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

 

[lobz]

Agree!

Let me re-state my point I guess for that person's sake, and to clarify what I meant.

We have very limited benchmarks right now. We can't just take the limited benchmarks, for example, from Anandtech, or Puget Systems, or The Verge - and say it's the fastest. Using just the benchmarks in AT's review is fraught with issues, as one example, since they ran only two relatively unconstrained somewhat real-world CPU test against other laptop chips, CB23 MT and Speedometer *** (please see below), where two AMD chips, including one at 15W TDP, beat it (and by "it" I mean the Mac mini with M1, not the MBP or MBA) pretty handily in one, and the M1 won pretty handily in the other. But we didn't even get a GB5 MT score against other AMD laptop chips. And in SPEC 2017 rate-N, the M1 in the Mac mini loses to the 4900HS.

My point being that right now, it's damn impressive at its power consumption, but probably due to sheer core/thread disadvantage, it still loses to Zen2-based laptop APUs in 2 of the 3 non-ST tests where it was compared to laptop chips in AT's review. Calling it the fastest laptop CPU or fastest laptop chip is at best premature, and most likely wrong for the types of benchmarks that are typically run by tech sites on CPUs/chips.

*** I am not sure what the setup of AT's Speedometer test was, but its results are off... I suspect they ran it in Safari on all computers. Speedometer on my Ryzen 3600 scores 140 in Chrome and 142 in Edge. On my iPhone it's 146 in Chrome. That they only got 140 for a 5950X tells you that this test is a bit fishy. I mean, if they want to use Apple's browser optimized for macOS/iOS and Arm on x86 Windows PCs as the comparison... whatever. But I think it must be pointed out that this test is a bit of a stacked deck.

Nah, Geekbench is fishy (I said the same when Zen 2 was suddenly better than Intel in GB4). Speedometer is completely useless for cross-platform comparisons. Foe terminology's sake, fishy doesn't mean intentionally skewed for me. That's reserved for the likes of Userbenchmark and the multiple decades-old Intel subsidiary benchmark suites.

 

[Gideon]

Speedometer is completely useless for cross-platform comparisons.

Could you elaborate why? It just runs simple TodoMVC apps in Different javascript frameworks and measures the page rendering speed. If the browser used is the same (just compiled to different platforms).

IMO seems ok-ishl as it actually renders stuff quicker on these setupsb(albeit in a trivially simplistic SPA). Still, users should see similar speedups in actual js- frontend applications. To me it certainly doesn't seem any worse than any other browser benchmark It at least is considerably harder to game than something like Octane or Kraken.

 

[lobz]

@Gideon Let's just say my 9900K system shouldn't be meaningfully faster than my P20 pro according to numbers I got 6 weeks ago when I ran the test both on my PC and my phone. That being said, I don't think you want me to start explaining how the 2 experiences differ.

(edit) P.S.: yeah I know, I'm basically a Chinese spy 😂

 

[mikegg]

Let's just say my 9900K system shouldn't be meaningfully faster than my P20 pro according to numbers I got 6 weeks ago when I ran the test both on my PC and my phone. That being said, I don't think you want me to start explaining how the 2 experiences differ.

(edit) P.S.: yeah I know, I'm basically a Chinese spy 😂

Explain.

All speedometer does is measure the completion times of a todo app in React, Ember, Preact, jQuery, etc. These are the most commonly used frameworks on the web and it's a good indicator of web performance.

FYI, my iPhone XR browses websites considerably faster than my Intel Mac. Speedometer is also clearly faster on my XR than Intel Mac. The benchmark matches my real-life experience.

 

[lobz]

Explain.

All speedometer does is measure the completion times of a todo app in React, Ember, Preact, jQuery, etc. These are the most commonly used frameworks on the web and it's a good indicator of web performance.

FYI, my iPhone XR browses websites considerably faster than my Intel Mac. Speedometer is also clearly faster on my XR than Intel Mac. The benchmark matches my real-life experience.

OK first of all, you showed me again how your downvote was nothing but a childish display of frustration, because you've (probably unintentionally) just explained to me why and how canned synthetic benchmarks shouldn't be used as trouncing cards when you must say something is 'tEh most fastestest'. Mock Cinebench on its everyday relevance all you want, but it's an actually good indicator of how a system would affect the time you needed to do your job in that particular type of workload. Then you try to tackle the situation with this absurd subjective experience statement, but all I can say is this: if that's true, be a true altruist and don't give away your Intel Mac for Christmas, just throw it into the trash. I'm not sure what you've done with that, but I've never seen a properly put together PC browsing the internet slower than any phone. At the very best case for any phone you shouldn't notice any difference, since web browsing is 100% instant on any good PC. Sorry mate but you're just being freaking ridiculous.

