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

 

[jeanlain]

Of course some would say that they are different platform and they only need to beat AMD, but I disagree.

So do intel, judging from their recent ad campaign against Macs.

 

[nxre]

Gotta wonder how they are going to compete in the mobile space with those numbers. Desktop space not so important for power, especially as we don't know how the M1 will scale up.

Of course some would say that they are different platform and they only need to beat AMD, but I disagree. There is enough crossover to say they are in direct competition.

I'd argue their biggest threat isn't even Apple, but the upcoming Qualcomm-Nuvia chips. If the claims are to be believed, thats's Apple like performance and efficiency for Windows laptops. Intel's x86 compatibility card won't work in keeping the ARM competition away forever.
So, one way or another, Intel will have to compete with Apple-like designs in the very near future.

 

[Eug]

I'd argue their biggest threat isn't even Apple, but the upcoming Qualcomm-Nuvia chips. If the claims are to be believed, thats's Apple like performance and efficiency for Windows laptops. Intel's x86 compatibility card won't work in keeping the ARM competition away forever.
So, one way or another, Intel will have to compete with Apple-like designs in the very near future.

Are we assuming only native ARM software for that "Apple-like performance"? Last year the Windows app library for ARM was pitifully bad, and translated x86 apps would likely be very, very slow on those chips, no?

That would be a big difference vs. Apple's M1 series.

https://twitter.com/x/status/1331736203402547201

 

[Hitman928]

Yes, I used those numbers.

What numbers did you use for Firestorm?

From Anandtech's M1max article, it uses very similar package power in ST loads as the 'E' cores quoted for ADL.

 

[scannall]

Are we assuming only native ARM software for that "Apple-like performance"? Last year the Windows app library for ARM was pitifully bad, and translated x86 apps would likely be very, very slow on those chips, no?

That would be a big difference vs. Apple's M1 series.

https://twitter.com/x/status/1331736203402547201

Apple has said the only thing keeping Windows off of M1 macs is Microsoft coming up with licensing for WARM.

 

[Roland00Address]

Some fun comparisons based on Anandtech's data

Alder Lake p-core is ~10% faster than Firestorm in SpecInt at 5x the power.
Alder Lake e-core is ~40% slower than Firestorm in SpecInt at 2x the power.

Seems like Apple really made the right choice by designing their own silicon, the efficiency gap is going nowhere.

I am curious how these chips will perform at different voltages targeting the 5w and 15w tdps area (and likely an in between area with ctdp up.). They will be slower but also more efficient and use less power.

Meanwhile M1Max or Pro hardware will be faster and likely consume the same or less power.

 

[nxre]

What numbers did you use for Firestorm?

From Anandtech's M1max article, it uses very similar package power in ST loads as the 'E' cores quoted for ADL.

I used 5W for Firestorm, but now that you point it out I was comparing core complex power on the M1 vs Package Power on ADL, didn't notice it, sorry.
I can't seem to find the core complex only power for Alder Lake, but I imagine it must be 3x-4x higher for p-core and 1x-1,5x higher for e-core instead of the 5x and 2x, but I can't be sure.

 

[nxre]

Are we assuming only native ARM software for that "Apple-like performance"? Last year the Windows app library for ARM was pitifully bad, and translated x86 apps would likely be very, very slow on those chips, no?

That would be a big difference vs. Apple's M1 series.

https://twitter.com/x/status/1331736203402547201

This is a good point, but I imagine Microsoft will put more effort in their x86 translation given they seem to want their own surface devices to run on ARM processors.

 

[dmens]

I used 5W for Firestorm, but now that you point it out I was comparing core complex power on the M1 vs Package Power on ADL, didn't notice it, sorry.
I can't seem to find the core complex only power for Alder Lake, but I imagine it must be 3x-4x higher for p-core and 1x-1,5x higher for e-core instead of the 5x and 2x, but I can't be sure.

