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

 

[biostud]

Legitimate new player in the PC hardware space for the first time in decades. I don’t see how anyone who bothers posting here wouldn’t be excited about that.

Because if your hobby is building gaming desktops, why would an Apple laptop excite you?

I acknowledge that it is an amazing SoC, but just as their phone SoCs are class leading, but as a pc enthusiast and one who likes to build my own desktops, neither products are really relevant to me and therefore meh.

 

[KentState]

Because if your hobby is building gaming desktops, why would an Apple laptop excite you?

I acknowledge that it is an amazing SoC, but just as their phone SoCs are class leading, but as a pc enthusiast and one who likes to build my own desktops, neither products are really relevant to me and therefore meh.

I was wondering if I'm the only one that felt this way. It's impressive what Apple can do with their proprietary hardware, but it's not something I would ever use at this point in time.

 

[igor_kavinski]

Apple's achievement is more important in terms of a peek into the future of what's possible in SoC designs from Intel/AMD and maybe even nVidia. Apple could inspire these companies to make gigantic jumps in computing efficiency. In the end, we win.

 

[Heartbreaker]

Because if your hobby is building gaming desktops, why would an Apple laptop excite you?

I acknowledge that it is an amazing SoC, but just as their phone SoCs are class leading, but as a pc enthusiast and one who likes to build my own desktops, neither products are really relevant to me and therefore meh.

Then why are you in this thread? Why waste your time on things that you have no interest in. Do you haunt the My Little Pony threads, to tell them you aren't into multi-colored ponies?

 

[Eug]

Apple's achievement is more important in terms of a peek into the future of what's possible in SoC designs from Intel/AMD and maybe even nVidia. Apple could inspire these companies to make gigantic jumps in computing efficiency. In the end, we win.

It’s more important for me because Apple is no longer hamstrung by Intel’s design choices.

I’m no creative professional but I sometimes do like to edit simple videos for the family. It always struck me as rather telling that I got better video editing performance on my 2017 iPad Pro 10.5” with A10X using LumaFusion than what people would get with 13” Intel MacBook Pros of the same generation.

 

[beginner99]

Apple's achievement is more important in terms of a peek into the future of what's possible in SoC designs from Intel/AMD and maybe even nVidia. Apple could inspire these companies to make gigantic jumps in computing efficiency. In the end, we win.

I hope you are right but I'm much more pessimistic. here the great advantage of x86 / the pc of it's universal standards and hence easy way to build custom solutions is also it's downfall because the software can never be so optimized, especially not closed source. Yeah if you do HPC you compile for your system but we are taking about end-user devices here.

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.
And still, the >$2000 pc laptop market isnt' that great as well. Let's not forget even an M1 Pro wouldn't be good for margins if put in <$1000 devices.
Intel/AMD issues is that the $800 device isn't much worse than the $2000 device really when it comes to the CPU and even me as "pro user" can't really see the need for the M1 Max GPU. I just run that stuff on a server as it will most likely run for hours anyway. The really big deal from M1 is the single threaded performance and accelerators.

 

[biostud]

Then why are you in this thread? Why waste your time on things that you have no interest in. Do you haunt the My Little Pony threads, to tell them you aren't into multi-colored ponies?

Because it is an amazing piece of hardware and I need to know if I'm missing out on anything. How would I know if I didn't participate?

 

[Doug S]

Thanks. I don't really shop laptops with discrete GPUs so I wasn't aware of the competitive landscape. I was mostly going off of reactions from twitter folk and some reviewers who were commenting on how thick and heavy it was, but they were probably using the M1 laptops as a reference point versus the x86 competitors.

The thickness complaints are funny, in light of how often I've seen someone whining about how Apple cares more than about making things thin than it does about usability. So Apple releases a laptop with a 99.9 Wh battery, the largest FAA regulations will permit, and revives missing ports people complained about losing, and then we have people complaining it is too thick?

I guess they could have used a battery half the size and still come out just fine in battery life and decided against the port revival, and been thinner and a lot lighter. Just goes to show that someone is going to be pissed off no matter what you do!

 

[Roland00Address]

Because it is an amazing piece of hardware and I need to know if I'm missing out on anything. How would I know if I didn't participate?

You could listen, when you speak you listen less and vice versa.

 

[Heartbreaker]

Because it is an amazing piece of hardware and I need to know if I'm missing out on anything. How would I know if I didn't participate?

By your standard listed above, you weren't going to like it no matter what.

And you don't need to participate in this thread, to find out more, you can just read the reviews, and Anandtech benchmarking.

And even if this thread contained vital info, you can read without commenting.

Dropping in to say: It's Meh, because it's not the kind of thing you are interested in, isn't helpful for anyone.

 

[biostud]

You could listen, when you speak you listen less and vice versa.

It is called actively listening, I try to be part of the discussion.

 

[biostud]

By your standard listed above, you weren't going to like it no matter what.

And you don't need to participate in this thread, to find out more, you can just read the reviews, and Anandtech benchmarking.

And even if this thread contained vital info, you can read without commenting.

Dropping in to say: It's Meh, because it's not the kind of thing you are interested in, isn't helpful for anyone.

I like it from a technical standpoint, but if it was a 57bn GPU I would be more excited. But I'll stop derailing the thread now.

 

[moinmoin]

Some Rust compilation benchmarks:

Some comparisons from the reddit thread

So not earth shattering, but looks pretty darn good for software development

Note that the numbers are likely not comparable since they are likely from native compilations (M1 compiling to ARM code, x86 compiling to x86 code) instead both compiling to e.g. ARM code. There's a complete lack of info in that thread in that regard.

