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

 

[Eug]

I feel for the poor sap who has to edit a 4k video with 16 GB of ram.

...

I think it was Linus Tech Tips complaining that Threadripper could only take 256 GB of ram, and that was just barely enough for his 4k video editing.

For some of the comparisons, FOR CERTAIN CONTENT, the M1 Mac was doing better than a 192 GB Mac Pro for 4K editing. YouTuber type 4K video editing, but 4K video editing nonetheless.

The big advantage here was that the M1 Mac was perfectly smooth in the actual editing process with butter smooth scrolling across the timeline, and perfectly clean playback. In contrast, the Mac Pro was stuttering through the same content.

As I mentioned before, it seems that Apple has purpose built the hardware accelerators to handle this sort of thing. It doesn't cover everything, and sometimes M1 fails hard once you hit 8K, but it's still remarkable what they've done... esp. considering this is a mobile SoC for ultrabooks.

Geekbench was used in Apple comparison, because it was the primary cross platform benchmark that would actually work on iPhones/iPads. I can't even think of anything else that easily fits the bill, unless you want to compile your own. Thus everyone used Geekbench. It's understandable that when it's pretty much all you have, you use it.

IMO scorn for Geekbench grew as performance of recent generations iPhone/iPad showed iPhone delivering desktop performance. The mindset developed that this was "Too good to be True" performance for a Smartphone SoC, and therefore Geekbench must be faulty.

Scorn for Geekbench began with its release, and lasted through Geekbench 3 IIRC. From what I gather, it gained a lot more respect with Geekbench 4 and then Geekbench 5.

The point you make is also probably true, but that is actually a later phenomenon.

Cinebench. I don't remember being prominent until Ryzen hit the scene. Since then it seems to be choice benchmark for AMD, and "AMD people" to show the core count advantage over Intel. If anything it's even less applicable than Geekbench to the real world. Geekbench is a composite benchmark. CB is a just one single task. In a way it's just one of the most simple embarrassingly parallel benchmarks out there.

Cinebench has been popular for just about forever even at AnandTech, at least when comparing Macs and Windows machines since it's cross platform and excludes the GPU. And yes, it's simple, which is actually one of its draws since anyone can run the benchmarks. In fact, there are databases out there which include scores of various Cinebench iterations, at stock clocks and overclocks.

Your experience with these benchmarks may be different though if for example you only started watching this stuff in say the last half-dozen years or so.

 

[Eug]

Shifting away from kerfuffle about some people attacking benchmarks they don't like:

Here is a new Arstechnica review of MBA. They were declined a review sample laptops, so this is his personal machine he bought to replace an Intel MBA:

Apple’s M1 MacBook Air has that Apple Silicon magic

One of my top 3 features of all time for Mac laptops is MagSafe. It's such a shame they killed it.

BTW, I see a lot of consumers choosing the fanless throttle-risking MacBook Air over the MacBook Pro, just because of that irritating Touch Bar. Thank the gods though that Apple at least listened a bit and brought back the physical ESC key.

 

[Qwertilot]

So let’s say the make a M1X chip for 16” with 8 big cores and a higher ram amount.

Do you think Apple would also use this same chip in the 13” pro even though it would be a different motherboard than the air and 13” M1 models?

Yes? Not the 13" pro which has the M1 in now, but they've got a load of 13" models with Intel in at the moment. Those are obviously getting replaced

 

[Doug S]

How are they going to massacre everything when their multithreaded performance is so far behind?

Where are they "so far behind"? Unless you are comparing them against PCs with 8 big cores (and probably SMT enabled as well) they aren't far behind.

And the "massacre" would come when new generations of the M* come with more big cores, eventually scaling up to the Mac Pro in a couple years. I expect at least 32 big cores in the high end there, maybe more. You're going to have to be comparing with some awfully big (and expensive) x86 hardware to put that "far behind".

 

[Qwertilot]

That seems like wishful thinking. How many people are willing to switch OS to get faster performance?

