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

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

 

[Eug]

I've been trying to figure out why macOS "feels" faster on M series than on Intel Macs, esp. with limited RAM, since it couldn't just be due to the raw CPU grunt power and SSD speed.

I came across these posts, that I had somehow missed last year.

https://twitter.com/x/status/1326238785181376512

Daring Fireball: The M1 Macs

Native code running on Apple Silicon is not 5 times faster than on Intel, generally, nor is Intel software running under Rosetta on Apple Silicon twice as fast as on Intel. But retaining and releasing NSObjects is so common on MacOS (and iOS), that making it 5 times faster on Apple Silicon than on Intel has profound implications on everything from performance to battery life. It’s almost silly how much faster this is natively, and quite remarkable that it’s twice as fast even translated through Rosetta.

Broadly speaking, this is a significant reason why M1 Macs are more efficient with less RAM than Intel Macs. It’s the combination of software and hardware designed together. You don’t have to take my word for this — this, in a nutshell, helps explain why iPhones run rings around even flagship Android phones, even though iPhones have significantly less RAM. iOS software uses reference counting for memory management, running on silicon optimized to make reference counting as efficient as possible; Android software uses garbage collection for memory management, a technique that, whatever you think of it philosophically, requires more RAM to achieve equivalent performance.

Further, I/O is faster on the M1 than on Intel-based Macs. So when you do run out of free memory and MacOS needs to use swap (virtual memory), it’s faster than ever. It’s not magic, of course. If you really need more than 16 GB of RAM, you need more than 16 GB. But faking it is a lot more fun than it used to be. If you’re using more than 16 GB of memory because you have a lot of small memory allocations that add up to more than 16 GB — like say, a slew of open browser tabs, a dozen productivity apps like Mail and Notes and Photos and so forth, a big dumb sloppy Electron app like Slack, and even something like Xcode, developing a typical application — you really might not notice any hitches with just 16 GB of RAM in these Macs.

---

BTW, this 12" Core m3 MacBook as a desktop stand-in is working great... until I run out of memory. I occasionally get the beachball while multitasking. This is not heavy multitasking, just moderate business type usage, with Safari and Chrome open with multiple tabs, Citrix Workspace, Messages, and perhaps Word, Calendar, Contacts, Mail, and Activity Monitor. Nonetheless despite having 16 GB RAM, I am sometimes using a fair bit of compressed memory, and then occasionally it hits the swap to an extent too. Presumably that's when I'm seeing the beachball. Looking through the list of applications with the most memory usage in Activity Monitor, there are no surprises, with no obvious indications of individual apps with big memory leaks. Just a lot of instances of browser processes, each using several hundred MB of memory.

Back in 2017 I had predicted that by 2022 I'd need about 12 GB RAM to give me breathing room for this type of usage. Apparently that isn't the case, since even 16 GB is becoming a noticeable limitation on Intel for this.

OS is Monterey 12 BTW.

 

[Doug S]

I think that apple is way too conservative with having no turbo at all. Something like Intel PL1 is still very useful, even though it should be way less aggressive with voltage and probably shorter. I still much prefer their take to the increasingly more aggressive turbos Intel (and even AMD more and more) have in their 15W let alone 25-45W laptops.

They've never had to think about it before, as such a thing is basically useless on a phone or even a tablet. In some sense you could think of mobile SoCs (especially the GPU) as being in "turbo" mode at their default clock, and the reduced clock rate they're forced into when running too long as their standard speed.

I don't know all the differences in process characterization between mobile and HPC cells for something like TSMC's N5, and whether there would be a useful amount of performance increase to be had if e.g. you were able to feed 2-4x the "default" power like Intel's PL1 settings allow (i.e. 65W "TDP" vs 241W PL1)

I think the most underrated reason Apple dropped x86 CPUs is power, even though everyone talks primarily about performance. Yeah, they needed to assure customers they aren't giving up performance by switching, but they waited to switch until they were able to do that without compromising their power efficiency goals. So even if Apple could add a turbo that increased performance by 10-20% at the cost of 4x more power, I don't think they'd want to. Either it would last such a short time it is hardly worth the bother, or it would result in hot and loud laptops that they don't want to sell.

For something like the Mac Pro where it isn't running on battery and isn't such a confined space that restricts the ability to move heat maybe it makes sense. It would probably still depend on how amenable a mobile process is to this, as they are unlikely to do a new floorplan and mask set to use an HPC process for just the Mac Pro SoCs. But it would be fun to see what it could do if they did!

