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

 

[IntelCeleron]

This is a fun little revisit. Performance Comparison: G4 vs. Core Duo

 

[ultimatebob]

This is a fun little revisit. Performance Comparison: G4 vs. Core Duo

Oh, look... it's Anand writing something positive about an Apple product. Who knew that would become his full time job someday

 

[Roland00Address]

Thinking about die size. 16 billion transistors with 120mm2 chip that is roughly. That is 133 million transistors per mm2. Now TSMC can get up to 173 million transistors per mm2 with similar logic but just for comparison 10nm intel is roughly 100 million transistors per mm2 best case scenario and 14nm intel is 37.5 million transistors per mm2 best case scenario.

Whatever number it "actually is" does not matter, it is just that apple and later amd has so much more "room to play" with the higher density tsmc process. We are seeing this now with those benchmarks which are like I said earlier...

"Spectacular, spectacular / No words in the vernacular / Can describe this great event / You'll be dumb with wonderment! ...So EX—CIT—ING, the mac users will stomp and cheer!"

 

[wlee15]

I've said this a thousand times, FFS.
Better branch predictors. Better prefetching. Better cache placement/replacement algorithms. Better ISA. More aggressive instruction fusion. Better fetch engine. Better TLB reach. Better memory controller.

How can you not comprehend this? If Apple substantially exceeds AMD's PPC, doesn't that mean that they do better than AMD on pretty everything that determines CPU performance? How else do you think the numbers work out?

The Zen 3 review noted that the prefetchers managed to foil the TLB+CLR thrash test which is designed to bypass prefetchers so better prefetchers are pretty debatable. Also something to note while Apple's cache has great latency they lag quite a bit in bandwidth. In fact clock for clock Zen 2 L2 cache is faster than the A13 L1D cache.

Also we have seen examples where an Apple processor struggles outside of it's walled garden.

Report: Super-Lightweight 12-inch MacBook Powered By Apple Silicon to Launch This Year

Some tests we did on the Apple Transition kit with Apple Silicone. Encoding the free animation movie bbb_sunflower_1080p_30fps to a MP4 format with HEVC and AAC with a video bitrate of 6000k and audio bitrate of 256k Encoding with FFmpeg 4.3.1 compiled versions for ARM and Intel (with x265...

forums.macrumors.com

Here a DTK gets beaten by a dual core i3 in x265. Now there are caveats since the DTK contains an older Apple 12z and x265 has extensive assembly optimization for x86, but lots of high performance x86 software have extensive optimization.

 

[tamz_msc]

Here a DTK gets beaten by a dual core i3 in x265. Now there are caveats since the DTK contains an older Apple 12z and x265 has extensive assembly optimization for x86, but lots of high performance x86 software have extensive optimization.

According to a poster further down in that thread, x265 has no SIMD optimization on ARM but x264 does. So a better comparison would be x264 not x265.

 

[HurleyBird]

How can you not comprehend this? If Apple substantially exceeds AMD's PPC, doesn't that mean that they do better than AMD on pretty everything that determines CPU performance? How else do you think the numbers work out?

Any of the big companies could exceed Apples PPC easily if they were willing to live with incredibly low clocks-- eg. Some humongously wide monster that only reaches 100 MHz.

Is Firestorm a better core than Zen 3 (just the core and not the rest of the uncore/package)? I'm not sure that it's obvious one way or the other. The former is on a more advanced process, doesn't seem like it's designed to scale well to high core counts (and it's just my intuition speaking, but I'm guessing that whatever refactoring AMD did to Zen 3 to allow for an 8 core CCX also allows for even larger core complexes), and has a significantly better memory controller (and no performance hobbling IOD). Despite all of those advantages going to Apple, Zen 3 still probably edges it out in overall performance albeit at a significant perf/w disadvantage. Of course, nothing designed to run at 5GHz is likely to be as power efficient as something designed to run at 3GHz. If AMD made a Zen-ish core targeting 3GHz they could build it wider, they could tighten up critical cache timings, and so on. If Apple was targeting 5GHz they wouldn't be able to be as wide, and certainly something would have to give in that big, tight L1 cache.

The real comparison is going to be Zen 4 and whatever succeeds Firestorm though.

Zen 4 is likely to be at least as big of a leap over 3 as 3 was over 2 (AMD has the low hanging fruit of the IOD lead anchor, they're likely to move towards exotic packaging, and they're getting a process and platform change).