Definitely not disagreeing with your basic conclusion but the M1 is giving these devices *much more* than extra CPU performance.

Passive, or near silent (mbp), running.
Considerable extra battery life.
Instant restart from sleep.
Enough iGpu to play games at a reasonable level.
Mildly cheaper.
etc

They're all non trivial selling points. No reason they shouldn't pick up some more market share.

Obviously with their using iPads for the 'cheaper' end of the market they won't get a massive percentage of overall laptop sales.

 

[name99]

I thought the M1 was an 8 core CPU?

This is kind of dancing around the terminology if you ask me. If the smaller more efficient cores are being tapped during multithreaded workloads, then the M1 is an 8 core CPU no matter which way you slice it.

Well this is the difference between people whose goal is to understand technology and people whose goal isredacted



Inappropriate language for the tech forums.


esquared
Anandtech Forum Director

 

[dmens]

In the x86 world, those specs at that price point is a joke. It is good for nothing but facebook and twitter, web browsing and tinkering

That is a damning indictment of the x86 ecosystem more than an attack on Apple "specs". Heh.

Here is a free pro-tip: the sole reason Firestorm is not outpacing x86 cores on single thread perf by enormous margins is because Apple does not bother playing stupid PCU games, like with Intel shoving 1.4 volts into a single core when it detects a single thread workload and pray the work is done within the turbo window before it has to throttle down and not blow through the thermal sink capacity.

 

[moinmoin]

I can imagine mobile games and Apple Arcade will ensure much better games support for ARM Macs going forward than ever happened with Intel Macs. Though it will be an interesting split, with console games (aside by Nintendo) being closer to (Wintel) PCs and mobile games closer to (ARM) Macs.

 

[Doug S]

Pretty much no one, right? That's exactly the way I see it, and it goes double for businesses. If better hardware = adoption of the platform were true, Apple would not have been bleeding market share in the phone space to Android for the past 3 years.

Anyone who thinks Apple is "bleeding market share to Android" doesn't understand the market. If you look at the market share of overall mobile phone sales Apple's share has been remarkably steady for years. The reason it looks like they are slowly losing share is if you only look at the smartphone market share. That matters because the cheapest phones out there are not smartphones they are so-called "feature phones". Android has been slowly moving further and further to the low end - you can get Android phones for $50 now but in places with truly low incomes there are plenty of people for whom that's still too much so even in 2020 there are over a half billion feature phones shipped.

This graph illustrates what I'm talking about, the growth in "smartphone" sales is a lot higher than the growth in overall phone sales. Apple's sales of the parts of the market they participate in are remaining steady or even growing (as some Android owners abandon the "high end" segment and figure a $150 phone is good enough) and all the Android growth is in lower and lower market ranges as they slowly eat the feature phone market from the top down as cheaper and cheaper Androids get built.

Global feature phone and smartphone shipments 2008-2020 | Statista

The forecast illustrates the global feature phone and smartphone shipments from 2008 to 2010, with forecasts from 2011 to 2020.

www.statista.com

 

[Doug S]

People are falling over themselves for the new M1 chip, but I don't get why people act like Apple came out of nowhere or are surprised that its good. This is the result of a decade of building possibly the premiere chip engineering team in the world. This move was rumored when they bought the one chip company like a decade back. Frankly, I'm surprised it took this long, considering how much resources they put towards it. It goes to show how difficult such a move actually is, and that its basically a complete overhaul needed to even begin such a move (which is why it took Apple and is taking others even longer). Frankly if it wasn't competitive with Intel (languishing under the weight of poor business management) and AMD (operating with what, 1/100, 1/1000 of Apple's resources) then it would've been almost outright embarrassing (see Nvidia's moves prior to buying ARM to try and get there).

