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

 

[Eug]

IIRC, that was shown as a (now fixed) reporting bug.

My view on this is: If this is a money making machine, then spend money boosting RAM to an excess even if the benefit is imperceptibly small.

But if this just a home/hobby computer, and makes no perceptible performance difference while swapping, then you don't "need" more RAM.

OTOH if your Windows machine starts stuttering while swapping then you "need" more RAM.

So I view the needs less RAM "myth" as not that far off the mark.

M1 Macs don't "use" less RAM, but when it runs out, fast swapping makes the consequences largely irrelevant, so you don't really "need" it.

However the other problem reported is that some users had been getting occasional random crashes on 8 GB machines when pushed. These crashes disappeared after they upgraded to 16 GB.

Many of these reports were not made until later on, after the initial reviews had already come out. That makes sense since people got a better feel for their systems after weeks of real world usage, and started noticing occasional random things like this.

I’m guessing the same may prove true for 16 GB users who should be on 32 GB.

 

[Eug]

Product quality wise? Looks so. Does Intel need to worry? Unless Apple suddenly goes for the lower budget mass market as well, not really.


Isn't that the issue that first came up with the original M1 launch where people first noticed these aggressive swap usages? I thought the topic was essentially resolved by the mention that usage of the SSD makes sense for improving performance and the wear it introduces is still perfectly well within the SSD's life expectancy?

1. See my post above about crashes.

2. It depends on just how many writes you’re actually doing. If you’re just very occasionally swapping heavily then I wouldn’t worry about. However, if this is a daily driver and you’re constantly on yellow memory pressure, I don’t think that’s a good idea.

3. Memory requirements will increase with time. This is true for both the the OS and with software applications, so you're already near capacity now, it's going to be an even bigger problem later, that is if you plan on keeping your machine a long time.

 

[Heartbreaker]

Isn't that the issue that first came up with the original M1 launch where people first noticed these aggressive swap usages? I thought the topic was essentially resolved by the mention that usage of the SSD makes sense for improving performance and the wear it introduces is still perfectly well within the SSD's life expectancy?

It was a bug, fixed in 11.4 update:

Report: macOS 11.4 update fixes M1 Mac SSD wear reporting issue

A data reporting error made it seems like M1 Macs were performing too many writes on the SSD.

www.macworld.com

 

[Heartbreaker]

However the other problem reported is that some users had been getting occasional random crashes on 8 GB machines when pushed. These crashes disappeared after they upgraded to 16 GB.

If you are crashing because you are using swap, that's a bug that needs to be fixed.

I'm on an ancient 6 GB machine. I live in swap, yet my PC doesn't crash.

 

[jamescox]

IIRC, that was shown as a (now fixed) reporting bug.

My view on this is: If this is a money making machine, then spend money boosting RAM to an excess even if the benefit is imperceptibly small.

But if this just a home/hobby computer, and makes no perceptible performance difference while swapping, then you don't "need" more RAM.

OTOH if your Windows machine starts stuttering while swapping then you "need" more RAM.

So I view the needs less RAM "myth" as not that far off the mark.

M1 Macs don't "use" less RAM, but when it runs out, fast swapping makes the consequences largely irrelevant, so you don't really "need" it.

If you are swapping a lot then there is almost no way that can happen without causing increased wear on the SSD. I probably would not purchase a laptop now with less than 32 GB expandability. What I would like is a powerful APU with a stack or two of HBM memory as HBM cache integrated. That should then be backed up by some LPDDR5, preferably as standard DIMMs. Even 8 or 16 GB of HBM cache backed up by some LPDDR5 would probably turn out exceptional performance and power consumption.

 

[IntelUser2000]

If you are crashing because you are using swap, that's a bug that needs to be fixed.

I'm on an ancient 6 GB machine. I live in swap, yet my PC doesn't crash.

LOL! My father uses a Windows 10 machine with 2GB of RAM. Doesn't crash either.

