Discussion Apple Silicon SoC thread

Eug · Nov 10, 2020

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

IvanKaramazov · Nov 11, 2020

IntelUser2000 said:
The MT scores are similar to Ryzen 7 4800U.

Unless I'm misunderstanding something, it's actually a bit faster MT than the Ryzen 9 4900HS, their higher-watt laptop part. The 4800U is a good bit slower than that. Zen 3 mobile will presumably change that equation whenever they come, though of course the M1 single core appears to be slightly faster than the Zen 3 desktop parts already.

Geekbench being such a short bench, the most interesting questions for me going forward are how much throttling there is on the MBA v MBP v Mini. Certainly the Air isn't going to sustain those numbers, but will be crazy fast for general usage. If the others can sustain that level of performance though, it will be truly something.

Eug · Nov 11, 2020

IvanKaramazov said:
Unless I'm misunderstanding something, it's actually a bit faster MT than the Ryzen 9 4900HS, their higher-watt laptop part. The 4800U is a good bit slower than that. Zen 3 mobile will presumably change that equation whenever they come, though of course the M1 single core appears to be slightly faster than the Zen 3 desktop parts already.

Geekbench being such a short bench, the most interesting questions for me going forward are how much throttling there is on the MBA v MBP v Mini. Certainly the Air isn't going to sustain those numbers, but will be crazy fast for general usage. If the others can sustain that level of performance though, it will be truly something.

Remember, these are parts that fit into that 10-15 watt Intel TDP envelope. With a fan in a Mac mini or a MacBook Pro, it should do very, very well. In a MacBook Air? The jury is still out on that one.

Too bad about that 16 GB memory limitation though. That said, for an entry level machine, that's fine for most people. Funny this "entry level" machine is way faster than every single computer I have in my house. In fact it beats my 2017 Core i5 iMac by a large margin. The difference though is I have 24 GB in my iMac. I even have 16 GB in my 2017 12" MacBook. I find for most of my type of usage, memory is more important than CPU speed these days. 16 GB is more than enough on my laptop, but for my desktop it can very occasionally get restrictive. (My needs are lighter than many in this forum, but nonetheless when you're running dual 27" screens, there is a tendency to heavily multi-task, and many business applications are feeling the bloat these days.)

I suspect what's going to happen with the MacBook Pro, the Mac mini, and the iMac is that they will all use M1 for entry level, limited to 16 GB, but they will also all have higher end chips (M1X?) that will be much faster and support up to 64 GB.

Are all those higher end Macs going to have the same GPU as each other though?

And then later on, Apple will release a new killer Mac Pro and maybe a killer iMac Pro too.

wlee15 · Nov 11, 2020

IvanKaramazov said:
Unless I'm misunderstanding something, it's actually a bit faster MT than the Ryzen 9 4900HS, their higher-watt laptop part.

There are lots of lower scoring 4900HS so I suspect users are playing around with TDP. The higher scoring 4900HS go up 1200 Single and around 8000 Multi.

name99 · Nov 11, 2020

Eug said:
? Cinebench is both a single-core and a multi-core benchmark.

Realistically it's an MC benchmark; it's a task that's optimally performed on as many cores as possible.

name99 · Nov 11, 2020

Eug said:
The one on the App Store is 5.25. The Geekbench scores out in the wild now are 5.3. I think the scores are legit.

Holy shit. 1687 single-core and 7433 multi-core for a frickin' fanless MacBook Air!

View attachment 33656

Geekbench says 3.2 GHz.

I wouldn't trust the 3.2GHz number. The mini gets essentially the same ST performance at what's claimed to be 3.0GHz. I'm guessing that GB still can't perfectly detect frequency, and the 6% or so boost over iPhone is the larger caches and the 128-wide memory.

Presumably we'll know for sure over the next few days.

HurleyBird · Nov 11, 2020

name99 said:
Realistically it's an MC benchmark; it's a task that's optimally performed on as many cores as possible.

All MT benchmarks are also ST benchmarks if you constrain then to a single thread

name99 · Nov 11, 2020

Eug said:
Remember, these are parts that fit into that 10-15 watt Intel TDP envelope. With a fan in a Mac mini or a MacBook Pro, it should do very, very well. In a MacBook Air? The jury is still out on that one.