I'm honestly a bit surprised there isn't more second guessing AMD's move to go x86 instead of ARM. Makes you wonder where they might would have been now had they gone ARM instead. Imagine if AMD had taken all the ideas that make Zen great, paired with many of the moves that Apple has made. We've seen others show that even without optimizations that ARM can be quite competitive in servers. AMD could have been sitting pretty there. With a competitive ARM core, perhaps Microsoft doesn't dilly dally Windows support. Further, Android would give them an alternative to build from. Heck, maybe Apple would've ditched Intel and foregone their own chips even. And AMD might would have bought ARM instead of Nvidia. When Zen 1 launched there might not have been as big of a performance shift, but I bet efficiency would have been bigger (and I think arguably pushed for quicker adoption; plus in servers they could just add more cores to get there), and they'd probably be even further along in the perf/W push. I think ARM would have actually been a benefit for AMD, as it would have gotten them competitive in laptops and servers faster. Instead of companies looking to do their own ARM designs to try and get away from the Intel situation, they'd have AMD chips. It also would've likely helped them in game consoles.

The other thing I'm a bit surprised hasn't been talked about more. Apple is now all in on ARM, and Nvidia just bought ARM. That might actually be a blessing though, as Apple has the legal resources to take on Nvidia should the latter try something. But, it also means that Apple doesn't get ARM to itself. Had Apple bought ARM then rolled out these, I bet the talk would be the death of Windows, and that Android might not be far behind. Perhaps we'd be looking at one or more of the PC 3 (AMD, Intel, Nvidia) partnering up.

The reason it took Apple this long is that they clearly wanted to wait until they would have no one doubting the move was for the benefit of their customers, rather than for Apple's own benefit. They could only do that if it was a clear performance win over the x86 Macs they replace, and even running under emulation was not a compromise.

I don't understand your thought that AMD "could have gone ARM". Who the hell would they be selling their ARM desktop chips to? What OS were they going to run? Who would be buying their ARM server CPUs, and what OS would they run? OK here "Linux" makes a good answer, but there's no reason to believe that AMD could make an ARM CPU that performs better than their x86 CPUs. If you think the ARM ISA has some sort of an advantage over x86, you're simply wrong. The ISA hasn't made any difference since the 90s, once transistors became so plentiful that translating CISC instructions into microops became reasonable the ISA became irrelevant from a performance perspective. AMD would have shot themselves in the foot if they tried to go ARM instead of sticking with x86.

 

[jeanlain]

Apple doesn't really have to do much of anything to become big in gaming, gaming is already moving the direction Apple is moving.

Apple has been openly hostile toward game streaming services and for a good reason. Game streaming goes against their business model and they're certainly not moving in that direction. They sell phones and tablets with high computing power. Traditional (local) gaming is one of the reason why people buy their devices.

 

[LightningZ71]

Have you been attempting to hide from the universe for the last year? The whole point of the Epic/Apple lawsuit is Apple's requirement that every single transaction in an app in their pay them 30% rent. Apple makes loads of money from the apps they host.

 

[Mopetar]

I think it's because most people didn't really believe that Apple iPhone SoCs were really desktop class. There wasn't a lot cross platform benchmarks that it could run, and people chose to believe Geekbench was invalid, rather than believe that iPhones had desktop class performance.

AnandTech has been showing SPEC results for a wide variety of the different benchmarks in the iPhone/iPad reviews for a while now and those results were consistent with the idea that they had some truly incredible performance. I think the people that dismissed Apple's SoC achievements were just not paying attention or burying their heads in the sand.

 

[dmens]

I don't understand your thought If you think the ARM ISA has some sort of an advantage over x86, you're simply wrong. The ISA hasn't made any difference since the 90s, once transistors became so plentiful that translating CISC instructions into microops became reasonable the ISA became irrelevant from a performance perspective. AMD would have shot themselves in the foot if they tried to go ARM instead of sticking with x86.

This is 100% wrong.

 

[LightningZ71]

Oh sweet baby (insert diety of your choice here)!

Here we go with yet another near religious argument over ISA pros and cons...

 

[NostaSeronx]

But when did the marketing happening with AMD slides and so on*?