 

[Heartbreaker]

If you are swapping a lot then there is almost no way that can happen without causing increased wear on the SSD.

Obviously if you are swapping a lot, then you use cause more SSD wear than not swapping.

But that doesn't mean you are on a trajectory to wear out your SSD in a couple of years. There was a reporting bug making it look like excess wear was happening, that would lead to early wear out.

It's very unlikely home users are going to wear out a high quality Apple SSD on a MBP, and I already said for a money making machine you go for the extra RAM.

 

[IntelUser2000]

AMD and intel make server CPUs and scale them down with all the downside. A server CPU doesn't need all these accelerators for encode/decode, dsps etc. hence they are missing and hence no support for them really in Windows software. Other thing is legacy compatibility. Why x86 is so cool. But why it is also being held back. Maybe this can trigger of both AMD and intel dropping at least the biggest baggage mostly hindering their designs.

It's not just x86. They are kicking ass in GPUs as well.

 

[jamescox]

. . . seriously?

Until IceLake, all AVX-512 implementations featured lower multipliers that would be inflicted on every core of a CPU even if only one core were handling AVX-512 instructions. Anyone using AVX-512 on a Skylake-SP CPU (for example) on a single core would nuke the performance on every core, even when the CPU was hosting multiple VMs on different cores. Very annoying! I think Intel managed to fix that on IceLake/RocketLake and TigerLake but i'm not 100% sure! Some lesser variant of that bogeyman may still exist.

The amount of re-engineering necessary to accommodate AVX-512 lead to massive amounts of die area being allocated just to AVX-512 support. The caches all had to be realigned for AVX-512 support. The less-often you used it, the more it would negatively affect the overall design of the chip.

And yes, the points @moinmoin brought up were also valid.

I don't want to thread derail, but you did mention AVX-512 complaints, so there they are. AVX-512 has been a mess. Something like SVE2 will work out much better. Hopefully Apple will choose to embrace SVE2 eventually.



Like I said, it's great until it isn't. One switch in algo and then you can't use it anymore. It's planned obsolescence unless you think that people will stay with the same codec forever. Let's face it, the dedicated hardware will always be faster and more power-efficient where applicable.



Please link those videos, or better yet text articles with that data?

How is that different from any other hardware? I have used a lot of really old hardware and it all suffers from not having hardware video codec support for the latest and greatest. If it is a desktop, you can add a more recent video card which might have support for more recent codecs. With a laptop, you mostly need a new laptop.

 

[Eug]

If you are crashing because you are using swap, that's a bug that needs to be fixed.

Perhaps a bug, but nonetheless, the result is a crash.

Here's an unrelated example, but I think it illustrates a point. I used to make a lot of PowerPoint presentations in Office 2016. Some of those presentations included 100+ relatively high resolution images. These individual presentations would take up several hundred MB worth of memory, but I've found that at times the memory usage would jump to several GB worth of RAM over time. If I had insufficient RAM, PowerPoint would eventually crash. However, on my 16 GB MacBook or 24 GB iMac, such crashes are much less common than on 8 GB machines (although they still do occasionally occur).

I'm convinced it's a bug somewhere in Office - memory leak? However, I haven't wanted to upgrade to Office 365 and pay a subscription hoping the bug has been addressed, since I have a perpetual licence for Office 2016 (and because I don't do this much anymore after switching jobs). BTW, interestingly, I don't have this problem in Office 2011... and Office 2011 is sooo much faster. I don't know how they managed to slow down Office so much with the exact same files, during that time.

It's very unlikely home users are going to wear out a high quality Apple SSD on a MBP.

For the most part, that is likely true.

However, and granted this is the minority, I'm talking about people on forums telling creatives they don't need as much memory because M1 is magic. Well perhaps from a performance perspective it is, relative to what we are used to, but the truth of the matter is that M1 machines seem to swap just as much as Intel machines, but also seem to have a less perceptible performance impact from that swap.

