Discussion Apple Silicon SoC thread

Page 101 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Eug

Lifer
Mar 11, 2000
23,584
998
126
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:

Screen-Shot-2021-10-18-at-1.20.47-PM.jpg

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:


M1 Ultra discussion here:


M2 discussion here:


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:

 
Last edited:

The Hardcard

Member
Oct 19, 2021
46
38
51
Not nearly as well as can dedicated hardware.

The dedicated hardware is huge, but there are already several hours on Youtube of both new chips crushing software codecs. These machines would have dominated the video editing market without the dedicated hardware. They are easily handling Canon, Blackmagic, and RED RAW footage on the CPU cores.

The ProRes hardware are just nails in the coffin. I’ll bet the Afterburner card was the result of working on the IP blocks for the M1s.
 

ashFTW

Senior member
Sep 21, 2020
306
229
96
If I was in the market for a Mac laptop I'd be wishing for a Macbook Air with a larger screen. I've always had laptops with a 17" screen, because I like a lot of real estate, but performance (especially GPU) isn't all that much of a consideration for me, nor is battery life since I hardly ever go more than a few hours without plugging it in. I care more about max memory config since I'd be running Linux on it with a Windows VM or two. I have zero interest in the DTR or gamer bricks, so usually there isn't much choice.

My ideal Macbook Pro given the power budget would have been 16 + 4 cores, and 8 GPU cores. CPU, large memory, and fast SSDs are much more important to me than GPU. That’s the reason I love my Threadripper workstation.
 
  • Like
Reactions: Arkaign

Eug

Lifer
Mar 11, 2000
23,584
998
126
iFixit full teardown:


SSD, HDMI, SD reader are soldered.

Headphone jack, USB-C, MagSafe are replaceable. However, since they are screwed into the chassis underneath the motherboard, you have to remove the motherboard first before you can access the screws for those ports.

As mentioned previously, the batteries are not glued on, but attached with iPhone-style pull tab adhesive tape. However, for two of the six battery segments you have to remove the trackpad to access their pull tabs.

Speakers are glued on, presumably because they are in constant motion, but you don't need to remove them to access the other components.
 

moinmoin

Diamond Member
Jun 1, 2017
4,944
7,655
136
No one admonishes Intel for offer AVX-512 support
*coughs*
I don't admonish Intel for the speed up AVX-512 can offer. I totally admonish Intel for the crap instruction set AVX-512 itself is, reinventing many parts of existing instructions, only to again stopping short of future proofing it for eventual further extensions where the whole spiel repeats once again.

Apple by standardizing hardware support (something Intel irritatingly doesn't manage with AVX-512 even in its very own products) does it in a sane way, making hardware acceleration easily usable through system libraries.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
While I completely expected the SSD to be soldered, it's a hard kill for our organization. I have had to replace multiple SSDs for power users both for filling them up and for actually hitting endurance limits with them. These were perfectly serviceable laptops that didn't warrant replacing. Now, and for the last several models, Apple wants me to dispose of or gut and pay a fortune to service perfectly otherwise usable equipment because they couldn't find it in their pocketbook to allow the use of M.2 drives? The marginal improvement they are possibly realizing from soldering is definitely not worth the cost it represents us.

No thank you!
 

Eug

Lifer
Mar 11, 2000
23,584
998
126
While I completely expected the SSD to be soldered, it's a hard kill for our organization. I have had to replace multiple SSDs for power users both for filling them up and for actually hitting endurance limits with them. These were perfectly serviceable laptops that didn't warrant replacing. Now, and for the last several models, Apple wants me to dispose of or gut and pay a fortune to service perfectly otherwise usable equipment because they couldn't find it in their pocketbook to allow the use of M.2 drives? The marginal improvement they are possibly realizing from soldering is definitely not worth the cost it represents us.
Fair enough, but I'm curious though, which laptops? Yes, all grades of memory and SSDs can die, but FWIW, Apple uses top tier quality memory and SSDs. (speed ≠ quality)

What I often see is people not using the same grade of memory that we find in Macs. Not saying this is your organization of course, but it still floors me that many small video editing outfits and individual consumers view OWC memory as top tier memory. This is oft repeated out there in internetland and I'm always just like WTF?!?
 
  • Like
Reactions: Viknet

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
Soldering the SSD is a stupid decision driven by greed. They don't want people getting the 512GB model and then upgrading it on the cheap.

They are using a custom mac chip to act as the memory controller, and it is connected by pci express. How would it work with multiple memory controllers if you could plug in a SSD?