Too bad about that 16 GB memory limitation though. That said, for an entry level machine, that's fine for most people. Funny this "entry level" machine is way faster than every single computer I have in my house. In fact it beats my 2017 Core i5 iMac by a large margin. The difference though is I have 24 GB in my iMac. I even have 16 GB in my 2017 12" MacBook. I find for most of my type of usage, memory is more important than CPU speed these days. 16 GB is more than enough on my laptop, but for my desktop it can very occasionally get restrictive. (My needs are lighter than many in this forum, but nonetheless when you're running dual 27" screens, there is a tendency to heavily multi-task, and many business applications are feeling the bloat these days.)

I suspect what's going to happen with the MacBook Pro, the Mac mini, and the iMac is that they will all use M1 for entry level, limited to 16 GB, but they will also all have higher end chips (M1X?) that will be much faster and support up to 64 GB.

Are all those higher end Macs going to have the same GPU as each other though?

And then later on, Apple will release a new killer Mac Pro and maybe a killer iMac Pro too.

What are you swapping to? VM off an HD is awful, but VM off a GB/s+ SSD I find I hardly notice.

name99 · Nov 11, 2020

HurleyBird said:
All MT benchmarks are also ST benchmarks if you constrain then to a single thread

Of course! The question is do they still give you useful info?
What do I learn from ST cinebench? Is there a paper/web page that tells me the way in which it hits or misses RAM? Is it bandwidth constrained? How much it uses vectors vs scalar, int vs FP? How difficult are its branches? ...

MT cinebench can be argued as, even if it's an opaque box, it represents a problem many people seem to care about. ST cinebench seems to me an opaque box solving a problem no-one cares about.

itsmydamnation · Nov 11, 2020

name99 said:
MT cinebench can be argued as, even if it's an opaque box, it represents a problem many people seem to care about. ST cinebench seems to me an opaque box solving a problem no-one cares about.

thats pretty average logic, who has cared about the performance of dijkstra since about '94? does that mean we invalidate any benchmark where dijkstra influences its "score"?

Hans de Vries · Nov 11, 2020

The Ryzen 7 5800X is saved by VAES in geekbench 5.3 it seems. Under Linux (which is much closer to MacOS) we get:

https://browser.geekbench.com/v5/cpu/compare/4653030?baseline=4651916

1769: ST for Ryzen 7 5800X @ 4.8 GHz
1719: ST for Apple M1 @ 3.2 GHz

itsmydamnation · Nov 11, 2020

Hans de Vries said:
The Ryzen 7 5800X is saved by VAES in geekbench 5.3 it seems. Under Linux (which is much closer to MacOS) we get:

https://browser.geekbench.com/v5/cpu/compare/4653030?baseline=4651916

1769: ST for Ryzen 7 5800X @ 4.8 GHz
1719: ST for Apple M1 @ 3.2 GHz

its there still an issue with complier version? with xcode using a newer compiler then the linux geekbench version?
As noted by Andrea in the A14 review there are some big optimisations with the newer visions that affect atleast some spec2k6 tests performance.

Hitman928 · Nov 11, 2020

itsmydamnation said:
its there still an issue with complier version? with xcode using a newer compiler then the linux geekbench version?
As noted by Andrea in the A14 review there are some big optimisations with the newer visions that affect atleast some spec2k6 tests performance.

As far as I can tell the Windows, Linux, and ARM versions were compiled with clang 9.0 and Apple versions were compiled with clang 11.0. This comes from the gb5 white paper.

name99 · Nov 11, 2020

itsmydamnation said:
thats pretty average logic, who has cared about the performance of dijkstra since about '94? does that mean we invalidate any benchmark where dijkstra influences its "score"?

Read what I said. The value of something like djikstra is that we know what it is doing, we understand that it's a reasonable proxy for code that engages in the manipulation of large graphs, which means it's going to stress things like prefetching and LSQ address aliasing.

name99 · Nov 11, 2020

Hitman928 said:
As far as I can tell the Windows, Linux, and ARM versions were compiled with clang 9.0 and Apple versions were compiled with clang 11.0. This comes from the gb5 white paper.

