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

 

[Nothingness]

It's actually kinda ironic that Intel flubbed it with an own goal - because with their own fabs they could outbid QC any day of the week with a comparable product.

Building a 5G modem, beyond the patents minefield, is incredibly complex, it's not only a fab/price issue.

BTW Apple bought the Intel modem division in 2019. But at that point Intel had already internalized everything after closing all R&D centers in Europe. So it comes as no surprise that Apple still doesn't have its own competitive chip. They probably bought only buggy unusable IP and some patents, and too few engineers with expertise.

 

[igor_kavinski]

I just don't like MacOS, but I can live with that given the benefits I get

What will it be then? M3 Max or M4 MBA/MBP? Look forward to your detailed user experience report including pain points (which hopefully you won't gloss over).

 

[Nothingness]

What will it be then? M3 Max or M4 MBA/MBP? Look forward to your detailed user experience report including pain points (which hopefully you won't gloss over).

I will be waiting for M4 MBP 😀 In the mean time, I will have learned enough of MacOS to not hate it, I hope.

Edit: at the moment the only pain point is MacOS. Along with the corporate spyware that comes with it; that also applies to Linux and Windows, security is paramount, but that comes with a price.
Another thing is that MBP 16" is heavy. But the felt quality over the Lenovo P1 is miles aways.

 

[soresu]

I don't disagree with that. I guess I would love the SD Elite X laptops too for the reasons you mentioned. MacOS is just so turdy that I would rather prefer going WoA than Apple anything. Unless Apple supported an official Linux distro.

It's interesting but still largely irrelevant to me until the ARM bois ramp up their SIMD pipes to match the x86 competition.

Yes they are doing really, really well in CPU scalar compute - but they are still mincing around in the shallow end of the pool with traditional CPU vector compute where the best out in the consumer market is only a quarter what Intel has and AMD will have once Zen5 launches.

 

[soresu]

They probably bought only buggy unusable IP and some patents, and too few engineers with expertise

At least at Apple there is a chance they might attract some former engineers from that division if they have a more palettable reputation.

 

[igor_kavinski]

Edit: at the moment the only pain point is MacOS.

How about trying dual boot with Asahi Linux?

Installing Asahi Linux on Apple M1 and M2 Macs

Get your Apple Silicon Mac up and running with Asahi Linux using our easy-to-follow instructions.

thetechdeck.hashnode.dev

 

[Nothingness]

It's interesting but still largely irrelevant to me until the ARM bois ramp up their SIMD pipes to match the x86 competition.

Yes they are doing really, really well in CPU scalar compute - but they are still mincing around in the shallow end of the pool with traditional CPU vector compute where the best out in the consumer market is only a quarter what Intel has and AMD will have once Zen5 launches.

X925 has 6 128-bit pipes all of which can do FMA. This is more than the 512-bit FMA throughput of Zen4. So they are getting better, slowly.

But yes I agree that Arm lags behind in SIMD performance. I'd love to see some 256-bit wide SVE CPU back into the game.

 

[Nothingness]

How about trying dual boot with Asahi Linux?

Installing Asahi Linux on Apple M1 and M2 Macs

Get your Apple Silicon Mac up and running with Asahi Linux using our easy-to-follow instructions.

thetechdeck.hashnode.dev

Corporate policies forbid that. They might be evaluating Asahi, I didn't check with IT. But as long as it's not officially sanctioned, that's a no go.

 

[soresu]

I'd love to see some 256-bit wide SVE CPU back into the game

Technically that was V1 so they were never in the game at the consumer level thus far.

Not sure how much FMA applies to my interest area in software video encoding though.

With ML starting to creep into the codecs it's a given that different strengths will start to rule the roost in that area eventually.

 

[igor_kavinski]

BTW Apple bought the Intel modem division in 2019. But at that point Intel had already internalized everything after closing all R&D centers in Europe. So it comes as no surprise that Apple still doesn't have its own competitive chip. They probably bought only buggy unusable IP and some patents, and too few engineers with expertise.

Must have pissed them off real good. Could be one of the reasons they may never go Intel again.

 

[poke01]

macos is good at video/photo editing(Apple has excellent video accelerators) and watching movies/tv shows for me(there this app called Infuse, which I will always use).
I much prefer Windows or Linux for work and gaming.

 

[soresu]

macos is good at video/photo editing(Apple has excellent video accelerators) and watching movies/tv shows for me(there this app called Infuse, which I will always use).
I much prefer Windows or Linux for work and gaming.

MPC-HC is plenty good for watching videos on Windows.

Plus Kodi is pretty multi platform.

 

[igor_kavinski]

MPC-HC is plenty good for watching videos on Windows.

Plus Kodi is pretty multi platform.

I think it's the display quality and HDR features on Macbooks that poke01 prefers. They do have some of the best displays.

 

[poke01]

MPC-HC is plenty good for watching videos on Windows.