What you want is a 2nd or 3rd PCI Express slot that does not use the apple memory controller.
 
Last edited:

Eug

Lifer
Mar 11, 2000
23,584
998
126
They are using a custom mac chip to act as the memory controller, and it is connected by pci express. How would it work with multiple memory controllers if you could plug in a SSD.

What you want is a 2nd or 3rd PCI Express slot that does not use the apple memory controller.
Yes, this is a good point. This keeps the controller in-house. Not only does this save money and reduce space used, it also reduces dependency on third party suppliers, and even more importantly, it keeps the base hardware more consistent even across different model lines. This reduces the chance of weird incompatibilities with third party peripherals and third party software, as well as with Apple's own peripherals, OS, and software.

Don't get me wrong though, from a personal point of view, I still prefer user upgradable SSDs. However, I think it is important to realize that this soldered SSD thing isn't just about money. As you point out, there are technical reasons to do this as well.
 

The Hardcard

Member
Oct 19, 2021
46
38
51
Yes, this is a good point. This keeps the controller in-house. Not only does this save money and reduce space used, it also reduces dependency on third party suppliers, and even more importantly, it keeps the base hardware more consistent even across different model lines. This reduces the chance of weird incompatibilities with third party peripherals and third party software, as well as with Apple's own peripherals, OS, and software.

Don't get me wrong though, from a personal point of view, I still prefer user upgradable SSDs. However, I think it is important to realize that this soldered SSD thing isn't just about money. As you point out, there are technical reasons to do this as well.

Those technical reasons include the packaging itself. For all the talk about architecture and process nodes , there going to need to be more investigation into how much the M1 Pro/Max package is contributing to both the speed and efficiency.

This is only going to accelerate. I wouldn’t be surprised at some point to see the storage is also part of the processor package. When a transistor is 2 nm wide, the width of a finger is really far away.

I think laptops and other small devices will be soon be faster than desktops, because no amount of high-wattage will catch up to a processor with a GPU less than 5 millimeters from the cores, RAM less than 10 mm away, and storage less than 15 mm away.

EDIT: hopefully a little more coherent.
 
Last edited:
  • Like
Reactions: podspi and Eug

jpiniero

Lifer
Oct 1, 2010
14,562
5,194
136
While I completely expected the SSD to be soldered, it's a hard kill for our organization. I have had to replace multiple SSDs for power users both for filling them up and for actually hitting endurance limits with them. These were perfectly serviceable laptops that didn't warrant replacing.

It's obviously not hurting sales. Last time the Macbook Pro had upgradable ram was 2012, and technically the 2016 and 2017 base 13" had upgradable SSD but doesn't sound like it was meant to be done by end users.

And yes people buy third party ram/ssd to avoid paying the OEM prices.
 

jamescox

Senior member
Nov 11, 2009
637
1,103
136
Yes, this is a good point. This keeps the controller in-house. Not only does this save money and reduce space used, it also reduces dependency on third party suppliers, and even more importantly, it keeps the base hardware more consistent even across different model lines. This reduces the chance of weird incompatibilities with third party peripherals and third party software, as well as with Apple's own peripherals, OS, and software.

Don't get me wrong though, from a personal point of view, I still prefer user upgradable SSDs. However, I think it is important to realize that this soldered SSD thing isn't just about money. As you point out, there are technical reasons to do this as well.
The soldered SSD is a big reason for me to not want to buy one of these. One reason that they solder it that doesn’t seem to have been brought up is manufacturability. They solder just about everything they can onto the mainboard. That is likely much cheaper than a separate card that needs to be installed by a human. With it on the mainboard it will be placed by a pick and place machine with no human worker needed. They were doing the gluing everything for similar reasons. Screws are a lot harder to put in by a machine than just having a machine squirt glue. I wonder if the screws are placed by machines now also.
 

jamescox

Senior member
Nov 11, 2009
637
1,103
136
Even just 4 channels of LPDDR5 would be plenty. If there are enough CUs just add in some sort of SLC/Infinity cache to memory controllers on die (like on the M1) and you could easily performance way above 3050 Ti.

If you're worried about pins, just put the memory on package like apple does.

Considering the mining craze, current GPU prices and the fact that Apple does huge integrated graphics now - OEMs should come around soon-enough for the amounts of money they can save alone.