Meanwhile, A14 was a comparatively modest improvement over A13 vs A13 over A12. This is despite advancing to a new process while A13 and A12 were both on 7nm.

I think it all depends on execution. Just comparing the last generational leap of both companies, AMD's execution looks better even though both are way above the industry standard. If A14 is an outlier, then they're about equal. How much low hanging fruit will be left for Zen 5 and A16 though? At some point someone is likely to experience severally diminishing returns. Maybe both will around the same time.

Last but not least, going forward who looks smarter will also depend on process. Poorer performance characteristics from new processes will benefit 'brainiacs', but if somehow there's a breakthrough in material science that bucks the trend and yields much improved frequencies, then suddenly 'speed demons' will look like the better bet. Netburst arguably would have been a fine foundation if the material science assumptions of its architects had come to pass.

 

[jeanlain]

Meanwhile, A14 was a comparatively modest improvement over A13 vs A13 over A12. This is despite advancing to a new process while A13 and A12 were both on 7nm.

This is not why I took from the recent anandtech review. The performance improvement appear linear (about +20% at each generation). The difference is that, while the A13 consumes more power than the A12, the A14 actually consumes less power than the A13.

As for raw single core performance, it seems that geekench is now dominated by Ryzen 5 CPUs under linux (not counting overclocked setups). Why are Linux scores that much higher than Windows scores (like 200 points higher)?

 

[coercitiv]

This is not why I took from the recent anandtech review. The performance improvement appear linear (about +20% at each generation). The difference is that, while the A13 consumes more power than the A12, the A14 actually consumes less power than the A13.

I believe @HurleyBird was talking about PPC in the context of increased density on 5nm. On A14 most of the gains came from higher clocks. That's why he's emphasizing Zen 4 vs. Firestorm successor, since we need to see the trends on each side.

 

[MarkizSchnitzel]

I don't think there will be a core truly that wide from x86 even then. Nor should there necessarily be, maintaining higher clocks might well be a better way to provide the performance.

The one big advantage ARM, especially 64-bit ARM has over x86 is that growing decode width grows it's power and complexity linearly. For ARM, 8-wide decode is ~ twice as power-hungry and large as a 4-wide decode. For x86, increasing decode width grows the complexity and power use of the decoders way faster than linearly, mainly because they have to build in massive muxes to line up instructions because instructions are wildly variable-width. This means that the ideal width of the machine from an engineering standpoint is much wider for an ARM machine than it is for x86. The width of M1 isn't free, they pay for it in many ways, including clock speed.

You seem like you know what you are talking about, or at least have a good general notion.
Would you be willing to educate a newb?

What are the pros and cons of either approach (width vs clocks) for a general usage in a consumer product?
By general usage, I mean, mostly browsing and using various webapps.


Also, if somebody wanted it, could they integrate the same amount and kinds of specialist silicon in a x86 processor? Like, for video and photo processing?
Asking, because somebody implied that these specialist silicone parts account for MOST, or at least MUCH of the efficiency and performance advantage M1 seems to have, yet, I am fairly certain that vast majority of normal consumers will never use there, ever, so not sure if there is incentive for Intel/AMD to even try this approach.

 

[Gideon]

Asking, because somebody implied that these specialist silicone parts account for MOST, or at least MUCH of the efficiency and performance advantage M1 seems to have, yet, I am fairly certain that vast majority of normal consumers will never use there, ever, so not sure if there is incentive for Intel/AMD to even try this approach.

According to leaks AMD is planning on using similar specialized silicon:

Van Gogh is a 4 Core APU that will release somewhere in 2021 (probably announced in January) it has a CVML block (computer vision / machine learning)

Spoiler: Leaked AMD roadmap

And all that special silicon has zero to do with SPEC or Geekbench results, that's not how hardware works, it's pure CPU perf.

 

[coercitiv]

What are the pros and cons of either approach (width vs clocks) for a general usage in a consumer product?
By general usage, I mean, mostly browsing and using various webapps.

From a consumer standpoint the clock approach equals lower price at ISO performance, the wide core approach equals lower power at ISO performance. However, that is assuming both approaches are handled by equally successful teams and have comparable diminishing returns. It also assumes a healthy, competitive market for the price part.

 

[MarkizSchnitzel]

From a consumer standpoint the clock approach equals lower price at ISO performance, the wide core approach equals lower power at ISO performance. However, that is assuming both approaches are handled by equally successful teams and have comparable diminishing returns. It also assumes a healthy, competitive market for the price part.

