Discussion Apple Silicon SoC thread

Page 150 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Eug

Lifer
Mar 11, 2000
23,579
992
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:

fkoehler

Member
Feb 29, 2008
193
145
116
My understanding is that Apple lost a fair number of senior engineers to Nuvia, and others. IIRC, last years and this years releases have been impacted. As you said, if A17 is still 'power optimized' vs actual IPC, then they'll probably be in a rut till 2024-25.

Though it is possible their CPU design team is a mess - if A17 isn't a pretty big jump that's when we might need to worry.
 

Doug S

Platinum Member
Feb 8, 2020
2,191
3,377
136
My understanding is that Apple lost a fair number of senior engineers to Nuvia, and others. IIRC, last years and this years releases have been impacted. As you said, if A17 is still 'power optimized' vs actual IPC, then they'll probably be in a rut till 2024-25.

Engineers come and go from all the big players. Just look at the resumes of any of the big names like Jim Keller and see how many different places they've worked. Sure losing more than a typical number off the top at once will hurt more than usual, but that shouldn't stop anyone in their tracks the way Apple has been the last two years.

People may be underestimating what Apple accomplished with the M1 Ultra, and what they will have accomplished whenever the 4 chip successor going in the Mac Pro is released. Without a single multi chip or multi socket design before, Apple designed something that blows away the competition as far as inter-chip bandwidth. That's a pretty damn good first effort, and took some pretty amazing engineering.

Correct me if I'm wrong but AFAIK they shipped the first multi chip GPU, which may not have made Nvidia quake in their boots but was not too bad for their first effort (and for the fact that all available cross platform benchmarks are heavily tilted in favor of PCs, and were ported to Metal not originally written for Metal, and definitely weren't written for TBDR)

The M1 Max had only one set of I/Os, they need three sets to do Mac Pro meaning the work wasn't done when M1 Max was done. So it isn't like they've been sitting on their hands, they've kept pretty busy. Doing these things the first time is at least an order of magnitude more difficult than iterating on it in subsequent generations - that is the difference between coming up the concept of how to do something no one else has done before vs building upon work that's already been done.

So I know some people have taken the suggestion Apple's designers have been busy elsewhere as making excuses for Apple recycling A14 with few changes (other than added address bits, improvements in cache, and other minor things in A15) for two straight years, but Apple would be stupid if they didn't have all their best people assigned to making the Apple Silicon rollout as successful as possible. You only get one chance to make a good first impression, if you screw it up and people abandon Mac for Windows winning them back would be far more difficult.

Once that Apple Silicon rollout is complete (and I'm guessing it already is, just because the Mac Pro hasn't shipped yet doesn't mean it hasn't been running in Apple's labs and maybe a few customer sites for a while now) you no longer need all your best people on that task, and they can get back to the problem of "what amazing things can we do with the ~70% more transistors we'll have available to us on M3/A17 thanks to N3?"
 

uzzi38

Platinum Member
Oct 16, 2019
2,556
5,531
146
Only as fast as a 5.4 GHz Zen 4 Core?
Apple how dare you!
Uh, no? Zen 4 at 5.7GHz scores >2200, dropping down to 5.4GHz would still net you >2000

It would be around as fast as a Zen 4 core at ~4.6-4.8GHz, by comparison. Which is kinda base clock territory now.

Don't get me wrong, Apple will still have a pretty substantial power efficiency lead at the same power. Like I'd expect that Zen 4 core at those clocks to still pull around 8W, which is much higher than where Apple peaks out at. But it's not fair to say the peak performance is the same.
 
  • Like
Reactions: Tlh97 and gdansk

Eug

Lifer
Mar 11, 2000
23,579
992
126
Multi core:
View attachment 67351

We should see a significant improvement next year with N3E.
Yeah, that 5455 multi-core score makes more sense. That's about 11% faster than A15.

BTW, that MT score - for a damn phone - is 40% faster than my
home desktop Core i5-7600, and well over 3X as fast as my work desktop Core i5-4278U. :p I really hope that M2 / M2 Pro Mac mini comes sooner rather than later.

At these performance levels, an ultraportable laptop with A16 would be great, assuming it could address sufficient RAM (and virtual memory) and had the appropriate I/O, with macOS.
 