Yeah that premium SKU might cost a bit and might not sell in crazy-volumes. But considering that it can be assembled from existing blocks and AMD makes more in a quarter they made in the whole 2016 (and Q4 will probably rival 2017), it should be worth it for mindshare alone. I mean, just look how relatively small the 128-bit memory controller and 8MB of extra cache is, it shouldn't be all that big of a problem:

Mf2IBG4.png
Adding some extra LPDDR5 as Apple did would obviously work, but an APU (possibly built out of modular / stacked blocks for cost efficiency vs. Apple massive, expensive, monolithic die) plus 1 or 2 stacks of HBM would be able to provide spectacular performance with exceptional power efficiency. Apple seems like they are getting far enough ahead that Windows laptop makers are going to need to do something. Apple still kind of caters to more of a niche market, but I see more and more people just using their iPhone or iPad for almost everything or using the cheaper Mac laptops.

The reason the Apple “isn’t good for games” isn’t because the hardware isn’t capable of gaming. It is still the case that games require a lot of optimization for the specific hardware to achieve good performance. Apple probably isn’t paying anyone to do that optimization. They would much rather sell to the much higher profit margin professional market. They don’t want to make cheap gaming machines that would dilute their brand. The 16 inch m1 pro based version starts at $2500 which is a bit pricy for gaming.

The things that make me lean towards buying a MacBook Pro is the fact that I am still using a 17 inch MacBook Pro from 2009 and a MacBook Air from 2012. That isn’t uncommon. It turns out that a lot of developers from my work are still using ancient MacBooks. While the everything soldered on one board solution isn’t easily reparable, it is likely very reliable. It is probably significantly less likely to have anything go wrong compared to more modular solutions. The battery is still the most likely thing to need replaced. I have been through quite a few batteries. Not having a replaceable ssd is a concern, but they can be made very durable. I have an Intel DC P3700 from about 2013 that I have beat on for years across multiple machines and it still has something like 95 percent write cycles left. It is a data center part, so using it for consumer means it will likely be completely obsolete before it actually wears out. Another thing that makes me want the MacBook Pro is the screen. That is probably well above anything else on the market, but is it worth the cost? Most of the other components are also top notch. I have used a few cheap windows laptops and have been astonished at how bad the track pads and such are. I do want a bigger laptop, so hopefully framework can come out with a 16 inch size in the next year or so. I wouldn’t mind waiting for the next revision of the m1 if I am going to go the MacBook Pro route.
 

Eug

Lifer
Mar 11, 2000
23,584
998
126
^^^ Speaking of gaming on Apple Silicon...


Screen Shot 2021-10-29 at 11.38.31 PM.png

This is a summary graph of the benchmarks, but for each test he went into more detail in the video. Most of the settings he used were more GPU intensive than CPU intensive. As you can see, there is fairly decent scaling as you increased the number of GPU cores, although in some tests the scaling wasn't as good on the 32-core GPU.

Overall, as expected, gaming wasn't very good. OTOH, if you really want to game on your Apple Silicon Mac, having the M1 Max sometimes was the difference between playable vs. annoying, at least with these more GPU-intensive settings. And often times, the M1 non-Pro was just useless.

However, one interesting thing he said what that in some games he was actually getting much faster frame rates through Crossover than the actual Mac ports. As in 50% faster.
 
Last edited:

StinkyPinky

Diamond Member
Jul 6, 2002
6,762
783
126
Yeah I had two 13" MB Pro's from 2013 that were still running nicely. THey are built to last that's for sure.

Anyway, I picked up the M1 Max 16" version with 32gb ram, it is connected to two 4K monitors, the primary of which can do HDR. Butter smooth so far as you'd expect. I have a gaming desktop so not really gonna game on this, but for interests sake I fired up Divinity Original Sin 2 (rosetta) and ran it unscaled at 4K on Ultra and it runs just fine. This game is a few years old so it's not current gen but for the record my 16" inch 2019 macbook I had with the 5500M mobile GPU had a real hard time at 1440p with this game.
 

DrMrLordX

Lifer
Apr 27, 2000
21,600
10,792
136
The only complaints I've seen are that it uses a lot of power

. . . seriously?

Until IceLake, all AVX-512 implementations featured lower multipliers that would be inflicted on every core of a CPU even if only one core were handling AVX-512 instructions. Anyone using AVX-512 on a Skylake-SP CPU (for example) on a single core would nuke the performance on every core, even when the CPU was hosting multiple VMs on different cores. Very annoying! I think Intel managed to fix that on IceLake/RocketLake and TigerLake but i'm not 100% sure! Some lesser variant of that bogeyman may still exist.

