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

 

[gdansk]

I suppose I should trade in my 2019 MBP 16 even if I hate the notch. I'm actually amazed it only has "11 hours" web battery life. But I suppose that's the display's fault? Or maybe the SoC is that much worse at idle low usage than M1 due to fewer efficiency cores.

 

[StinkyPinky]

Since you can max out the chip on the 14", I'm thinking the cooling requirements for it must be pretty modest.

 

[eek2121]

I suspect it may not be quite that fast, but even 3070 level performance mobile would be amazing. If only Microsoft could include ARM support in Windows 11 I'd straight up sell my desktop since it would be strong enough at gaming to not need one.

Ian said on Twitter earlier the the Max variant was 10.4 TFLOPs, which is around 3060 levels of performance.

 

[Gideon]

Really impressive what they managed to pull off! The price is just high enough that I'll wait for actual reviews and benchmarks though.

I mean it's absolutely undestandable for all they offer , but just tiny a bit on the high side to insta-order. The best bang-for-buck configuration for work costs 3309€ over here (14" version with best M1 Pro SoC and 32GB RAM, 1TB of disk) bumping the ram to 64GB and disk to 2TB (needs M1 Max) is 4400€. That's latter is above my pain threshold, no matter the performance really (even if it's my employers money). The former is about the cost of my previous work laptoop (2018 Coffee lake 15" Macbook pro) which is a bit too new to really justify it (after all I don't need the perf, particularily as I already have a M1 mac mini)

If there is anything to nitpick about this release at oll, it's that they didnt add more cores to the already humongous M1 Max Die.

I mean they already designed a 423,6 mm² GPU (with more transistors than A100 ampere) with a miniscule CPU on it. Adding 2-4 big cores and bumping small core count to 8 would have barely moved the needle on die-size but would be extremely helpful for compiling and other parallel tasks (especially with that much memory bandwidth on the table) Thermals shouldn't be an issue as they can downclock more on all-core-loads.

But I guess we have to wait to the Mac Pro replacements to get crazy core-count CPUs.

 

[Eug]

I am shocked. I never thought they would do the 32core GPU. They did and it has 57 Billion Transistors. Holy shoot. That's gonna be expensive.

So I printed out your screen grab and measured the pictures off the page, and then compared the sizes. (I'm hoping these are provided to scale.)



M1 Pro ~247 mm2
M1 Max ~436 mm2



[EDIT]

The scaling in the picture is off. The shared dimension of M1 Pro and M1 Max are off by 5%, and those are probably slightly out of scale vs M1 too. Real life die sizes are supposedly 241.7 and 383.5 mm2.

https://twitter.com/x/status/1450286632729518089

[/EDIT]

1st Geekbench is lower at 11542 than I was guessing, which was around 13000 for this 10-core variant.

 

[Ajay]

Hmm, still now large iMac (27/30). Would have thought that there could easily be a M1 Pro and a M1 Max version of that. Apple must be really tight on wafer allocation.

 

[eek2121]

Hmm, still now large iMac (27/30). Would have thought that there could easily be a M1 Pro and a M1 Max version of that. Apple must be really tight on wafer allocation.

Likely supply chain issues. Things are a mess right now. I have parts sitting in China with no way to get to the US.

 

[StinkyPinky]

Ian said on Twitter earlier the the Max variant was 10.4 TFLOPs, which is around 3060 levels of performance.

3060 desktop.

 

[biostud]

Apple can destroy x86 business if they are ok selling these chips to other OEMs.

That's not their business plan. They simply want to own the store where all purchases include 30% Apple tax. And the way to achieve this is to monopolize the best hardware and fuse it with their optimized software. Why on earth would they share anything, when the end goal is to get everybody into the Apple store?

 

[Gideon]

1st Geekbench is lower at 11542 than I was guessing, which was around 13000 for this 10-core variant.

The ST score is actually surprisingly lackluster considering it's the essentially the same as my Iphone 13 mini (i was hoping they managed to clock it a bit higher for burst loads as A15 is clocking significantly higher than A14).