Look at the M1 review if you want to isolate the single core power, Pro/Max have much more significant power overhead with the other portions of the SoC. 5-6W is correct for 1C max. All core loading will be lower per core.

 

[Doug S]

Apple has said the only thing keeping Windows off of M1 macs is Microsoft coming up with licensing for WARM.

It is strange that Microsoft isn't interested in licensing Windows/ARM but that's not the only thing keeping it off M1 Macs. There are no Windows drivers for the Mac hardware, in particular the GPU. The M1 Linux people will eventually solve that problem, and those drivers could be ported to Windows. Once they reverse engineer a decent accelerated M1 GPU driver, in addition to porting that to Windows someone would have to write a DX -> Metal translation layer.

Between whatever performance is lost through the reverse engineering missing a few tricks, that DX translation overhead, and the Apple GPUs using deferred rendering instead of the immediate rendering DirectX assumes, the GPU performance on Windows would not live up to the relative performance of Apple's CPU cores. So I doubt they'd be a big threat to x86, at least not for people who expect GPU performance comparable to what they can get in the Windows x86 market.

Even if Qualcomm's Nuvia core does like M1 and delivers CPU performance that beats x86 (especially at a given power level) and has comparable GPU performance (by using the same AMD/NVidia GPUs x86 laptops use) that's no guarantee it would be successful. People complain about the lack of games on macOS, well there aren't exactly a lot of Windows/ARM ports of popular PC games either. Any x86 emulation on Qualcomm's Nuvia cores will be much slower than Rosetta 2, since using a similar technology is simply not a viable option for Windows.

Heck, Microsoft hasn't even ported all of its own Windows software, which doesn't indicate that much of a commitment to the platform. They've abandoned all five non-x86 Windows ports in the past (including the ARM port from the Windows 8 days) so developers are rightly reticent about devoting much time to porting or especially to optimizing the ports they do ship.

 

[gdansk]

After reading a few reviews I've decided on a base MBP 16. I think the 8+2/16 M1 Pro is overkill for what I need. A 16" with the regular M1 would be fine but they don't make it. Maybe they'll make a big MacBook Air in the future.

 

[Geegeeoh]

They've abandoned all five non-x86 Windows ports in the past (including the ARM port from the Windows 8 days) so developers are rightly reticent about devoting much time to porting or especially to optimizing the ports they do ship.

But back then x86 was just the fastest platform.

The one thing going for x86, for "normal consumers", is that it's just good enough anyway.

 

[Shivansps]

I'd argue their biggest threat isn't even Apple, but the upcoming Qualcomm-Nuvia chips. If the claims are to be believed, thats's Apple like performance and efficiency for Windows laptops. Intel's x86 compatibility card won't work in keeping the ARM competition away forever.
So, one way or another, Intel will have to compete with Apple-like designs in the very near future.

It already doesnt matter. ARM Windows is already a valid platform for everyday use, other than a perf hit for running x86-64 executables, eveything is the same.

You would only hit a wall with device drivers.

The only major block right now is Microsoft itself. Try to get a license and a normal installer for ARM Windows for Raspberry PI4... you just cant. Thats also the reason why they dont develop a VC4 gpu driver for ARM64 Windows.

The other problem is the hardware that is not up to the task, Apple may be beating both Intel and AMD HARD, but they also completely destroy ARM cpus avalible on the market even more. Im looking forward to future hardware, but normal ARM cpus have a problem called "small core" that is not suited for Windows (in order), Apple doesnt have this problem. So ARM can only go "big.Bigger" for notebooks and desktops, what would not be a issue until you realise that they actually need to make a silicion that targets notebooks and desktops only.

 

[Eug]

M1 Pro First Impressions: 2 Core management and CPU performance

He has some interesting graphs and commentary in this article. It seems Apple may be managing power utilization on M1 Pro/Max more actively than some may have expected.