Also if you truly perceive your phone to be considerably faster than browsing on an actually OK desktop system, I must quote House M.D. and say, 'I have no knowledge of alien physiology', so it's not a debate I can participate in.

 

[Qwertilot]

Also if you truly perceive your phone to be considerably faster than browsing on an actually OK desktop system, I must quote House M.D. and say, 'I have no knowledge of alien physiology', so it's not a debate I can participate in.

Some calm perhaps? There's some absurdly JS intensive sites out there, and phones have been spending huge effort in optimising everything for them - Safari seems scarily fast.
Also a bunch of potential side advantages in a top end phone like fully up to date wifi, storage, ram etc.

When the base CPU speed is as close as it is here, why shouldn't they sometimes get ahead?

I've definitely found a website I use where my M1 MBA is indeed, visibly much faster than my OK, if mildly old (Ivy i5, sata SSD), desktop. Some website doing a 3D terrain fly through of a route.

Could be a whole number of things involved there of course! Not interested in finding out what

 

[Thunder 57]

OK first of all, you showed me again how your downvote was nothing but a childish display of frustration...

I must admit I had the same problem just a day or two ago. You may have seen. Thankfully we were able to resolve it somewhat cordially. I've got your back though, if I see a non-deserved downvote on you here I'll bump it back up.

 

[DrMrLordX]

It says "Written in C", "optimized in Assembler". I would expect they have the original C code, and use compiler directives to override it with the Assembler code only on appropriate architecture.

And guess what? The "original C code" doesn't necessarily support SIMD at all. In fact, for it to support any SIMD and compile to a native binary on M1, it would have to support NEON and nothing else.

Rosetta 2 can't translate AVX

:

It can translate everything up to SSE4.x . Any SIMD-capable application that expects to run on non-AVX-capable hardware - x86 or otherwise! - has code paths for older SIMD ISAs. Some older applications have codepaths going all the way back to SSE2 (the minimum for x86_64). What you're seeing in the Phoronix FLAC benchmark is a non-SIMD native binary vs. SSE4.x Rosetta translation.

Anyway. Phoronix results are "curious" to say the least. But I certainly wouldn't base my decision process on them.

It's a first shot at compiling and running FOSS on M1. Someone would have to go in and make sure NEON was fully-supported in all SIMD-capable applications and recompile with M1 as a target. Some of the native M1 compiles on Phoronix probably do give us an idea of how well M1 can be expected to perform, and some of them don't. For now, it's up to the reader to decide which results are representative of actual performance.

Whilst Rosetta does not support translating the AVX family of SIMD instructions, it does support translating the SSE family of SIMD instructions. FLAC has codepaths that make use of SSE intrinsics in addition to codepaths that make use of AVX intrinsics. The reason that the FLAC benchmark shows greater performance when run via Rosetta is most likely due to the fact that SSE SIMD instructions are translated to NEON SIMD instructions and achieve a greater throughput than can be achieved by compiling the fallback FLAC C codebase to ARM.

Yeah, that!

This nails it. Why aren't people seeing what you're seeing? Are they reading completely different benchmarks?

In order to make M1 out to be "the fastest" mobile CPU, you have to lean on GB5 and SPEC numbers.

 

[Thunder 57]

Some calm perhaps? There's some absurdly JS intensive sites out there, and phones have been spending huge effort in optimising everything for them - Safari seems scarily fast.
Also a bunch of potential side advantages in a top end phone like fully up to date wifi, storage, ram etc.

When the base CPU speed is as close as it is here, why shouldn't they sometimes get ahead?

I've definitely found a website I use where my M1 MBA is indeed, visibly much faster than my OK, if mildly old (Ivy i5, sata SSD), desktop. Some website doing a 3D terrain fly through of a route.

Could be a whole number of things involved there of course! Not interested in finding out what

That's pretty old. I loved my Ivy desktop, but that I got it in early 2013 and it was out before then. Plus, a SATA SSD vs NVMe? I know, computers last a lot longer these days so 7-8 years isn't that big a deal anymore. But Ivy didn't even have AVX2 (Not that it would matter here).