AmericanLocomotive said:
I literally have quoted their statements in my posts. That's what " " marks mean - they indicate I am quoting someone or something.

I've already been over every single one of these points, but for the last time before I give up:

Here are their actual statements:
" Our high‑performance core is the world’s fastest CPU core in low‑power silicon. "
- What does this statement mean? Are they saying it's the highest performing "low power silicon" CPU? Is it saying it's the highest performing CPU core period, and it just happens to be made using low-power silicon? Who knows? It's purposely ambiguous.
"Comparison made against the highest-performing CPUs for notebooks commercially available at the time of testing. "
- Notice it doesn't name a specific CPU or say "The highest performing notebook CPU". Them having "CPUs" as plural is pretty suspect. "Highest performing CPUs" could just mean "i7", and still technically be the truth. Them not listing the notebook models is also concerning. Maybe they're talking about the highest performing Chromebook CPU? We don't know.

“World’s best CPU performance per watt”
Qualified with this statement:
"Comparison made against high-performing CPUs for notebooks and desktops, commercially available at the time of testing.
- What does "high performing" mean? A 4700U offers much better perf/watt than a 3800X. Is a 4700u a "high performing" CPU? It will only get about 1/2 the R20 score of a 3800X, but consumes 1/9th the power. Why don't they say "against the highest perf/watt competing processor" ?

"2x faster CPU performance/Matches peak PC performance at 25% of the power"
Qualified with:
"Comparison made against latest‑generation high‑performance notebooks commercially available at the time of testing."
- Once again, what are "high-performance" notebooks? Is it a DTR running a desktop chip at 105w TDP? Is it an "H-Class" notebook chip? Could it be certain "U" class parts like a 4800U? They don't say. Obviously a 4800U offers tremendously better perf-watt than a notebook running a desktop chip. Furthermore, the phrase "latest-generation" is suspect because "latest generation" doesn't mean best. Intel's TGL CPUs are technically the "latest generation" of mobile CPUs in general (since they're newer than Renoir), but still do not offer anywhere near the same MT performance OR efficiency as a 4800U.

Or how about this gem:

"Up to 15 hrs of wireless web browsing"
Qualified with:
"The wireless web test measures battery life by wirelessly browsing 25 popular websites with display brightness set to 8 clicks from bottom."
- Which websites? What defines "popular?" Why not didn't they use an industry standard website browsing benchmark? What does "8 clicks from the bottom" even mean!? Is "8 clicks from the bottom" the same brightness on a MacBook Air and MacBook Pro? What is it in an industry-standard unit like nits, so I can compare it to another brand's systems?

All of their phrases and qualifications have been chosen very, very carefully. Intel, AMD and nVidia typically call out exact specifications of systems they are comparing against. The only systems Apple called out rather specifically were their own previous gen units, using quite lackluster CPUs. There's a reason why they aren't "naming names" with competing Windows-based systems. Compare that to AMD's recent launches - the Ryzen 5000 series and upcoming Radeon RX 6000 series where they offered hard numbers of their chips, and competitors chips right off the bat, they were later backed up and verified by 3rd party review. That's confidence. Apple has a lot of very wishy-washy qualifications for this new chip that screams to me they don't have the confidence to back those numbers up.

Look, you want Apple to steelman their marketing copy. Fine, in an ideal world we'd all demand that.
But given that you (and most of the people on your side) are strawmanning rather than steelmanning your arguments against Apple, I just can't take your crocodile tears very seriously.

AmericanLocomotive · Nov 11, 2020

itsmydamnation said:
its there still an issue with complier version? with xcode using a newer compiler then the linux geekbench version?
As noted by Andrea in the A14 review there are some big optimisations with the newer visions that affect atleast some spec2k6 tests performance.

Hitman928 said:
As far as I can tell the Windows, Linux, and ARM versions were compiled with clang 9.0 and Apple versions were compiled with clang 11.0. This comes from the gb5 white paper.