Taken together, these results show that process allocation to cores in the M1 Pro and Max is carefully managed according to QoS (as in the M1) and between the two groups of P cores. This management aims to keep the second group of P cores unloaded as much as possible, and within each group of P cores loads lower-numbered cores more than higher-numbered. This is very different from the even-balancing seen in symmetric cores, and in the M1.

The end result is that the two E cores in the M1 Pro/Max are significantly faster (in some respects, at least) than the four E cores in the M1, although the E (but not the P) cores are slowed when running on battery alone.

Because of this sophisticated asymmetric core management, measuring CPU performance in the M1 Pro/Max is more complex than when cores are managed symmetrically. While running on battery alone shouldn’t impair the performance of CPU-bound tasks run at higher QoS, you should expect background services run on the E cores alone to take longer.

 

[Heartbreaker]

It is strange that Microsoft isn't interested in licensing Windows/ARM but that's not the only thing keeping it off M1 Macs. There are no Windows drivers for the Mac hardware, in particular the GPU. The M1 Linux people will eventually solve that problem, and those drivers could be ported to Windows. Once they reverse engineer a decent accelerated M1 GPU driver, in addition to porting that to Windows someone would have to write a DX -> Metal translation layer.

Note when Apple said that it was up to Microsoft they were NOT talking about running Windows on the bare Metal of M1 Macs. They were talking about Running Windows in a VM.

It has been demonstrated that Windows 10/11 of WOA, works fine in M1 VMs, so, the only thing keeping Windows from running legally on M1 VMs is a license from Microsoft, something Microsoft seems totally against.

Microsoft Says No to WOA Support on M1 Macs

Microsoft has finally gone on record about whether it will support Windows on Arm on Apple’s M1 Macs. And it’s not good news.

www.thurrott.com

It may not be so strange that Microsoft is blocking this, as it makes Microsofts own WOA machines look weak in comparison. The designed for Windows ARM chip, running native on the metal, can't come close to the performance of WOA running in a VM on a M1 Mac.

 

[Eug]

The Information: Apple's Road Map for Mac Chips Shows Likely Advantage Over Intel

It's behind a paywall so I don't know how much of this is supposed to be a true leak or if a lot of it is just conjecture.

MacRumors summary

- Mac Pro will use M1 Max derivative with at least 2 dies.
- M2 Pro/Max chips will be 2 dies with more cores, on 5+ nm.
- M3 series will be up to 4 dies and up 40 GPU units (presumably M3 Pro/Max), some on 3 nm. Code names Ibiza, Lobos, Palma.

If true, does this mean there will be a bigger performance delta between M3 MacBook Air (1 die) and M3 Max MacBook Pro (4 dies)? I'm thinking no, and perhaps this would mean the iPhone and iPad would get 1 die (Ibiza), iPad Pro and MacBook Air would get 2 dies (Lobos), and MacBook Pro would get 4 dies (Palma).

Note: I am posting this from an m3 Mac.

 

[Eug]

Here is a different type of benchmark:

https://twitter.com/x/status/1455971162060697613

Also, MKBHD has yet another type of benchmark. He actually used to lug around an iMac Pro to industry events for editing his YouTube videos. He had a industrial strength iMac Pro foam hard case and everything. Now that has been retired as he has replaced it with an M1 Max 16", which BTW, outperforms his 28-core Intel Mac Pro with Afterburner with his workflow.



That should say M1 Max at the bottom. Note that those MBP times are on battery. Plugged-in and on high performance mode, he got 3'49" on the M1 Max MBP.

 

[Heartbreaker]

I think he also called it a $50K Mac Pro.

 

[Schmide]

Between whatever performance is lost through the reverse engineering missing a few tricks, that DX translation overhead, and the Apple GPUs using deferred rendering instead of the immediate rendering DirectX assumes, the GPU performance on Windows would not live up to the relative performance of Apple's CPU cores. So I doubt they'd be a big threat to x86, at least not for people who expect GPU performance comparable to what they can get in the Windows x86 market.