 

[mikegg]

At the very best case for any phone you shouldn't notice any difference, since web browsing is 100% instant on any good PC. Sorry mate but you're just being freaking ridiculous.

Also if you truly perceive your phone to be considerably faster than browsing on an actually OK desktop system, I must quote House M.D. and say, 'I have no knowledge of alien physiology', so it's not a debate I can participate in.

Yea no. my iPhone XR has much better single-core performance than my Macbook Pro 2015 and loads websites significantly faster and has no visible stuttering when browsing websites.

And as speedometer and other single core tests demonstrate, the M1 should be faster at web browsing than the 4900HS.

Javascript is single threaded. It shouldn't be a surprise to anyone that Apple SoCs dominate web browsing performance.

Some calm perhaps? There's some absurdly JS intensive sites out there, and phones have been spending huge effort in optimising everything for them - Safari seems scarily fast.
Also a bunch of potential side advantages in a top end phone like fully up to date wifi, storage, ram etc.

When the base CPU speed is as close as it is here, why shouldn't they sometimes get ahead?

I've definitely found a website I use where my M1 MBA is indeed, visibly much faster than my OK, if mildly old (Ivy i5, sata SSD), desktop. Some website doing a 3D terrain fly through of a route.

Could be a whole number of things involved there of course! Not interested in finding out what

This is correct.

 

[amrnuke]

Saying that the M1 is the best overall laptop chip you can get is not controversial. Saying the M1 has the highest performing CPU in a laptop chip is not controversial.

Except that's not what you said, is it?
You said it's the fastest laptop CPU, period.
You said it's the fastest laptop chip, period.

So the question is, did you mis-speak, and intend to say that the M1 has the fastest core, or are you ignoring the benchmarks that test the (and I'll emphasize this) whole laptop chip / CPU rather than a small portion of it?

 

[mikegg]

Saying that the M1 is the best overall laptop chip you can get is not controversial. Saying the M1 has the highest performing CPU in a laptop chip is not controversial.

Except that's not what you said, is it?
You said it's the fastest laptop CPU, period.
You said it's the fastest laptop chip, period.

So the question is, did you mis-speak, and intend to say that the M1 has the fastest core, or are you ignoring the benchmarks that test the (and I'll emphasize this) whole laptop chip / CPU rather than a small portion of it?

It's the same thing. I didn't say anything different.

M1 CPU is the fastest, best performing, best overall, speediest laptop CPU out there. Period. Not controversial.

M1 chip (entire SoC) is the fastest, best performing, best overall, speediest laptop chip out there. Period. Not controversial.

Is it more clear to you now?

Edit: Added "period" and "not controversial" since you seem to get caught up when I either write it or not write it.

 

[mikegg]

I also don't get the obsession of highly MT tests where 8-core chips win very slightly. Once 6 or 8 big-core M chips start rolling out they will be very competitive even with AMDs 12 and 16 core desktop chips.

AMD mobile CPUs only definitively win in Cinebench multithread against the M1, which is a fairly niche product that historically favors AMD CPUs. For example, in Cinebench, Zen2 is competitive with Comet Lake in single-core even though in real life apps and in just about every other single-threaded benchmark, Comet Lake wins handily.

M1 beats or is competitive in just about every other multi-threaded application/benchmark against Renoir. And we haven't even seen comparable benchmarks where M1 accelerators such as the neural engine or encode/decode units accelerate traditionally multi-threaded applications beyond what Renoir can bruteforce.

 

[lobz]

Some calm perhaps? There's some absurdly JS intensive sites out there, and phones have been spending huge effort in optimising everything for them - Safari seems scarily fast.
Also a bunch of potential side advantages in a top end phone like fully up to date wifi, storage, ram etc.

When the base CPU speed is as close as it is here, why shouldn't they sometimes get ahead?

I've definitely found a website I use where my M1 MBA is indeed, visibly much faster than my OK, if mildly old (Ivy i5, sata SSD), desktop. Some website doing a 3D terrain fly through of a route.

Could be a whole number of things involved there of course! Not interested in finding out what

So your M1 system is faster than a system using an architecture from literally almost a decade ago. Can anyone come up with a framework in which we could actually argue around facts? I personally can't, I just start to feel like I'm trying to speak about practical life in an insane asylum.