The amount of re-engineering necessary to accommodate AVX-512 lead to massive amounts of die area being allocated just to AVX-512 support. The caches all had to be realigned for AVX-512 support. The less-often you used it, the more it would negatively affect the overall design of the chip.

And yes, the points @moinmoin brought up were also valid.

I don't want to thread derail, but you did mention AVX-512 complaints, so there they are. AVX-512 has been a mess. Something like SVE2 will work out much better. Hopefully Apple will choose to embrace SVE2 eventually.

Which would be a problem if the base CPU sucked but it's also great.

Video encoding is probably one of the biggest resource consumers out there for both pros and home users. Having excellent video encoder HW is a big win for a large segment of the buying population.

Like I said, it's great until it isn't. One switch in algo and then you can't use it anymore. It's planned obsolescence unless you think that people will stay with the same codec forever. Let's face it, the dedicated hardware will always be faster and more power-efficient where applicable.

They are easily handling Canon, Blackmagic, and RED RAW footage on the CPU cores.

Please link those videos, or better yet text articles with that data?
 

jeanlain

Member
Oct 26, 2020
149
122
86
However, one interesting thing he said what that in some games he was actually getting much faster frame rates through Crossover than the actual Mac ports. As in 50% faster.
For Sleeping Dogs, to be specific. I think this is an openGL game. But even then, being faster under emulation is crazy. I suppose there's a bug somewhere that reduces frame rate and saturate one CPU core. See how performance doesn't change much going from 16 to 32 GPU cores.
 
  • Like
Reactions: Eug

TESKATLIPOKA

Platinum Member
May 1, 2020
2,348
2,843
106
And here I thought people frequenting this website would know better than to equate package power with TDP. Unless the reviewer explicitly monitored package power consumption, it is not possible to compare the two given how different laptop OEMs will calibrate profiles for the same chip.

Case in point, the 5900HS has a 35w TDP, which Asus (the only OEM w/ a 5980HS laptop) happily ignored in the implementation with power profiles that draw 65w & 45w from the CPU, respectively. https://www.ultrabookreview.com/46631-asus-zephyrus-g14-2021-review/
I frequent that site, so I knew that.
It would be pretty surprising if Tim Schiesser who did these tests didn't know this as well.

In the review you posted ROG Zephyrus G14 GA401QM R9-5900HS + 3060 has 4 power profiles for CPU.
Turbo 65W
Performance 45W
Performance Battery 35W
Silent 25W
The author wrote he tested that laptop with Turbo profile(65W), the score was in Cinebench was 12865 points.

In the techspot review I posted the scores are:
R9 5900HX 11885 points for 45W
R9 5980HS 11024 points for 35W
From this It's pretty clear they had a lower power budget and considering R23 is by default at least a 10 minutes render loop, the author also used the default settings, so these scores should be from a sustained power draw.

At the end Apple did a very good job, efficiency is not many times better like some here believe, at least not in this bench, but this efficiency difference is also not only due to a better process.
I am eagerly waiting for notebookcheck and how much power does M1 MAX draw during gaming.
 
Last edited:

eek2121

Platinum Member
Aug 2, 2005
2,929
4,000
136
And you can add the Screen, Speakers, and microphones to that as well. In almost every dimension that matters, these are the best laptops you can buy.

Gaming being a notable exception. Don't buy a Mac for gaming rule, still holds.

I never owned an Apple product, or a Laptop, but these laptops are so impressive, it makes me want to, but they are so expensive that I won't.
Well it is important to note that gaming is not the only area where ARM based Macs are lacking. The lack of server chips and (high performance) cloud based servers means that Apple is putting developers in a bad spot.

With Intel, you could just run stuff inside a docker container on the same Linux distro as production, and if was generally pretty easy to debug as well as do performance profiling. With Apple chips using ARM, you are now facing a choice of whether to use emulated x86 locally, or hope that your Linux distro and tooling has an official ARM port.

In addition, debugging and profiling become much more difficult, because your tooling and your code may exhibit completely different behavior under ARM vs. x86.

I’ve already seen this play out with graviton2 vs. Xeon vs EPYC instances. Our software, the libraries, and the tooling used benefit from decades of x86 development and optimizations. Graviton was a completely different beast and we would have needed twice as many servers to handle our workloads.

Note that most of our stack is open source except our application code, so this isn’t our fault.

Apple needs to address this issue. Our company is going out of it’s way to NOT buy these machines because of this.