Last edited:

richardskrad

Member
Jun 28, 2022
51
44
51
A16
Single-core: 1887
Multi-core: 5455

Holy crap. What a time to be alive. The A16 is a monster. Until the day comes when AMD and Intel cal do this level of performance without the presence of fans, Apple Silicon will have the title as the best chips on the market.

 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Engineers come and go from all the big players. Just look at the resumes of any of the big names like Jim Keller and see how many different places they've worked. Sure losing more than a typical number off the top at once will hurt more than usual, but that shouldn't stop anyone in their tracks the way Apple has been the last two years.

There are good teams and there are legendary ones and the latter is what the Alpha CPU design team within DEC was, which included Keller. Keller was a well known guy but anyone who were following microprocessors at the time knew the Alpha team in whole was stellar in their design and execution.

Little before Intel transition there was a startup named PA Semi, which employed many of those legendary engineers plus from big ones such as Intel. The specs for that chip was amazing too but it never got used. However PA Semi did get bought out by Apple.

I am quite certain it's the PA Semi team that allowed the Apple Ax cores to be good as they are today.

So when you have lead architect and many others leaving it's not a good sign. It's likely not a coincidence the slowdown happened just a few years after the team members left. It takes few years to put a chip from concept to production.

We shouldn't underestimate the impact of excellent engineers. It takes years if not decades of training to reach that point, assuming the person is capable of doing so. Just like other engineering professions(such as architects and ship builders) they skill they have can't easily be replaced by a newcomer fresh out of school.

The project was so good is why even the remnant is excellent. But like I repeated before, it took Intel 2 years before the problems started, and an additional year before it got serious when Brian Kraznich became CEO and started firing many employees. But it wasn't until 5 years later we knew they were in real trouble.
 
Last edited:

poke01

Senior member
Mar 8, 2022
632
601
106
Apple said in the keynote A16 focused on efficiecy, we shall see. 20% lower power on P cores while also increasing clock by 200Mhz is good but lets see we can get close to that claim.

For 3 years Apple compared the A series to the competition or their own chips that are 2 gens behind. When Apple compares their new A series chip to its predecessor that is when IPC increased greatly, hopefully they do that next year.
 

Eug

Lifer
Mar 11, 2000
23,579
992
126
There are good teams and there are legendary ones and the latter is what the Alpha CPU design team within DEC was, which included Keller. Keller was a well known guy but anyone who were following microprocessors at the time knew the Alpha team in whole was stellar in their design and execution.

Little before Intel transition there was a startup named PA Semi, which employed many of those legendary engineers plus from big ones such as Intel. The specs for that chip was amazing too but it never got used. However PA Semi did get bought out by Apple.

I am quite certain it's the PA Semi team that allowed the Apple Ax cores to be good as they are today.

So when you have lead architect and many others leaving it's not a good sign. It's likely not a coincidence the slowdown happened just a few years after the team members left. It takes few years to put a chip from concept to production.
When the PA Semi purchase happened, naive little me wondered if Apple would be doing something with PowerPC again. I'm glad they didn't. :)

I still wonder why so many key people left. Money is always an incentive, but usually it's not just the money.

 

Heartbreaker

Diamond Member
Apr 3, 2006
4,219
5,220
136
When the PA Semi purchase happened, naive little me wondered if Apple would be doing something with PowerPC again. I'm glad they didn't. :)

I still wonder why so many key people left. Money is always an incentive, but usually it's not just the money.


Signing bonus + fresh stock options can be pretty big money though, and those leaving for startups are potentially looking at MANY millions: Ex Nuvia guys...
 
  • Like
Reactions: Mopetar

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,785
136
Apple said in the keynote A16 focused on efficiecy, we shall see. 20% lower power on P cores while also increasing clock by 200Mhz is good but lets see we can get close to that claim.

Typically companies use efficiency when they have less to show for than what they want to. 20% efficiency increase at the CPU level is how much at the platform level? You are buying a phone or a laptop not the CPU so who knows what the actual gain is at the device level. It could be 3-4% or it could be 10%. Whatever it is, it'll be less than what they claim. Also keep in mind we don't know what arbitrary point they chose to compare efficiency. Is that worst case or not?

I still wonder why so many key people left. Money is always an incentive, but usually it's not just the money.