AMD Analyst Day - Barcelona, Fusion and Beyond - PC Perspective

AMD Analyst Day - Barcelona, Fusion and Beyond AMD is currently holding an analyst day in which they talk about the company's current status as well as the

pcper.com

" + Title + "

2007-ish.

Compared to 2006:

AMD Analyst Day Platform Announcements

www.anandtech.com

Comparable to Torrenza's Package Integration/Core Integration(Fusion being Core Integration, but maybe with R--- product chiplet fusion will be a thing).

 

[nxre]

What browser are you using? All the major browsers nowadays are multithreaded, though they may go about it in different ways.

Web-browsing, the most common task in any computer, has been multi-threaded for years.

Browsers have been multi threaded, web browsing has not. Javascript is still a fundamentally single threaded languague.

Code compiling might not be inherently multithreaded, but it seems to respond well to parallelization. You see the same thing in the Spec GCC sub test.

Independent modules of the same code can be compiled in parallel which does not meant that compiling is a multithread workload even if it can benefit from it in certain occasions. You can google this in more detail as I grossly simplified it.
On the topic of games, the bottleneck is still single thread performance in most cases when it comes to CPU.
I dont understand what is hard to get about the idea that single thread code will never be obsolete: if your code requires the result of a previous operation to proceed, it cant be parallel.

 

[Bam360]

If all or most tasks could be hugely multithreaded, there would be no need to invest a huge amount of transistors on improving the core, because you would get much more performance and efficiency by cramming tons of cores and making each of them much simpler.

 

[Gideon]

The graph I posted with 4-8 simultaenous thread occupancy on an Android phone includes every workload needed to render a page, hence there's much more work to do on the CPU than just one JavaScript thread.

Web browsing is multi-threaded.

Yeah, as it touched a subject I'm a bit more familiar with,I'll add my 2 cents.

Modern browsers are actually quite multi-threaded.

JavaScript on a single page in a single tab is indeed single-threaded. But only if:

• That page doesn't use Web Workers - with which you can run things in other threads
• Doesn't use WebAssembly libraries (which does support multi-threading now)
• Doesn't have nested Iframes from other pages on different domains (that in both Chromium and Firefox run in a separate process even, let-alone thread)

And JavaScript is just a small part of the list of things that browsers do to actually deliver your page.
I'll take Firefox as an example (as they have nice blog-posts explaining stuff I'll link below), but many things from here apply to Chrome as well:

• Composition is done off main-thread and can be quite taxing
  • (and in fact nowadays GPU is used quite extensively as well, see Firefox Webrender for instance)
• CSS layout can be very parallel if designed well
• HTML Parsing can be parallel (Though this is a different engine not used in FF currently, the Gecko engine also supports that)
• And yet again Chrome and Firefox have different processes for every domain you have open. (for firefox it's very recent, see project fission)

And all of that doesn't even account to the fact that JavaScript itself is not just interpreted but also JIT (Just In Time) compiled - meaning the hot parts of your code get compiled to native a code to speed it up.
That happens on-the-fly and is also done off-main-thread.

Browsers aren't parallel in a sense that say Cinebench is (they don't tax all your cores to 100% with embarrassingly parallel workloads) but they absolutely do use multiple threads.

Just try running your browser in a virtual machine with a single vCPU, won't be a pleasant excercise even with a fast processor and using hypervisors (e.g. minimal overhead).
In fact Just going from a 2C/4T cpu to a 4C/8T CPU is often a noticable speedup while browsing (if you also have other stuff open). Less so with more cores, but the it's still there.

The reason for that misconception is simple - Benchmarks.

Old benchmarks (Octane, Kraken etc) are the worst, testing some fringe 100% atypical javascipt functionality (that sometimes get's special paths in browsers just to look good). For instance Firefox totally redesigned their JS engine (Warp) and got real-life site loading improcements from 12-20% on JS heavy sites (Google Docs, Reddit, Netflix). These benchmarks showed significant regressions (despite actual browsing experience improving greatly).