That isn’t the only area where Macs fall short either, so if you are on a PC, make sure you investigate whether your workloads are compatible with ARM and the macOS itself before you run out and purchase one. I have a Mac application that does NOT run on ARM or rosetta AT ALL, for instance.
 

The Hardcard

Member
Oct 19, 2021
46
38
51
. . . seriously?

Until IceLake, all AVX-512 implementations featured lower multipliers that would be inflicted on every core of a CPU even if only one core were handling AVX-512 instructions. Anyone using AVX-512 on a Skylake-SP CPU (for example) on a single core would nuke the performance on every core, even when the CPU was hosting multiple VMs on different cores. Very annoying! I think Intel managed to fix that on IceLake/RocketLake and TigerLake but i'm not 100% sure! Some lesser variant of that bogeyman may still exist.

The amount of re-engineering necessary to accommodate AVX-512 lead to massive amounts of die area being allocated just to AVX-512 support. The caches all had to be realigned for AVX-512 support. The less-often you used it, the more it would negatively affect the overall design of the chip.

And yes, the points @moinmoin brought up were also valid.

I don't want to thread derail, but you did mention AVX-512 complaints, so there they are. AVX-512 has been a mess. Something like SVE2 will work out much better. Hopefully Apple will choose to embrace SVE2 eventually.



Like I said, it's great until it isn't. One switch in algo and then you can't use it anymore. It's planned obsolescence unless you think that people will stay with the same codec forever. Let's face it, the dedicated hardware will always be faster and more power-efficient where applicable.



Please link those videos, or better yet text articles with that data?

I haven’t come across written articles yet, but here are a couple of Youtube links:


 

Heartbreaker

Diamond Member
Apr 3, 2006
4,223
5,225
136
Like I said, it's great until it isn't. One switch in algo and then you can't use it anymore. It's planned obsolescence unless you think that people will stay with the same codec forever. Let's face it, the dedicated hardware will always be faster and more power-efficient where applicable.

You are still better off until that day, when you theoretically meet the codec that levels the playing field, so it's still a better choice.

Also Look at these guys doing serious video editing, it's multiple sources, in multiple formats, so a single new codec being introduced that doesn't have HW support, is still going to leave the solution that has very good HW for everything else ahead.

Also I would bet that the HW acceleration is a mix of HW and SW so small changes in an algorithm won't invalidate it. You most likely just not going to get HW support for completely new codecs.
 

DrMrLordX

Lifer
Apr 27, 2000
21,600
10,792
136
Also I would bet that the HW acceleration is a mix of HW and SW

That's possible. Anything beyond that observation is speculation, but the M1 Max in particular has a lot of GPU resources available . . .

@The Hardcard

Having problems picking data out of those vids. The first one may be using hw acceleration and the second one just has a M1 Max running solo without any comparative data I could find. Doesn't someone just have some handbrake benchmark runs on their M1 Max or similar?

For anyone that has or is getting one of the new Macbook Pros, we have a handbrake thread in this subforum:


Any contributions to that thread would be appreciated. I'm pretty sure handbrake may support some of the hw acceleration available under M1 Max (or if it doesn't now, it may in the future), but for the sake of gathering data, I'm pretty sure that can be disabled/bypassed.
 
Last edited:

Nothingness

Platinum Member
Jul 3, 2013
2,390
724
136
Note that most of our stack is open source except our application code, so this isn’t our fault.
Since it contains code from your company, what makes you so confident it‘s not your fault?

That isn’t the only area where Macs fall short either, so if you are on a PC, make sure you investigate whether your workloads are compatible with ARM and the macOS itself before you run out and purchase one. I have a Mac application that does NOT run on ARM or rosetta AT ALL, for instance.
You mean Cinebench? If you want to be trusted you should provide some more data because your track record doesn’t inspire confidence. So what is that open source sw stack that doesn’t run well on Arm? What is that software that doesn’t run at all on an Arm Mac? Obviously these things exist, but claims with no evidence can just be discarded and are useless.
 

Nothingness

Platinum Member
Jul 3, 2013
2,390
724
136
For anyone that has or is getting one of the new Macbook Pros, we have a handbrake thread in this subforum:


Any contributions to that thread would be appreciated. I'm pretty sure handbrake may support some of the hw acceleration available under M1 Max (or if it doesn't now, it may in the future), but for the sake of gathering data, I'm pretty sure that can be disabled/bypassed.
To be fair you would also need some people to rerun with Handbrake 1.4 on their x86 machine (and with HW acceleration disabled too of course) as this was the first version to officially support Mac M1.