 

[Gideon]

So even if Apple could add a turbo that increased performance by 10-20% at the cost of 4x more power, I don't think they'd want to

I agree with you post overall, but this part might not be quite as clear cut.

When I accidentally ran my CPU at a fixed clock, due to an accidental BIOS change, the only place I really noticed it (in day-to-day use) was during the startup of different applications. It just wasn't as snappy as it should have been, otherwise the system seemed to work normally.

Boosting for a few seconds for brief lightly-threaded workloads (such as opening apps) might actually make the device feel snappier.

Power draw for such short burst shouldn't really matter (remember, Apple can choose how often they allow it). I do agree that the mobile cells might make it impractical though.

 

[Ajay]

Anymore rumors about upcoming large screen iMacs? Obviously next year now, along with the upgrade mac mini. Just curious if anyone has weaseled out more info from the tight-lipped Apple and its suppliers.

 

[Eug]

Anymore rumors about upcoming large screen iMacs? Obviously next year now, along with the upgrade mac mini. Just curious if anyone has weaseled out more info from the tight-lipped Apple and its suppliers.

Leakers are claiming 2022 H1, and 27" (not bigger).

 

[Ajay]

Leakers are claiming 2022 H1, and 27" (not bigger).

Same CPUs as the MacBook Pros, I assume?

 

[Eug]

Same CPUs as the MacBook Pros, I assume?

Yes, that is the claim.

Some have claimed Apple may use the moniker "iMac Pro", but I suspect that could just be conjecture.

P.S. I caved, and ordered a Mac mini...

...but it's just a used 2014 model to tide me over until next year. Hopefully it works fine when it arrives.

At this point I'm still waffling between a hypothetical M2 Mac mini vs. a hypothetical M1 Pro (8-core) Mac mini. I suspect the M1 Pro will come first, at around the same time as the iMac in 2022 H1, but either way if I have a functional 2014 Mac mini, I may just wait until the summer anyway to make use of Apple's education promotion. What I'm afraid of though is the M2 Mac mini not showing up until 1 year from now. By that time, the edu promotion will be over.

Ironically, for the Mac mini I'm supposed to get, I'll be putting an M2 SSD in it, and the computer I'm using in the interim until that arrives is an m3 MacBook.

 

[Mopetar]

You don't happen to own a BMW M4 do you?

 

[ashFTW]

X5 M

 

[Mopetar]

Somehow in this crazy world of shortages, shipping delays, and other logistics issues the Macbook Pro I ordered shipped about 3 weeks sooner than was originally scheduled.

 

[gdansk]

Somehow in this crazy world of shortages, shipping delays, and other logistics issues the Macbook Pro I ordered shipped about 3 weeks sooner than was originally scheduled.

What luck. 14 or 16?
My 16" is still 3 weeks out

 

[Mopetar]

What luck. 14 or 16?
My 16" is still 3 weeks out

16" with M1 Max. I've mainly been getting stuff setup all day (mainly issues with some applications that don't have a release compiled for ARM yet) so I haven't really spent much time running anything to put it through it's paces, but it does feel good to use.

Obvious the keyboard isn't as good as a mechanical, but if you're coming from one of the laptops with the really thin keys, this feels a lot better to type on. I'm one of those people who thought the touch bar was interesting and honestly I think they could have kept it in addition to bringing back the physical function keys, but it doesn't seem like something that ever good a lot of use so I can see why they got rid of it. It's also nice to have MagSafe back.

The notch hasn't bugged me as much as I suspected that it might at first, but I haven't been running a lot of applications in full screen and it's not obtrusive as long as it's tucked away in the menu bar. It seems like even when you do go to fullscreen, that part of the screen is still treated as a menu to avoid having any of the app obscured. Unless the brightness is really cranked up on the screen, it almost gives the illusion that the blacked out part of the screen isn't part of the actual display, but it's slightly noticeable depending on the lighting in the room or the display settings.

So far I'm really enjoying it. I don't think the fan has kicked in once yet, but then again I've mainly been installing and configuring software so far, but I'll try a few things later to see how well it handles a typical workload. Hopefully you'll luck out like I did and get yours early, because it does feel like the best MacBook Pro since the old 17" unibody models when they first had quad-cores in terms of how much of a lead forward it was in terms of design and performance.