Speedometer 2.0 is much better, cause it at least renders an actual javacsript SPA (in different js frameworks) but it's still very simplistic and doesn't do stuff that is actually done on most websites: No embedded twitter/facebook frames, no google analytics, no ads (separate domains) nor adblock (which can be quite taxing on ad-heavy sites). No SVG or huge image-rendering, no worker-threads reading/writing from IndexedDB (separate thread).

All of these things benefit from more cores (at least more than 1-2) in the real world, but aren't shown in any benchmarks.

 

[Doug S]

Happy birthday JavaScript, you are 25 today on Dec 4th. 25 years later we have computers with the M1 that can actually run you as intended 🥳 🎂 🎉 !

Don't worry, before long web designers will come up with even more complex and convoluted Javascript that makes even an M1 seem slow.

 

[Kenmitch]

That would suck if true....I have no need to worry about it as I don't use Apple's products. For those who do it might be of some consern.

Your new Mac's speedy SSD might not last as long as it should | Macworld

 

[moinmoin]

I'm starting to think Apple's approach to the cores is more similar to AMD's Zen than to ARM's big.LITTLE, but with two big advantages.

With core designs there are essentially three different approaches:

• Optimize the chip for absolute performance. This is the way the desktop x86 chips went, pursuing ever higher frequencies. To achieve those the design is made less denser thus need a bigger area.
• Optimize the chip for lowest possible power usage. This is the approach ARM's little cores and Intel's Atom cores took, in both cases features that consume much power are either cut or significantly trail full featured performance/high frequency chips.
• Optimize for power efficiency. This is (used to be?) the standard approach for new ARM designs, the chip is full featured and very efficient at it, but there is no headroom for pushing performance further through increasing the frequency. This both ensures the chip is always efficient and allows for a very dense design.

AMD's approach with Zen has been to optimize the cores for power efficiency, but then relinquish the density for achieving higher absolute power. The result is that Zen can be used across the whole range of products, mobile, desktop and server as well as ST and manycore MT workloads. The downside being Zen cores being relatively big compared to ARM cores on the same process node. And especially with Zen 3 we can see well how power efficiency carries the MT performance while higher ST performance on too many cores at once is limited by TDP.

And I think that balance between MT performance and ST performance within a shared TDP is what led Apple to the 8+2 configuration. Apple's efficiency cores are optimized for power efficiency and as such perfect for MT workloads, all while wasting little space for the cores. Apple's high performance cores are optimized for absolute performance and as such perfect for ST workloads. In the ideal case all 10 cores can max out at the same time without breaking the TDP budget. The advantage versus AMD's approach then is that with this design Apple can ensure both predictable reproducible MT and ST performance at once, whereas for Zen at unlimited TDP MT workloads would turn the chip into a highly inefficient all core ST performance furnace.

 

[naukkis]

Part of the reason efficiency cores are much more efficient is that they have a more limited instruction set, which is less of a problem for someone like Apple since the control both hardware and software and can design both with a shared goal in mind.

What? As I know every big.LITTLE configuration have identical instruction set support for both big and little cores, including Apple. With incompatible instruction sets it would be impossible to switch between core clusters making whole arrangement pretty much useless.

 

[repoman27]

That was with a discrete GPU. They are using "Quad" (or "Eight" on the Max) channel LPDDR5.

LPDDR5 channels are generally 16-bit, so it would actually be 16 or 32-channel. They're using LPDDR5-6400 in x128 packages... which is rather exotic. And not at all what I was expecting seeing as Apple has never previously shipped a product utilizing LPDDR5. Pretty bonkers.

 

[jeanlain]

EDIT: In case you don't understand, they aren't measuring real performance. They are "extrapolating" perf/watt.

They aren't measuring real performance? So they're lying I suppose?
First you claimed they only used iGPUs, then that they only measured on battery, and now that you've been proven wrong, you assert that they haven't actually measured performance. What's next?

EDIT; you think that the dimmed part of the curve is extrapolation from Apple's part, based on measures on battery power? That Apple couldn't be bothered to plug the laptop in to measure actual performance in this situation?