215
6,818
136
Really, that's the first I've heard. Assuming they did want another x86 Mac, which does make a certain amount of sense, why go with Intel? AMD has far and away the better HEDT and sever CPUs right now.

Unless it's contractual obligations, Apple has no reason to stick with Intel.

It sounds like you're thinking hardware production when jpiniero means software support.

In terms of releasing Intel-based hardware? Apple's going to drop x86 like a hot potato. But software will likely last longer. Apple didn't ditch PowerPC support in macOS until Snow Leopard in 2009, three years after Intel Macs swept aside PPC in a matter of months. That's three years of software support after the lineup had switched.

There are no guarantees the same will happen here, but it seems likely that Apple will keep Intel support going until it either hits a technical barrier (to do X, we need to ditch Intel) or it knows the Intel user base is rapidly dwindling.
 

JasonLD

Senior member
Aug 22, 2017
487
447
136
Really, that's the first I've heard. Assuming they did want another x86 Mac, which does make a certain amount of sense, why go with Intel? AMD has far and away the better HEDT and sever CPUs right now.

Unless it's contractual obligations, Apple has no reason to stick with Intel.

2 year transition doesn't mean they will be ending x86 support after 2 years. Just means they won't be doing hardware update with x86.
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,340
5,464
136
Oh it'll still be AMD. AMD might actually be the holdup since I would think the GPU options will include Pro versions of the various Navi chips.

Yes, the Leaks are for a new RDNA2 card, for the Mac Pro. Not a new Mac Pro.

The point is, a new GPU card, is not an Intel Mac Pro refresh.

There is negligible chance they are releasing any new Intel HW, now that their hands are full with developing and release ARM Mac HW on a tight timeline.
 
  • Like
Reactions: Tlh97

smalM

Member
Sep 9, 2019
72
77
91
Well there's a difference between "when a wafer enters production" and "when a wafer exits production" - something like two months with modern processes, possibly up to three (I understand it was close to three in 7nm DUV due to all the multiple patterning steps, but hopefully EUV has reduced the number of wafer processing steps somewhat)

N3 will be used for next year's iPhone SoCs, so the first HVM wafers may begin exiting in H2 but they will definitely be entering it in Q2, otherwise TSMC would miss the launch date for their biggest customer's biggest product.
Good pioints.
But typically TSMC's targets for entering HVM mean exactly that, entering.

N7 has 78 mask layers, N5 59 (14 EUV); N3 is estimated to have 21~24 EUV mask layers and I couldn't find a total number.