 

[Jwilliams01207]

I think Apple marketing screwed this one up if M1 is as fast as most of us think it is. The performance claims look like the typical BS and by now a lot of people know the pattern and have it set to ignore. To the extent M1 is impressive, they should have actually shown it to be so.

 

[Eug]

The first of the Linus Tech Tips Apple Silicon MacBook Pro reviews is up now.

They only have the 8-core Pro and 10-core Pro for now, since they haven't received all their Maxes yet. They claim they spent $30K in MacBook Pros.

As expected, the Pros did very well in most non-gaming tests. And it just annihilated the competition when it came to battery life. You'll see that the graph isn't just measured in hours now. It has day in the x-axis.



However, for some reason with hardware encoding turned on in Video Toolbox, Handbrake encoding was abysmal. Lower is better.

 

[jeanlain]

However, for some reason with hardware encoding turned on in Video Toolbox,

There must be a bug somewhere with handbrake or video toolbox on the M1 Pro. Maybe it fell back to software encoding.

As expected, performance in games was poor when they used Rosetta games and CPU-bound situations. CS:GO is an openGL game, which adds further CPU overhead for the translation from openGL to Metal.
I'm not sure why Civ VI performs so poorly. It cannot just be Rosetta. I suppose the game is not well optimised for macOS to begin with.
In fact, the performance delta between PC and M1 Macs probably correlates with the performance delta between macOS and bootcamp on an intel Mac.

 

[moinmoin]

The progress report writers of the Dolphin emulator looked at M1 Max. Since it's primarily ST limited not such huge difference to M1:



"As an application that depends on single thread CPU performance, the CPU power available to Dolphin is basically the same for both SoCs and the performance is similar. The improvements we're seeing here are from removing every bottleneck that constrained the M1, thanks to the M1 Max's cache, memory bandwidth, and significant GPU improvements."

"What isn't shown on that chart is the absolutely monsterous GPU the M1 Max delivers. Native res for the display (5x native for the 14in) is trivial, and many games can run fullspeed at 8k (12x native) in our tests. The original M1 choked on 4x native, hence picking 3x for the original testing."



"Now that we have direct access to M1 hardware, we took the opportunity to revisit our framepacing tests that previously revealed framepacing issues on Macs with Intel graphics, and confirmed that it is not an issue on M1. Macs with Intel graphics STILL suffer from this problem."

 

[tamz_msc]

Benchmarking the Apple M1 Max

Understanding the Hardware Capabilities of Apple's flagship SOC

tlkh.dev

Numpy, general Python and ML benchmarks for the M1 Max.

 

[DrMrLordX]

@tamz_msc Good find, and the methodology looks solid too.

 

[Doug S]

I'd been waiting for someone to test the memory bandwidth the GPU could grab, since it had been observed the CPUs could get "only" 200 GB/sec, and most of that in just a couple cores. Some had suggested that the bandwidth was dedicated to different uses, but the tests showing the GPU was able to grab 330 GB/sec demonstrates that's not the case.

So the limitation on the CPU is probably architectural, and will likely be addressed with the M2 generation (which may get a further memory bandwidth boost from LPDDR5X)

 

[DeviantBunny]

Does anyone know of an analysis that investigated the impact of the relatively large M1 chips' L1 instruction cache on the performance of object oriented code?

 

[vinhom]

Would it be possible to keep the base memory capacity at 8GB using ddr5 memory?

 

[Heartbreaker]

Optimum Tech compares the M1 Max MBP to his 5950x/RTX 3090 desktop:

 

[Oyeve]

My company's new M1 iMacs barely run faster than the 10+ year old iMacs they replaced. Barely. Creative Suite is only a bit faster.

 

[Mopetar]

Are you running native binaries? Seems like Rosetta has big performance hit in some cases:

Source: https://petapixel.com/2021/03/10/real-world-test-photoshop-for-apple-silicon-is-a-game-changer/

Even still, something is going very wrong if it's slower than 10-year old iMac hardware. Those computers would be running Haswell-era CPUs or perhaps even prior i5/i7 models.

 

[gdansk]

I know this is anecdotal but my new M1 Pro MBP16 crashes due to out of memory more than my old 9750H MBP16. Both on Monterey, both 16GB. Maybe some memory leak in ARM versions?

But otherwise it seems generally faster and quieter. Work test suite completes a bit faster. And the battery life is much longer.