Thanks

So, are Macbooks in the same market? Would PC OEMs now give hell to AMD/Intel to copycat/emulate?
Apple is de facto a trendsetter in mobile space. It's definitely common market there. Not sure if it's the case for laptops.

According to leaks AMD is planning on using similar specialized silicon:

Van Gogh is a 4 Core APU that will release somewhere in 2021 (probably announced in January) it has a CVML block (computer vision / machine learning)

And all that special silicon has zero to do with SPEC or Geekbench results, that's not how hardware works, it's pure CPU perf.

Aha, ok.
I was thinking more of the performance claims apple made, like multiple times faster in task X and/or Y.

 

[mikegg]

There may not be much to expand to. You obviously won't get any nVidia GPUs, and even AMD probably won't bother making native drivers unless they get an OEM deal which doesn't seem likely. USB4 might be fast enough for storage.

I see AMD making drivers for Mac Pros and iMacs. I don't think Apple will abandon discrete GPUs completely and I don't think they will make their own discrete GPUs. Thus, to satisfy graphics-intensive workloads, they will partner with AMD.

 

[Eug]

Interesting.

Cinebench R23 by default now runs for 10 minutes to factor in potential throttling. However, there is now the option of a 30 minute thermal stress test as well. It will be interesting to compare the results of the new M1 MacBook Air in the three different test modes:

1. Old style single run
2. 10 minute default test
3. 30 minute stress test

 

[Gideon]

But regarding mac's and their walled garden, this is an example of a not-so-good approach:

Jeffrey Paul: Your Computer Isn't Yours

 

[coercitiv]

So, are Macbooks in the same market? Would PC OEMs now give hell to AMD/Intel to copycat/emulate?

It will always come down to price vs. performance equation. As long as AMD/Intel offer a competitive mix of perf/dollar and perf/watt, OEMs will be able to build competitive devices just like smartphone makers are still competing with Apple. And I have a feeling Microsoft won't wait to see if AMD/Intel can pull it off and double down on getting ARM onboard as well.

 

[mikegg]

It will always come down to price vs. performance equation. As long as AMD/Intel offer a competitive mix of perf/dollar and perf/watt, OEMs will be able to build competitive devices just like smartphone makers are still competing with Apple. And I have a feeling Microsoft won't wait to see if AMD/Intel can pull it off and double down on getting ARM onboard as well.

I wrote this before:

A lot of buyers want a Mac but Intel Macs have been out of reach for them due to the price. With ARM Apple can finally start to penetrate this market, using the same strategy as the iPhone SE.

People are and should be comfortable with owning a Mac. iPhone owns ~40-50% of the U.S. market share. iPad 65%. They're already familiar with Apple's designs and walled garden. Up until now, Apple couldn't make a sub $1,000 laptop that wasn't crap (said by Steve Jobs himself). They will be able to with ARM.

I think Apple is going to increase their laptop/desktop market share from ~10% today to ~20-30% in the next few years. I'd buy Apple stocks now.

I think the ARM switch is going to be THAT impactful.

Imagine a year from now, Apple releases the M2 chip. Then they reuse the M1 chip for a $700 Macbook Air, ala iPhone SE strategy. This $700 laptop will still be significantly faster, more portable, and have a much longer battery life than any $700 Windows PC.

Price has always been the limiting factor for Macbook adoption. People who have iPhones want a Macbook to stay in the ecosystem but Macbooks couldn't compete with the garbage $500 PCs at the lowerend. Having their own chips and not having to pay Intel 30% of the cost of a laptop will allow Apple to make cheaper, and better laptops.

Remember, Apple is transitioning into a Service-oriented company. They want to sell you subscriptions to News+, Music, Arcade, Fitness, iCloud, TV+, etc. The more Apple devices you own, the more likely you are to also buy their services and the longer you stay in their ecosystem. To do this, Apple needs to sell cheap versions of their iPhone, iPad, and Macs.

 

[coercitiv]

Apple M1 vs Intel 6-core & AMD 580X in Affinity Photo:



original source

 

[Carfax83]

This is supposedly an Apple A12z dev kit in Cinebench R23:

 

[coercitiv]

Imagine a year from now, Apple releases the M2 chip. Then they reuse the M1 chip for a $700 Macbook Air, ala iPhone SE strategy. This $700 laptop will still be significantly faster, more portable, and have a much longer battery life than any $700 Windows PC.

Yeah, I'm sure it's the expensive dual-core preventing Apple from making this Mac cheaper. At least they gave up on selling it with a spinner instead.