Immediate mode (forward) is just a mode in DirectX, or any other renderer, and assumes nothing. It's just the old style of begin, draw something, end. Deferred just means the renderer is more flexible in how it schedules the rendering. Other than very simple and old graphics everything uses deferred these days.

Apples CPUs are by far powerful enough to drive any GPU so relative performance is moot.

The only thing keeping Apple from being a threat to Windows GPU performance is Apple. Then again they wouldn't be Apple if they were flexible enough to open up.

 

[Geegeeoh]

Here is a different type of benchmark:

https://twitter.com/x/status/1455971162060697613

Oh c'mon, the whole workflow won't "scale" with the build times.

 

[Doug S]

The Information: Apple's Road Map for Mac Chips Shows Likely Advantage Over Intel

It's behind a paywall so I don't know how much of this is supposed to be a true leak or if a lot of it is just conjecture.

MacRumors summary

- Mac Pro will use M1 Max derivative with at least 2 dies.
- M2 Pro/Max chips will be 2 dies with more cores, on 5+ nm.
- M3 series will be up to 4 dies and up 40 GPU units (presumably M3 Pro/Max), some on 3 nm. Code names Ibiza, Lobos, Palma.

If true, does this mean there will be a bigger performance delta between M3 MacBook Air (1 die) and M3 Max MacBook Pro (4 dies)? I'm thinking no, and perhaps this would mean the iPhone and iPad would get 1 die (Ibiza), iPad Pro and MacBook Air would get 2 dies (Lobos), and MacBook Pro would get 4 dies (Palma).

Note: I am posting this from an m3 Mac.

I take that to mean that the Mac Pro we'll get next year will be the rumored "compact" Mac Pro, and it will max out at 16 big cores, 4 little cores, and 64 GPU cores.

That would explain the need for next year's Ice Lake based Mac Pro, to give the high end model a boost while waiting for the M3. If these rumors are true, then the true "high end" ARM Mac Pro that replaces that Ice Lake Mac Pro won't appear until the M3 in a couple years or so. With all the new transistors N3 will allow I'd be surprised if they don't have more GPU and/or big cores, but who knows maybe they'll use most of that new area to grow the L2 and SLC caches instead.

This does NOT mean you will see the multichip versions in a MacBook Pro - that's just not workable in that form factor either size wise or cooling/power budget wise.

The claim the Mac Pro will use an M1 Max "derivative" suggests the M1 Max die is itself a "chop" of an even larger die, which includes the off chip communication necessary to support two chips talking to each other. Maybe a DDR5 controller but I'm skeptical the "low end" model will support DIMMs, so if it comes without DIMMs to me that doesn't rule out the high end M3 based model having them later.

 

[Heartbreaker]

The Information: Apple's Road Map for Mac Chips Shows Likely Advantage Over Intel

It's behind a paywall so I don't know how much of this is supposed to be a true leak or if a lot of it is just conjecture.

Claims about 3rd generation series of chips at this point seem very dubious. The Mac Pro stuff just seems to be repeating what Gurman said long ago.

 

[Doug S]

Claims about 3rd generation series of chips at this point seem very dubious. The Mac Pro stuff just seems to be repeating what Gurman said long ago.

I take all Apple rumors with a grain of salt since they do their best to keep it all quiet.

The third generation chips are likely near tapeout if they aren't already, given that N3 is entering risk production right about now. So they are a known quantity, it is just a question of whether The Information found someone in a position to know that and willing to leak it.

 

[gdansk]

Oh c'mon, the whole workflow won't "scale" with the build times.

Truly. If compiles are faster I'm not going to work much more.
I'm sure there's an xkcd about it.

 

[Hitman928]

Truly. If compiles are faster I'm not going to work much more.
I'm sure there's an xkcd about it.

But of course:

Compiling

xkcd.com