Plus Kodi is pretty multi platform.

yes but I like the fancy design of the app. If theres one thing macOS devs do, they make their apps look fresh. https://firecore.com/infuse

I cosume a lot of movies and shows, so I got a full license. License is tied to your Apple ID so you can use it on Apple TV, Mac, iPhone etc. They increased the price now but I got it when it was cheaper. It also connects to Jellyfin, if I want to operate from a PC. Nice app and devs are awesome. I don't pay for Netflix etc so it evens out.

 

[soresu]

I think it's the display quality and HDR features on Macbooks that poke01 prefers. They do have some of the best displays.

Ye unfortunately HDR support on Windows is still pretty cack.

 

[FlameTail]

Ye unfortunately HDR support on Windows is still pretty cack.

Even Android has HDR figured out. What an embarrassment for Microsoft.

With ML starting to creep into the codecs it's a given that different strengths will start to rule the roost in that area eventually.

Isn't Matrix units better for AI/ML? That's what SME is for.

X925 has 6 128-bit pipes all of which can do FMA. This is more than the 512-bit FMA throughput of Zen4. So they are getting better, slowly.

But yes I agree that Arm lags behind in SIMD performance. I'd love to see some 256-bit wide SVE CPU back into the game.

Chips&Cheese article noted that Oryon had strong vector performance.

Also Apple M4 can compute 512 bit vectors on the SME block using SSVE (Streaming SVE).

Another thing is that MBP 16" is heavy.

You might want to wait for M5 MBP then. Apple is moving to a thinner design (and probably adopting OLED as well), in the same vein as the new M4 iPad Pro.

But the felt quality over the Lenovo P1 is miles aways.

Aah, Macbooks lined with felt!

 

[soresu]

Even Android has HDR figured out

Android is also used as a TV OS remember.

They didn't really have much choice but to get it working.

I'm not altogether sure that nVidia didn't help them with it considering SHIELD TV was one of the first Android TV devices.

 

[Nothingness]

Also Apple M4 can compute 512 bit vectors on the SME block using SSVE (Streaming SVE).

The SSVE performance of the SME block of the M4 is poor. It's lower than the aggregated NEON performance of the P cores. For instance I'm not sure one could write a faster codec with it.

 

[johnsonwax]

TBH I'd expect owning stock to be a good source of distorted opinions too. If not bigger than not owning. Easily makes one more likely to really reach for the good news, ignore/try to downplay the bad, and get the "convince others" impulses.

Yeah, this is true. You have to resist focusing on the company as the thing that matters and focus on the market the company is in as the thing that matters. Everyone is susceptible to seeking out reinforcement of their opinions. If you overindex on the company, you also lose perspective on what would be better investments. If you focus on the market, and try to understand what the relationships of the various players are in that market, you'll see where others are better positioned for that market and diversify to those other players.

 

[Doug S]

Just because the people you rub shoulders with on a semi regular basis are "eyes wide open" Apple buyers doesn't mean that the great unwashed masses that make up likely 95+% of Apple's bottom line correspond to this level of self awareness in their purchasing habits.

The idea that Apple buyers are "eyes wide open" is ludicrous, I agree. The idea that ANY company's customers in the consumer sphere are is ludicrous! Whether it is iPhone or Android, Mac or PC, the overwhelming majority of sales are to people who haven't much clue when it comes to tech. Qualcomm is relying on that to sell ARM PCs to people, because most don't understand what the difference is between ARM and x86. If you asked them, most would confuse ARM with AMD, or think Qualcomm is just another company making CPUs - which in their mind means they make the same CPUs that AMD and Intel do - because they SURE AS HECK don't know what an "ISA" is, nor would they care if you tried to explain it to them.

And that's fine, they don't need to understand that stuff to full utilize their smartphone or PC. One could argue that they'd make "better" purchase decisions (where "better" inevitably means better in the opinion of the person making the claim) if they understood the tech better, but so what? Some people reading this may know PCs inside and out, but couldn't tell you what's what under the hood of their car (or even know how to open it without looking at the manual) but that doesn't impact their ability to use the car for the reason they purchased it - to get from point A to point B.

 

[igor_kavinski]

Some people reading this may know PCs inside and out, but couldn't tell you what's what under the hood of their car (or even know how to open it without looking at the manual) but that doesn't impact their ability to use the car for the reason they purchased it - to get from point A to point B.

Did we meet somewhere? Coz you just described me, minus knowing PCs inside and out (I learn something new everyday). But yeah, cars, printers, women. None of those things make sense to me. Oh, women are not products? Sorry, I got confused coz they seem to come in so many varieties and some of them actually want you to sample them and spend inordinate amounts of time trying to make themselves look appealing to potential... (yeah, not gonna say it!)

 

[Doug S]

Building a 5G modem, beyond the patents minefield, is incredibly complex, it's not only a fab/price issue.

BTW Apple bought the Intel modem division in 2019. But at that point Intel had already internalized everything after closing all R&D centers in Europe. So it comes as no surprise that Apple still doesn't have its own competitive chip. They probably bought only buggy unusable IP and some patents, and too few engineers with expertise.