MT is impressive for a laptop, but also only about 25% faster than my undoubtably old Ryzen 3700X.

Cosidering I get the same ST score out of a fanless tiny phone, I hoped for a bit more (not too much, but say, even 10-15%)

 

[eek2121]

So I printed out your screen grab and measured the pictures off the page, and then compared the sizes. (I'm hoping these are provided to scale.)

View attachment 51590

M1 Pro ~247 mm2
M1 Max ~436 mm2



1st Geekbench is lower at 11542 than I was guessing, which was around 13000 for this 10-core variant.

View attachment 51591

So if that is indeed a normal score, Apple’s minor perf advantage appears to come entirely from the node. If the chip didn’t have the two efficiency cores, if would have lower multicore performance than Ryzen H)

3060 desktop.

lol no. 3060 mobile.

EDIT: Apple was very misleading with the graphs. Ask @uzzi38 what he thinks about it.

 

[Eug]

So I printed out your screen grab and measured the pictures off the page, and then compared the sizes. (I'm hoping these are provided to scale.)

View attachment 51590

M1 Pro ~247 mm2
M1 Max ~436 mm2

Hmmm.... If M1 Pro is really just a chopped M1 Max, then these images are not to scale. In my pic, M1 Pro is 19 mm wide but M1 Max is 20 mm wide. So they're off by at least 5% per dimension. But nonetheless, Max is ginormous!

CPU/GPU upgrade pricing.

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

I was wondering about the GPU binning (postulating 14-core and 28-core variants) but I didn't think they were going to do CPU core binning as well. Now we just have to find out about the clock speeds too, but no mention about that was made.

Anyhow, my wallet is safe for now, since the updated Mac mini did not materialize. I will soldier on with my Mac Pro 3 GHz 8-core Xeon X5365 for now, for my secondary computer. (Its speed is mostly fine for what I use it for, but it's a Frankenmachine with some idiosyncrasies, it sucks power and is sometimes loud, and it's stuck on OS X 10.11 El Capitan.)

If Apple releases a new Mac mini with new case and lots of ports, then I may get that, with M1 Pro 8-core CPU (6+2) and 14-core GPU. I'd take an M2 Mac mini as well, but I suspect that would get crippled in terms of available ports. Ironically though, if I do get an M1 Pro Mac mini, it would be way, way, way faster than my main machine, which is a Core i5-7600 iMac.

 

[Ajay]

That's not their business plan. They simply want to own the store where all purchases include 30% Apple tax. And the way to achieve this is to monopolize the best hardware and fuse it with their optimized software. Why on earth would they share anything, when the end goal is to get everybody into the Apple store?

And subscribing to Apple Services (Apple Music, Apple TV, Apple+ for the bundle). The Apple 'tax' of 30% seemed reasonable in the beginning to help build all that infrastructure, but now seems too high; that said, I do appreciate using the Apple store an am not a fan of needing to buy my software/subscription on the web instead (more passwords, more companies with my credit info).

 

[StinkyPinky]

So if that is indeed a normal score, Apple’s minor perf advantage appears to come entirely from the node. If the chip didn’t have the two efficiency cores, if would have lower multicore performance than Ryzen H)



lol no. 3060 mobile.

EDIT: Apple was very misleading with the graphs. Ask @uzzi38 what he thinks about it.

What am I looking at here? That is clearly inferior (on paper, maybe not in the real world).

And I'd expect them to be misleading, I wonder when the reviews drop

 

[Abwx]

Interestingly they state that CPU TDP is 30W and 55W for the GPU.

Claimed efficency seems to me a bit optimistic, we ll see how it does in real benches, though.

MacBook Pro mit M1 Pro/Max: Apple bringt extreme Leistung und mehr Anschlüsse zurück

Im MacBook Pro 14,2" und 16,2" feiern Apples M1 Pro und M1 Max mit deutlich mehr Leistung ihre Premiere. Vorbestellt werden kann ab sofort.

www.computerbase.de

 

[eek2121]

Interestingly they state that CPU TDP is 30W and 55W for the GPU.