It looks like the Windows AND Linux x86 binaries are still using CLANG/LLVM 9.0, while the Apple ARM version is being compiled with CLANG/LLVM 12.0 ("Clang 12.0.0 (clang-1200.0.32.27)" ). Apple x86 is using CLANG 11.0

CLANG/LLVM 12.0 isn't even officially released yet. I'm really hoping Geekbench will update the other binaries to LLVM 12.0, otherwise it's a bit sus having a benchmark that's supposed to be "platform agnostic" using very different versions of the same compiler.

Hitman928 · Nov 11, 2020

AmericanLocomotive said:
It looks like the Windows and Linux x86 binaries are still using CLANG/LLVM 9.0, while the Apple version is being compiled with CLANG/LLVM 12.0 ("Clang 12.0.0 (clang-1200.0.32.27)" )

CLANG/LLVM 12.0 isn't even officially released yet. I'm really hoping Geekbench will update the other binaries to LLVM 12.0, otherwise it's a bit sus having a benchmark that's supposed to be "platform agnostic" using very different versions of the same compiler.

Where did you find the info about them using clang 12 for Apple?

itsmydamnation · Nov 11, 2020

name99 said:
Read what I said. The value of something like djikstra is that we know what it is doing, we understand that it's a reasonable proxy for code that engages in the manipulation of large graphs, which means it's going to stress things like prefetching and LSQ address aliasing.

And yet its performance scales almost exactly like other renderers , what it provides that other renderers dont is a nice consistent point of reference to compare.

AmericanLocomotive · Nov 11, 2020

Hitman928 said:
Where did you find the info about them using clang 12 for Apple?

Just append ".gb5" onto any of the Geekbench individual test result URLs. It will tell you the compiler version used. Windows and Linux x86 use CLANG 9.0. Apple X86 uses CLANG 11.0, Apple ARM is using CLANG 12.0

Don't get me wrong, I don't expect hugely major performance differences, but for fairness I think it's important they all use the same compiler version. We're also getting into a gray area here. For example, CLANG/LLVM (and most compilers) support CPU-specific optimization. You can pass the command argument "-march=znver3" into CLANG 12.0, and it will generate highly optomized code that will basically only run on Zen 3 based CPUs. That's not particularly practical, so most commercial software won't use such optimizations. But with these current Apple ARM binaries, there's only one architecture so you can use these optimizations. Is it "fair" to AMD and Intel that Apple gets to use highly optimized code, while AMD and Intel don't for the sake of "compatibility"?

Should there be an "Intel", "AMD" and "Apple" version of Geekbench?

amosliu137 · Nov 12, 2020

Hitman928 said:
Is that your own simulation? If so, can you run single threaded simulation?

itsmydamnation said:
its there still an issue with complier version? with xcode using a newer compiler then the linux geekbench version?
As noted by Andrea in the A14 review there are some big optimisations with the newer visions that affect atleast some spec2k6 tests performance.

A14 use clang 11.0 in iOS to get 1600. Intel in mac uses clang 11.0 too. So clang 12.0 is not magic here.

Eug · Nov 12, 2020

.

Eug said:
It's official folks! 1700+ single-core for the MacBook Pro. However, the multi-core is actually noticeably worse in this one test.

MacBookPro17,1 - Geekbench

Benchmark results for a MacBookPro17,1 with an Apple Silicon processor.

browser.geekbench.com

Geekbench 5.3: 1714 / 6802

Hmm... In retrospect, the early CPU-Monkey GB5 scores - 1634 / 7220 - are right in the right ballpark.

So, did they have some inside track before everyone else, or are they just good guessers?

Interestingly though, they used to have the same 1634 / 7220 for their A14X score, but now they've changed it to 1622 / 6879.

They still have 1.8 / 3.1 GHz listed for the clock speed for both M1 and 14X though.

beginner99 · Nov 12, 2020

The funny thing is why they are limiting their devices now on the RAM side. 16Gb of course is plenty for office+ browsing but the default is 8GB. But why would I need 4 such beefy cores just for browsing? heck I had no issue browsing on my 4 year old smartphone which is probably 10xtimes slower in these micro benches than this m1.