 

[Heartbreaker]

FYI this was the exact marketing speech Intel had with the launch of Tiger Lake: ST performance wins in notebooks, GPU lead is paramount, ML features improve user experience even more. This has not slowed down Renoir even a bit.

I think you need a better criteria than: "has not slowed down Renoir", since there are multitude of factors far beyond performance, including market positioning, pricing, availability, starting market share, compostion of the lineups, goodwill, etc... that affect who sells better.

I am not saying Tiger Lake is better than Renoir, just that this is useless criteria. It's like saying, why hasn't your supposedly high quality restaurant slowed down McDonalds?

Something a little less nebulous please.

 

[lobz]

It's the same thing. I didn't say anything different.

Are you deliberately pretending being illiterate, so people stop contradicting you out of pure pity? Sadly, it's an actual question. I wanna know.

 

[mikegg]

So your M1 system is faster than a system using an architecture from literally almost a decade ago. Can anyone come up with a framework in which we could actually argue around facts? I personally can't, I just start to feel like I'm trying to speak about practical life in an insane asylum.

Edit: Nevermind. You weren't quoting my post. This is an erroneous response.

 

[lobz]

Some calm perhaps? There's some absurdly JS intensive sites out there, and phones have been spending huge effort in optimising everything for them - Safari seems scarily fast.
Also a bunch of potential side advantages in a top end phone like fully up to date wifi, storage, ram etc.

When the base CPU speed is as close as it is here, why shouldn't they sometimes get ahead?

I've definitely found a website I use where my M1 MBA is indeed, visibly much faster than my OK, if mildly old (Ivy i5, sata SSD), desktop. Some website doing a 3D terrain fly through of a route.

Could be a whole number of things involved there of course! Not interested in finding out what

We probably frequent vastly different kinds of websites then. My chrome experience has been mostly instant on my desktop. Can you describe me how you perceive faster than instant browsing? Don't show me benchmark numbers, but please describe the actual experience. Also if it's possible, use a good system that's at least comparable in age to your M1 rig.

 

[mikegg]

Are you deliberately pretending being illiterate, so people stop contradicting you out of pure pity? Sadly, it's an actual question. I wanna know.

Why don't you actually argue how what I said was different instead of using ad hominem?

Like seriously.

How many times do I have to say the exact same thing?

 

[lobz]

Why don't you actually argue how what I said was different instead of using ad hominem?

Like seriously.

How many times do I have to say the exact same thing?

How about you actually argue with the people who did point out the exact contradictions instead?

 

[mikegg]

How about you actually argue with the people who did point out the exact contradictions instead?

What were the contradictions? List them out. I've love to see someone actually word by word, point out where the contradictions are. No one has done that yet.

 

[Qwertilot]

That's pretty old. I loved my Ivy desktop, but that I got it in early 2013 and it was out before then. Plus, a SATA SSD vs NVMe? I know, computers last a lot longer these days so 7-8 years isn't that big a deal anymore. But Ivy didn't even have AVX2 (Not that it would matter here).

It won't die, and I haven't had anywhere to hand it down to, so it'd have been a total mother board or system replacement. It wasn't really until Zen2 that there was something dramatically fast enough to motivate the effort/expense. Starting to hoard Might well be the grown up mac mini when they release that, because I rather like small, quiet, efficient machines. Also Acorn nolstagia.

It does, of course!, still do most websites very quickly. The specific website I'm thinking of runs about the same speed on my desktop as on my 1st gen ipad pro. Rather basic 3D terrain fly through rendered from mapping data. Those devices very useable but do need periodic pausing to pop textures in. The M1 MBA doesn't.

I'd assumed it was a network based bottleneck but I guess not. Must(?) be doing all the rendering in comically inefficient CPU Javascript. A bit of a daft example but, hey, real world

 

[Eug]

Surfing comparisons on a phone vs. desktop are often pointless, considering they are often served different versions of the same sites. On many sites, even my iPhone 6s with A9 seems zippy. It doesn't take much horsepower to run a mobile site.

A more meaningful comparison is an iPad running iPadOS 13 or later vs. a Mac desktop since they get served the same site versions.