Luckily most creatives are smarter than that. But then we have the opposite issue... Some people are unnecessarily getting 64 GB just because.

 

[Heartbreaker]

But then we have the opposite issue... Some people are unnecessarily getting 64 GB just because.

It seems like you are encouraging this issue, with this talk about needing more memory if you happen to tip into swap.

Pros already know what they need. It's the home users that will read statements like this and think they need more memory than they really do.

You can run on swap all the time without it being much of an issue at all.

 

[Gideon]

M1 Macs don't "use" less RAM, but when it runs out, fast swapping makes the consequences largely irrelevant, so you don't really "need" it.

I agree as long as we are talking about regular Joes, whose workloads consist of browsing, office, streaming media, etc.

The only time I would recommend them going fro a 16GB M1 mac (non-pro) is when they plan to keep the machine for 5+ years.

But for professionals it's not that clear cut.

First, it only swaps up to a little less than 2x the memory you have, then the following dialog appears and you're forced to close apps:

Second, I actually have an M1 mac-mini as my work machine and I can instantly tell when it starts swapping (IDE search and autocomplete slows down a lot). It's still usable but definitely not snappy. I then usually have to close all non-essential programs, restart the browsers and IDE. This gets me below the "brown" memory pressure threshold, but it's far from ideal.

This is the only real complaint I have about my work machine, which the eventual 14" Macbook pro should solve.

 

[DrMrLordX]

How is that different from any other hardware? I have used a lot of really old hardware and it all suffers from not having hardware video codec support for the latest and greatest. If it is a desktop, you can add a more recent video card which might have support for more recent codecs. With a laptop, you mostly need a new laptop.

It isn't any different. If you rely on your video card or something like quicksync for encode, then you are going to be in for a rude awakening if you rely on an unsupported codec, or even an unsupported resolution/quality level. There are reasons why the PC community for years has benchmarked encoding software with settings that (usually) take hw-based encode out of the equation.

 

[Geegeeoh]

How is this hardware encoder in terms of quality/file size?

Usually I encode with CPU because hardware accelerated, while faster, do use a LOT more space.

 

[naukkis]

If you are swapping a lot then there is almost no way that can happen without causing increased wear on the SSD.

Swapping to SSD actually isn't increasing it's wearing much. Reading from SSD isn't harmful, and if swapping algorithm is at least somehow decent dirty pages won't get swapped out. They could even implement things like USB-stick boost as swap read speed is what is mostly needed and simultaneously writing swapped data to hard disk won't slow things much at all.

 

[Mopetar]

LOL! My father uses a Windows 10 machine with 2GB of RAM. Doesn't crash either.

Can that even run minesweeper without swapping?

 

[BorisTheBlade82]

My track record? I wasn’t the guy claiming Apple’s hardware was faster than a 3080 or 3090. I wasn’t the guy claiming the M1
Pro/Max was 500% more efficient than AMD/Intel.

I wanted to just ignore these comments but honestly, I can't. I stated several times that this applies to M1 (read: not M1 Pro, not M1 Max) in single threaded workloads. I have numbers to prove that. You and your friends in your reality distortion field have so far failed to come up with any numbers to prove me wrong. All you lot are doing is senseless and factless bashing.

Even when Andrei is writing exactly the same in his M1 Max article you guys just do not seem to be able to realize. So just go and read that article. And if you already read it, go read it again. And if you still do not accept reality I suggest you start working on your reading comprehension as there seems to be a severe lack. Just as an example: He is using words like "absurd" when comparing Apple's competitive advantage for a reason. Or is he also a member of the so called hype train? Are you really suggesting that your words count more than his? If so then that is another proof of you disregarding reality.

 

[jamescox]

It isn't any different. If you rely on your video card or something like quicksync for encode, then you are going to be in for a rude awakening if you rely on an unsupported codec, or even an unsupported resolution/quality level. There are reasons why the PC community for years has benchmarked encoding software with settings that (usually) take hw-based encode out of the equation.