At high levels it's less likely it's about the money and looking for better projects to work on. Possible toxic conditions in the company may play a role as well. There are certainly rumors out there about it since 2017.

These guys are absolutely at the top of the game and very ambitious. If the company they work for won't allow them to fully realize their dreams then no amount of money will convince them.
 

Doug S

Platinum Member
Feb 8, 2020
2,191
3,377
136
Signing bonus + fresh stock options can be pretty big money though, and those leaving for startups are potentially looking at MANY millions: Ex Nuvia guys...


One of the problems of being a company with a market cap of nearly $3 trillion is that offering stock options isn't much of an incentive because the stock price simply can't go up percentage wise like it can at smaller companies.

Everyone in Silicon Valley takes a moonshot at working for a startup at least once during their career, and dreaming it hits a home run and they end up collecting $5 million for 2 years of 80-100 hour weeks.
 

Eug

Lifer
Mar 11, 2000
23,579
992
126
FWIW, Nanoreview.net claims to have an A16 Antutu score.
I don't know from where, so I can't confirm its legitimacy, but here ya go.


Overall A16 score is +7.5% over A15, but that's skewed by the memory score.

A16A15
CPU227104 (+4.8%)216602
GPU349695 (+3.9%)336667
Memory136718 (+21.0%)112950
UX138952 (+5.3%)131916
Overall851991 (+7.5%)792323

EDIT:

Comparing this to the results posted by Antutu itself (see below), I don't think this is legit. It looks like they made some guesses and posted them.
 
Last edited:
  • Like
Reactions: trivik12

Doug S

Platinum Member
Feb 8, 2020
2,191
3,377
136
More Antutu scores. These appear to be legit, posted by Antutu itself.


View attachment 67470


That's a big gain in GPU - I wonder if that's just from LPDDR5 or a bigger GPU clock jump than the CPU got, or if they made some bigger updates to the GPU than they did the CPU? The MEM subtest shows little gain despite LPDDR5. I know zilch about Antutu, which exactly is it testing? Maybe latency is a big component of it, so a 50% bandwidth gain doesn't provide much improvement since the latency would see little or no improvement?
 

mikegg

Golden Member
Jan 30, 2010
1,740
405
136
More Antutu scores. These appear to be legit, posted by Antutu itself.


View attachment 67470

View attachment 67472

View attachment 67471

I expect a bunch of Geekbench leaks to start coming in this week too.
Seems like a pretty good gain across the board - especially if the chip increased in efficiency at the same time. This is year 3 on a 5nm-based node. So it's pretty amazing what Apple has done from the A14 to A16.

People like to write about how iPhone SoC gains have slowed, which is certainly true due to diminishing returns, but these gains are within a 2 year period which is basically one generation of Zen.
 

poke01

Senior member
Mar 8, 2022
632
601
106

mikegg

Golden Member
Jan 30, 2010
1,740
405
136

3NE next year for A17 and M3.
Hopefully, M3 will launch in Spring or Summer 2023. When will N3E be ready?
 

Doug S

Platinum Member
Feb 8, 2020
2,191
3,377
136
Hopefully, M3 will launch in Spring or Summer 2023. When will N3E be ready?


Spring of next year is impossible if M3 is N3E, it would have to be late summer or fall. TSMC's official schedule for N3E would appear to suggest it wouldn't be ready in time for A17, but there have been rumors that it is going better than expected and TSMC might pull it in. The graph TSMC supplied comparing N3E to N5 yield looks good, but you'd expect that a second generation N3 family process should compare well to a first generation N5. I would think N3E vs N5P would be a fairer comparison.

I suppose if the "Pros get the new SoC, non-Pros get last year's SoC" is the new iPhone strategy rather than a one-off for this year they'd buy a bit more wiggle room on when N3E has to be ready to make a September launch.

I would also say it is quite possible Apple doesn't use N3 OR N3E for A17, but their own flavor of N3 that's probably very similar to N3E but perhaps not identical. Apple has used (at least as far as we know) the standard TSMC process in the past because their release schedule was timed exactly with the release of a new process. But with TSMC launching N3 "late" (and it appears N2 is on the same timing) maybe Apple's plan is to instead use their own tweaked version. AMD is not using a bog standard TSMC process, why should we assume Apple must?