The baseband software is the difficult part because there are SO many carriers, so much complexity in how everything ties together in all the countries around the world, etc. By comparison the hardware itself is much easier. The software team at Intel was 3000 people in India, and those who have worked with India teams before know how quickly the best ones turnover to get a raise at another company, while the useless ones stay forever. Dunno if Apple kept that team intact or tried to start fresh with their own or what.

I've mentioned before my rule of thumb for project work where 10% of people do 80% of the work, 80% of people do 40% of the work, and 10% of people are so incompetent that fixing their screwups costs the project 20%. I would adjust that based on my experience working with teams in India (admittedly only a couple year's worth spread across about a decade) to say that 5% do 100% of the work, and I'm not sure what the heck the rest do other than show up and collect a paycheck.

The baseband software will be made easier if Apple made it LTE and 5G only, and didn't support 2G and 3G at all. Yes I know that just because the US has phased out 2G/3G doesn't mean everywhere has and many places don't even have it on the roadmap. But still having 2G or 3G in a country doesn't mean that there is actually a lot of land area that is only covered by 2G/3G and still has no LTE and 5G. The reason they are keeping 2G/3G around is because they have a lot of older phones, or stuff like alarms or smart meters using it, not because they haven't deployed at least LTE widely.

 

[The Hardcard]

(I can't think up a less political analogy for this issue, so please excuse it mods)

That sounds not so unsimilar from various times I have seen people say:



Just because the people you rub shoulders with on a semi regular basis are "eyes wide open" Apple buyers doesn't mean that the great unwashed masses that make up likely 95+% of Apple's bottom line correspond to this level of self awareness in their purchasing habits.

It's your basic echo chamber problem.

Much like what we occasionally have here in this forum of hardware enthusiasts where we fool ourselves into thinking that the vast majority of PC buying public think just like we do, when even a moments introspection or a visit to your local PC retailer would reveal that just isn't the case.

I am not claiming the majority of Apple buyers who are eyes wide open. I am just saying that they exist. Apple’s products with the current pricing has an appeal to people who understand what they are paying for.

Nearly every device user I know is not tech savvy. Most are not even interested in tech, they just want things done that require tech devices. Most of the Apple buyers, most of the Android buyers most of the Windows buyers, and in fact, I know non-savvy Linux users as well.

However, I dispute the claim that there is nothing to Apple but marketing deception. There are large numbers of technically savvy people, who are very aware of what the alternatives are, and have experiences with those alternatives, who choose to buy Apple devices.

My point is that if there was nothing beneficial to Apple on a technical level for the prices they charge, if it was 100% a marketing trick, there would be none of those buyers. And Igor seems to imply that there is no way Apple can make a single sale of any product unless the buyer has no understanding they can do better elsewhere.

He has also implied that it is only people spending extra money to look cool to other, sacrificing technical capacity per dollar for oohs and ahhs as they whip out the shiny logo, soaking up status points while suffering lower productivity with a less capable tool.

Apple does plenty of things from hardware through operating system, choices and software, and with pricing on the cherry on top that mean, it makes perfect sense why they have a smaller marketshare. It does not surprise me at all that Windows and Android have much larger marketshare.

But I contend that Apple is a valid choice for a certain number of people based on technical hardware merits and operating system and software decisions. In fact, I think that if everyone was given the opportunity and wanted to take the time to learn and experience all of their options that Apple marketshare would go up at the current pricing. A slightly larger, but still minority marketshare to be sure.

 

[soresu]

Chips&Cheese article noted that Oryon had strong vector performance.

Currently in ARM world that just means X1 or better, as 4x 128 bit has basically become the standard for perf/beeg cores since that time.

 

[soresu]

Isn't Matrix units better for AI/ML? That's what SME is for.

The codecs that use it aren't really anywhere in the market at the moment so I wouldn't get too excited.

VVC/H.266 has some, and likely it's successor ECM/H.267 will have more - but VVC has only just got its first hardware decode implementations with little content out there, and ECM is years off even getting to standards level yet.

(proprietary codec licensing pools b0rked the long term potential of HEVC popularity and pretty much flatlined enthusiasm for VVC which already has multiple separate patent pools, it seems like the patent contributors will never learn their lesson - this is basically the reason AV1 exists in the first place)

AOM's AV1 successor codec AV2 will definitely make use of ML in some capacity, but exactly how much is hard to say at this point, and AOM haven't even made any announcement about roadmaps for it so I wouldn't expect it anytime soon.

Opus audio 1.5 does make much more use of ML vs the original codec, but I think the newer ML features are not backwards compatible with earlier decoders, so there could be fragmentation issues going forward 🤦‍♂️

By the time any of this is sorted out there will be more AI/ML feature support in pretty much every processor out there, so there's little point comparing the hardware landscape at the moment based on what currently exists.