Spoiler

 

[mikegg]

Yeah, I'm sure it's the expensive dual-core preventing Apple from making this Mac cheaper. At least they gave up on selling it with a spinner instead.

Spoiler

View attachment 33719

Yeah, I'm sure it's the expensive dual-core preventing Apple from making this Mac cheaper. At least they gave up on selling it with a spinner instead.

Spoiler

View attachment 33719

I don't understand your post. Are you agreeing with me? Disagreeing? Why or why not? Put a little more effort into your post please.

 

[Carfax83]

Apple M1 vs Intel 6-core & AMD 580X in Affinity Photo

Why is the vector performance so low on the intel CPU?

I've never heard of Affinity Photo. Is that benchmark constrained to 128 bit vectors or something?

 

[MarkizSchnitzel]

I wrote this before:



I think Apple is going to increase their laptop/desktop market share from ~10% today to ~20-30% in the next few years. I'd buy Apple stocks now.

I think the ARM switch is going to be THAT impactful.

Imagine a year from now, Apple releases the M2 chip. Then they reuse the M1 chip for a $700 Macbook Air, ala iPhone SE strategy. This $700 laptop will still be significantly faster, more portable, and have a much longer battery life than any $700 Windows PC.

Price has always been the limiting factor for Macbook adoption. People who have iPhones want a Macbook to stay in the ecosystem but Macbooks couldn't compete with the garbage $500 PCs at the lowerend. Having their own chips and not having to pay Intel 30% of the cost of a laptop will allow Apple to make cheaper, and better laptops.

Remember, Apple is transitioning into a Service-oriented company. They want to sell you subscriptions to News+, Music, Arcade, Fitness, iCloud, TV+, etc. The more Apple devices you own, the more likely you are to also buy their services and the longer you stay in their ecosystem. To do this, Apple needs to sell cheap versions of their iPhone, iPad, and Macs.

You have articulated well the exact reason why I, a layman and a newb, have posted here in the 1st place.
I despise walled garden approach, and apple by extension, with every fiber of my being.
That said, if most of what is written on this chip and battery life pans out, and the price does drop to what you suspect it might (and the service based model does seem plausible), it would be stupid NOT to get this. Especially since I just need a browser and a file manager. I suspect many consumers are like this.

 

[coercitiv]

I don't understand your post. Are you agreeing with me? Disagreeing? Why or why not? Put a little more effort into your post please.

I am disagreeing, as it should have been transparent from the $1100 price tag they put on an AIO with an antiquated 2c CPU, 8GB of 2133mHZ RAM, 256GB of storage and FHD screen. The price of this AIO is not about BOM alone.

The $700 Windows laptop you mention comes with a decent FHD IPS screen, 16GB of RAM, 512GB SSD and a 6 core CPU. The equivalent on the MBA side right now is not the $1000 part, but the $1400 part with upgraded RAM and storage. It's a beautiful, premium device with excellent performance and autonomy, but it's not just $200-300 more expensive relative to the current offering in the PC market. It's 100% more expensive.

 

[Qwertilot]

You have articulated well the exact reason why I, a layman and a newb, have posted here in the 1st place.
I despise walled garden approach, and apple by extension, with every fiber of my being.
That said, if most of what is written on this chip and battery life pans out, and the price does drop to what you suspect it might (and the service based model does seem plausible), it would be stupid NOT to get this. Especially since I just need a browser and a file manager. I suspect many consumers are like this.

The macs aren't turning walled garden as such. There's two distinct things going on with Apple's edge.

The first is sheer resources - Apple have more and have used them very well with CPU design etc. Even if the technology was equalised they'd probably still be at least a bit ahead.

The second, directly answering your question about putting NPU units in, is about software/ecosystem. Yes people could but there's no point putting in new technology like this unless people will actually use it.

With the Apple sillicon, every Mac is getting an NPU, the operating system is built to expect it, the compilers are build to use it etc etc. So it'll just work and have a considerable impact. Ditto with GPU compute offload. That requires the developers to be on board to some degree, but Apple do seem to have done that rather better over time.

The PC/Windows ecosystem seems to suffer from massive software inertia. AMD say tried for ages to push GPU compute offload in their PC APU's, to very minimal real world effect. Even Intel's 'new' instruction sets like AVX512 have been rather crippled in terms of support because they're only supported by a few chips.

For a long time that really wasn't that important a problem as they were way ahead but with these Apple silicon chips being - quite frankly - embarrasingly superior? You really do have to wonder.