Ok, maybe for developers for compiling it helps but who will buy a Mac now if he develops x86 software? Can it even be used to make "general" ARM software or will everything get compiled for apple cores?

shady28 · Nov 12, 2020

beginner99 said:
The funny thing is why they are limiting their devices now on the RAM side. 16Gb of course is plenty for office+ browsing but the default is 8GB. But why would I need 4 such beefy cores just for browsing? heck I had no issue browsing on my 4 year old smartphone which is probably 10xtimes slower in these micro benches than this m1.

Ok, maybe for developers for compiling it helps but who will buy a Mac now if he develops x86 software? Can it even be used to make "general" ARM software or will everything get compiled for apple cores?

Well, there's no reason to own a Mac if you're only doing Windows development. Boot camp and all yes, but I don't see very many people doing that in code production / business environments. A real mac user is more likely to be using VMWare Fusion if they have to have some Windows App running.

On the Mac itself, developing for iOS using hardware (Apple Silicon based mac) that is like native iOS hardware (A11, A12, etc) will probably compile way way faster, and it will definitely run an iPhone or iPad simulator much faster since it won't have to emulate the SoC (edit: still has to emulate a lot, but not to as low a level).

I would hazard a guess 90% of XCode developers are primarily iOS developers.

Gideon · Nov 12, 2020

beginner99 said:
The funny thing is why they are limiting their devices now on the RAM side. 16Gb of course is plenty for office+ browsing but the default is 8GB. But why would I need 4 such beefy cores just for browsing? heck I had no issue browsing on my 4 year old smartphone which is probably 10xtimes slower in these micro benches than this m1.

Ok, maybe for developers for compiling it helps but who will buy a Mac now if he develops x86 software? Can it even be used to make "general" ARM software or will everything get compiled for apple cores?

You can add some web and universal-app developers to the list.

We do web-developing on macs (not exclusively, but many of us). For most backend stuff the OS doesn't really matter. Most interpreted and Virtual machine based stuff doesn't really care at all: NodeJS, JVM based stuff (Kotlin, Java, Scala), Ruby, Python, you name it, runs usually without issues. Even things that do compile to native machine-code (Golang and Rust) usually run mostly fine on the web-related stuff. We also do frontend stuff and apps, and if you want to build them universally for both Android and iOS, then unfortunately mac is the best option due-to vendor lock-in (Android SDK runs anywhere, xcode doesn't).

Now granted, that was with x86. With Arm things get a bit more complicated. We have to redo some docker containers to ARM as well, etc, some libraries will probably break, etc. That's why we're planning on buy a Mac Mini for evaluation how well different stacks would run and if 16GB will be bearable (it definitely won't be everywhere but hopefully in most projects)

mikegg · Nov 12, 2020

Whoever reviews the M1 chip, I hope they will incorporate benchmarks for things like machine learning, HDR video processing, cryptography acceleration, and storage controller.

The M1 dedicates a lot of silicon to non-CPU and non-GPU accelerators. It would be unfair to only measure CPU and GPU tasks.

mikegg · Nov 12, 2020

shady28 said:
Well, there's no reason to own a Mac if you're only doing Windows development. Boot camp and all yes, but I don't see very many people doing that in code production / business environments. A real mac user is more likely to be using VMWare Fusion if they have to have some Windows App running.

On the Mac itself, developing for iOS using hardware (Apple Silicon based mac) that is like native iOS hardware (A11, A12, etc) will probably compile way way faster, and it will definitely run an iPhone or iPad simulator much faster since it won't have to emulate the SoC (edit: still has to emulate a lot, but not to as low a level).

I would hazard a guess 90% of XCode developers are primarily iOS developers.

The market for pure Windows development is now very small. No software really only works for Windows nowadays. This isn't the early 2000s. Most applications have moved to the cloud, to the browser, or to mobile. Pure Windows apps are only for ultra niche professions.

Thus, if you're doing apps on the browser, mobile, or cloud, an ARM Mac is generally fine. I'm sure Docker support will be added very soon.

Discussion Apple Silicon SoC thread

Lifer

Member

Lifer

Senior member

Senior member

Senior member

Platinum Member

Senior member

Senior member

Platinum Member

Senior member

Platinum Member

Diamond Member

Senior member

Senior member

Member

Diamond Member

Platinum Member

Member

Member

Lifer

Diamond Member

Platinum Member

Golden Member

Golden Member

Golden Member