Claimed efficency seems to me a bit optimistic, we ll see how it does in real benches, though.

That GPU perf vs. power graph is complete marketing nonsense. Read the fine print, they didn’t compare it with the 3070, they compared it with Intel’s Integrated graphics! Talk about lipstick on a pig.

EDIT:

 

[Ajay]

Interestingly they state that CPU TDP is 30W and 55W for the GPU.

Claimed efficiency seems to me a bit optimistic, we ll see how it does in real benches, though.

MacBook Pro mit M1 Pro/Max: Apple bringt extreme Leistung und mehr Anschlüsse zurück

Im MacBook Pro 14,2" und 16,2" feiern Apples M1 Pro und M1 Max mit deutlich mehr Leistung ihre Premiere. Vorbestellt werden kann ab sofort.

www.computerbase.de

Apple, as usual, is being really fuzzy about this (like the super tiny text I can't read). Given their track record, I'd give them the benefit of the doubt, but yes, the proof will be found in detailed reviews.

 

[jeanlain]

That GPU perf vs. power graph is complete marketing nonsense. Read the fine print, they didn’t compare it with the 3070, they compared it with Intel’s Integrated graphics! Talk about lipstick on a pig.

?
They compared to many laptops, including one with a 3080 mobile.

 

[jeanlain]

Apple, as usual, is being really fuzzy about this (like the super tiny text I can't read)

These are screenshot from the video stream I think. You can read the notes from Apple's websites.

 

[scannall]

Apple, as usual, is being really fuzzy about this (like the super tiny text I can't read). Given their track record, I'd give them the benefit of the doubt, but yes, the proof will be found in detailed reviews.

Like any product regardless who makes it, waiting for reviews of the actual product is a good idea. I am excited to see what the actual reviewed performance is. It should be spectacular.

 

[eek2121]

Apple, as usual, is being really fuzzy about this (like the super tiny text I can't read). Given their track record, I'd give them the benefit of the doubt, but yes, the proof will be found in detailed reviews.

I pasted the text above. IMO it is ridiculous.

 

[eek2121]

?
They compared to many laptops, including one with a 3080 mobile.

They supposedly compared efficiency to a 3080 mobile. However, the 3080 was *on battery* (which throttles when on battery to conserve battery life) to the M1 Pro Max (which does not throttle when unplugged). Why not compare them both plugged in and compare perf/watt? The graph, despite claiming relative performance, is marketing fluff. I bet we will see around 3060 levels of performance at 25%-30% less power.

Don’t get me wrong, decent for a first attempt at a GPU, but Apple botched it with absurd claims and marketing fluff.

 

[jeanlain]

You haven't watched the video stream. I suggest you watch it. On battery, the M1 MAX is 3.3x faster than the PC with 3080 mobile. The grey curves shown above represent PC laptops plugged in. For the Macs, being plugged in or not makes no difference.

 

[Schmide]

Looks like the RAM, IO and SLC are off the GPU and the CPU L2 feeds off that.

Considering the new trend in HPC is to run the slingshots off the GPUs, it could be?

It's symmetry is beautiful.

Edit: although the slight skew of the 8 performance cores is somewhat triggering.

 

[eek2121]

You haven't watched the video stream. I suggest you watch it. On battery, the M1 MAX is 3.3x faster than the PC with 3080 mobile. The grey curves shown above represent PC laptops plugged in. For the Macs, being plugged in or not makes no difference.

It baffles me that you don't understand this. The test is completely irrelevant UNLESS the laptop is plugged in. With my RTX 3070 mobile, clocks drop from over 1100 MHz to 200 Mhz when the plug is removed. This isn't because the card can't run off the battery, it's to keep the battery from dying in 5 minutes. The test was completely nonsensical. The only valid way to run this test would be with both laptops plugged in.

EDIT: Also when you unplug, the GDDR6 downclocks as well.