In that context, my iMac Core i5-7600 sometimes feels faster than my iPad 7 with A10 for surfing, but the iPad 7 is surprisingly good overall. My iMac feels roughly the same as my iPad Pro 10.5" with A10X though, but then again, both of them feel fast, and going to a faster machine (like a 2018 iPad Pro 12.9" with A12X) doesn't seem like much of an improvement. Side by side I can notice a difference, but in real world usage I don't care because they're all good.

It should be noted that the iPad Pro 10.5" with A10X and the iPad 7 with A10 have the same single-core speed, but the iPad Pro still sometimes feel faster just for surfing. Not all the time, but sometimes with more complex websites. That said, even the A10 is fine for most people. Also, I'd say my MacBook m3-7Y32 is roughly in the same ballpark as the iPad 7, although I tend to multi-task on the MacBook more.

OTOH, my iPad Air 2 with A8X is clearly slower. Still very acceptable, but definitely slower.

My Macs are on Big Sur, and my iPads are on iPadOS 14.

tl;dr:

iPhone vs desktop surfing comparison is not a good one, since they are served different content. Better to compare iPad vs desktop/laptop.

iPad Air 2 (A8X with 2 GB RAM) - Kinda slow but tolerable.
iPad 7 (A10 with 3 GB RAM) - Very decent, but sometimes the slower SoC is noticeable.
2017 MacBook 12 (m3-7Y32 with 16 GB RAM) - Very decent, but sometimes the slower CPU is noticeable.
2017 iPad Pro 10.5" (A10X with 4 GB RAM) - Snappy
2017 iMac 27" (i5-7600 with 24 GB RAM) - Snappy
2018 iPad Pro 12.9" (A12X with 4 GB RAM) - Snappy. Perhaps slightly better than iMac and iPad Pro 10.5, but not enough for me to care in real world usage.

 

[lobz]

Surfing comparisons on a phone vs. desktop are often pointless, considering they are often served different versions of the same sites. On many sites, even my iPhone 6s with A9 seems zippy. It doesn't take much horsepower to run a mobile site.

A more meaningful comparison is an iPad running iPadOS 13 or later vs. a Mac desktop since they get served the same site versions.

In that context, my iMac Core i5-7600 sometimes feels faster than my iPad 7 with A10 for surfing, but the iPad 7 is surprisingly good overall. My iMac feels roughly the same as my iPad Pro 10.5" with A10X though, but then again, both of them feel fast, and going to a faster machine (like a 2018 iPad Pro 12.9" with A12X) doesn't seem like much of an improvement. Side by side I can notice a difference, but in real world usage I don't care because they're all good.

It should be noted that the iPad Pro 10.5" with A10X and the iPad 7 with A10 have the same single-core speed, but the iPad Pro still sometimes feel faster just for surfing. Not all the time, but sometimes with more complex websites. That said, even the A10 is fine for most people. Also, I'd say my MacBook m3-7Y32 is roughly in the same ballpark as the iPad 7, although I tend to multi-task on the MacBook more.

OTOH, my iPad Air 2 with A8X is clearly slower. Still very acceptable, but definitely slower.

My Macs are on Big Sur, and my iPads are on iPadOS 14.

tl;dr:

iPhone vs desktop surfing comparison is not a good one, since they are served different content. Better to compare iPad vs desktop/laptop.

iPad Air 2 (A8X with 2 GB RAM) - Kinda slow but tolerable.
iPad 7 (A10 with 3 GB RAM) - Very decent, but sometimes the slower SoC is noticeable.
2017 MacBook 12 (m3-7Y32 with 16 GB RAM) - Very decent, but sometimes the slower CPU is noticeable.
2017 iPad Pro 10.5" (A10X with 4 GB RAM) - Snappy
2017 iMac 27" (i5-7600 with 24 GB RAM) - Snappy
2018 iPad Pro 12.9" (A12X with 4 GB RAM) - Snappy. Perhaps slightly better than iMac and iPad Pro 10.5, but not enough for me to care in real world usage.

Your posts bring some sense into this thread, thanks

I also wanna mention, in 90% of all websites, I actually DO MISS the features not present on the mobile version and it drives me crazy to the point where I check 'desktop version' in the settings and start zooming in desperately. What a mess

 

[Eug]

BTW, where I might see more differences in speed is with window resizing. Older slower machines have less smooth window resizing. Fortunately, I'm not resizing the window with every page visited.

Judging by the reviews, M1 is very fast for window resizing, but they don't usually comment on this metric.