 

[Hitman928]

Note that the numbers are likely not comparable since they are likely from native compilations (M1 compiling to ARM code, x86 compiling to x86 code) instead both compiling to e.g. ARM code. There's a complete lack of info in that thread in that regard.

Some replies in the thread talk about this. It is expected to take longer when compiling x86 code though how much is unknown. One poster posted a time comparison and x86 took about 1.5-2.5% longer than compiling for ARM but it was one of the short, simpler compile jobs so who knows if that translates to the longer, more complex jobs.

The jobs under test also obviously stop scaling at some point with added cores as the 5950x is only about %30 faster than the 5800HS despite having double the cores and close to 4x the power limit so most likely that 30% is probably accomplished just through higher frequencies and increased L3 cache.

 

[moinmoin]

Some replies in the thread talk about this. It is expected to take longer when compiling x86 code though how much is unknown. One poster posted a time comparison and x86 took about 1.5-2.5% longer than compiling for ARM but it was one of the short, simpler compile jobs so who knows if that translates to the longer, more complex jobs.

The jobs under test also obviously stop scaling at some point with added cores as the 5950x is only about %30 faster than the 5800HS despite having double the cores and close to 4x the power limit so most likely that 30% is probably accomplished just through higher frequencies and increased L3 cache.

All nice info. Doesn't change the fact that the numbers simply are not comparable unless both systems compile to the same target.

 

[Gideon]

Note that the numbers are likely not comparable since they are likely from native compilations (M1 compiling to ARM code, x86 compiling to x86 code) instead both compiling to e.g. ARM code. There's a complete lack of info in that thread in that regard.

While true, for rust it should usually be in the ballpark. There was one post that covered cross-compilation or ripgrep and the difference was nelgilible

 

[moinmoin]

While true, for rust it should usually be in the ballpark. There was one post that covered cross-compilation or ripgrep and the difference was nelgilible

View attachment 51978

Good to know. Is that also true on x86?

 

[Gideon]

Good to know. Is that also true on x86?

What do you mean? The table shows timesfor both a x86 mac and ARM mac both compiling for both x86_64 and ARM. Full quote:

I was curious about this also, so tested it. Cross compiling ripgrep --release on a 9880HK MBP16 (8 core/16 thread) vs an M1 MBP13 (old 4p/4e core one):
MBP16 (x86_64): x86_64: 29.81s aarch64: 29.10s
MBP13 (aarch64): x86_64: 22.41s aarch64: 22.01s

 

[LightningZ71]

There is no question of "if AMD can make a comparable" APU/SOC to what Apple has with the M1 pro and max. The question is, "are enough laptop makers interested enough to purchase sufficient quantities of them to make the project make financial sense?"

 

[moinmoin]

What do you mean? The table shows timesfor both a x86 mac and ARM mac both compiling for both x86_64 and ARM. Full quote:

Sorry, I just looked at the MBP part and didn't see that it included older Intel parts. Thanks.

 

[Eug]

Running Cinebench R23, CPU power for M1 Pro 8-core (6+2) maxes out at 21.3 Watts, with entire package drawing 24.3 Watts. DRAM power is listed at 0.9 Watts and GPU is listed at 0.013 Watts.

Intel 13" MacBook Pro 2.0 GHz Core i5-1038NG7 CPU was 39 Watts with package drawing 41 Watts according to Intel Power Gadget. Note that the Core i5-1038NG7 has a TDP of 28 Watts.

For Xcode compile, M1 Pro did it in 111 sec, and the fans never ramped up.
The Intel machine did it in 309 sec, with the CPU hitting 100C and fans at full blast.

4K ProRes RAW to ProRes 422 export:
67 seconds on M1 Pro 8-core
107 seconds on Mac Pro 12-core W-3235 + Afterburner card
165 seconds on Mac Pro 12-core W-3235, no Afterburner card
600 seconds* on 13" Intel MBP

*Reviewer gave up on the Intel MBP half way through after 5 minutes, so he estimated the encode at about 10 minutes.

 

[JasonLD]

M1 Pro/Max Macbook Pros might be the first laptop in a long time that I can actually set it on my lap for several hours without risking a burn lol. This is one hell of an impressive machine.

 

[gdansk]

I guess they could have used a battery half the size and still come out just fine in battery life and decided against the port revival, and been thinner and a lot lighter. Just goes to show that someone is going to be pissed off no matter what you do!

For reference, the last generation MBP 16 (2019) also had an airline-limited battery ("Built‑in 100‑watt‑hour lithium‑polymer battery") while being a bit lighter. Not that I'm complaining but that battery doesn't justify the weight increase.

 

[Doug S]

For reference, the last generation MBP 16 (2019) also had an airline-limited battery ("Built‑in 100‑watt‑hour lithium‑polymer battery") while being a bit lighter. Not that I'm complaining but that battery doesn't justify the weight increase.

The design is a bit different though, since the x86 one had glued down batteries almost impossible to replace, while the new one uses batteries with the pull tape like the iPhone that anyone comfortable enough to twist a few screws on their laptop can replace. Not sure how much that affects weight but generally anything designed to be modular is going to weigh more.

 

[moinmoin]

The design is a bit different though, since the x86 one had glued down batteries almost impossible to replace, while the new one uses batteries with the pull tape like the iPhone that anyone comfortable enough to twist a few screws on their laptop can replace. Not sure how much that affects weight but generally anything designed to be modular is going to weigh more.

But does making it more modular add that much to the weight? It's not like the connectors are made of lead (I sure hope!).