I still don’t understand what your argument against the MacBook Pro hardware acceleration is about. Either people will use the hardware acceleration or not. It is probably an exceptional value for those that do. All of the benchmarks have their place, but for those that are going to use the hardware acceleration, the non-accelerated performance is irrelevant.

Let’s say you use the hardware accelerated codec for 5 years and then something else comes along or you need to switch editing software or something. You seem to be saying that some other hardware is more future proof (I guess?) because it performs slightly better when no hardware acceleration is used. I have been there with a lot of old computers and I would say it doesn’t matter. Based on past experience, 5 years from now, a high end MacBook Pro or a high end Windows solution will be roughly in the same class without hardware acceleration. That class is “really slow”compared to a new system with hardware acceleration and significantly faster general hardware.

 

[Doug S]

People seem strangely focused on encoding performance here for reasons I cannot understand. Unless you are spending hours running encodes every day (in which case you have a preferred workflow and can just benchmark that instead of worrying about how it performs with software you don't use) who cares how fast it is? It will fall into one of three classes - "cup of coffee", "lunch break" or "overnight". Unless the performance difference is large enough to move it from one class to the next, improvements in encoding performance are pretty irrelevant except for fanboy "my CPU is better than yours" battles.

Are there really people here who are 1) not video professionals and 2) spend several hours a day running encodes? If so, I'm curious to hear what you're doing, why, and what percentage of people you think are doing the same.

 

[Geegeeoh]

I have over 20 TB of h264 video. I'm transcoding to h265 to save space.
Sadly it looks like hardware acceleration won't help a bit.

I don't think many have this need.

 

[DrMrLordX]

I still don’t understand what your argument against the MacBook Pro hardware acceleration is about.

It's about what the CPU can do unaided by dedicated hardware, period. It's about, what can your system do with a new workload you weren't planning to run when you bought the machine, or a workload that the manufacturer decided not to support with dedicated hw?

People seem strangely focused on encoding performance here for reasons I cannot understand.

You really don't know? People have been building encode boxes around here (and on other enthusiast forums and A/V forums) for decades now. Media PCs are expected to carry out those tasks at numerous different quality levels and/or resolutions that may not be supported by dedicated hw.

 

[Carfax83]

People seem strangely focused on encoding performance here for reasons I cannot understand. Unless you are spending hours running encodes every day (in which case you have a preferred workflow and can just benchmark that instead of worrying about how it performs with software you don't use) who cares how fast it is? It will fall into one of three classes - "cup of coffee", "lunch break" or "overnight". Unless the performance difference is large enough to move it from one class to the next, improvements in encoding performance are pretty irrelevant except for fanboy "my CPU is better than yours" battles.

Are there really people here who are 1) not video professionals and 2) spend several hours a day running encodes? If so, I'm curious to hear what you're doing, why, and what percentage of people you think are doing the same.

I personally rarely encode anything, but I still pay attention to a CPU's encoding performance because encoding is an excellent test of a CPU's raw computational performance and microarchitecture. Encoding is one of the most demanding tasks that end users might run on their systems, along with gaming.

 

[igor_kavinski]

M1-max Benchmarks - OpenBenchmarking.org

Unfortunately, I can't figure out how to do a comparison with other CPUs. Help, please?

 

[Nothingness]

I personally rarely encode anything, but I still pay attention to a CPU's encoding performance because encoding is an excellent test of a CPU's raw computational performance and microarchitecture.

I second this. I never had to encode anything, but I'm interested in knowing encode performance which beyond a facet of CPU performance, can also show whether a port to a new ISA is late or not.

Encoding is one of the most demanding tasks that end users might run on their systems, along with gaming.

I'd say browsing is also very demanding. Code size is much larger than encoder and uses many things encoding software usually don't (e.g. JIT). And it's much more used than encoding so its performance matters more to end users

 

[DrMrLordX]

I have personally done transcodes in the past, and now sometimes I stream stuff via OBS where software encode performance is important.