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

 

[Heartbreaker]

Yeah, and too a fault. There is either a design issue or a serious lack of needed optimizations.

The thing we shouldn't lose sight of for M1 devices is that they are full service SoC with a large GPU, and other dedicated accelerators, so Apple appears to use fairly conservative clocking some sound reasons.

Like keeping the CPU section heat and power budget low so it doesn't pull down other parts of the SoC. Lower clock speed means a LOT more chips will make the cut. This is not like AMD and Intel where they have different binned parts. Everything M1 is in the same speed grade.

If they increased the CPU speed, the power budget would start to interfere with the GPU/Accelerator power/heat budgets and less chips would meet the cutoff.

Now if Apple was just making a CPU without all the SoC baggage, and they were doing multiple speed bins, then yeah, the could probably have some 16 Performance core, CPU only parts that would post more impressive open source benchmarks, and yay, more flexing for nerds in the DIY community...

But Apple isn't building a part like that. Apple is building a full service SoC aimed directly at speeding up the exact jobs most of the high end Mac audience is doing like Video Editing/Graphics editing/SW development.

The important benchmarks for Apple are Video Editing Suites, Adobe Tools, and Compile times, and in those they are doing a fantastic job, where it actually counts.

 

[LightningZ71]

Or just the laws of physics. That is a LOT of transistors in that tiny space (even though its a hhuge processor). Not having space between the individual chips with an oversized heat spreader and a more robust heat sink will definitely negatively impact thermal management.

 

[Eug]

This is from the teardown.

BTW, basically everything (except RAM) is modular on the machine. It was mentioned that the SSD is removable, but so are all the ports, etc.

 

[DrMrLordX]

Not having space between the individual chips with an oversized heat spreader and a more robust heat sink will definitely negatively impact thermal management.

Not really. The temps are seriously low at around 3050 MHz. Apple had more thermal headroom available. The fans aren't even ramped up that high.

 

[Heartbreaker]

This is from the teardown.
...

BTW, basically everything (except RAM) is modular on the machine. It was mentioned that the SSD is removable, but so are all the ports, etc.

Not too bad from a repair-ability perspective. The main things that are wear items are ports and SSD, and those are not soldered in.

 

[Doug S]

Let's not forget that M1 is Apple Silicon v1.0.

They might be willing to push things a bit more in M2 or M3, but we are never going to see the crazy power budgets that Intel and AMD consider OK. Apple does not want to build Macs that sound like a jet taking off from a carrier when pushed hard. But perhaps allowing the fan to become audible when pushed hard is OK

There is probably some driver optimization still to be done on the GPU side since Apple Silicon GPUs have so many more cores than than the iPhone/iPad SoCs that came before, and that's before we consider the potential issues around making them work well across SoC dies. We've seen Intel, AMD and Nvidia get some pretty good sized GPU performance gains from driver fixes after release, especially of a brand new generation of GPU. Apple is likely to do the same.

 

[Eug]

Mac Studio appears to be Apple's answer to those wanting machines where lack of noise is of paramount importance.

They are not squeezing these chips for every last drop of performance. They are configuring them to be very powerful, but also very quiet, with decent yields.

The heatsink on the Ultra is massive and is fed by two large blowers. This ought to work well for those using Pro Tools for example. Yes, many high end pros will have the hardware in separate soundproof rooms, but those smaller shops will have these machines on their desks next to their music equipment, so a quiet machine is necessary.

Noise was a common complaint for these types of owners of the iMac Core i7 and Mac mini Core i7. This was solved with the Intel iMac Pro, but was never fixed with Intel Mac mini line.

Could they stuff the M1 Pro and M1 Max in the Mac mini? Sure, but especially with the Max they could run into the same noise issues. As it stands now, I suspect Apple will release an M1 Pro model to replace the so-called "high end" Intel Mac mini, but at this point I'm betting the M1 Max won't make it into the Mac mini. An M1 Pro Mac mini would not necessarily be whisper quiet either under heavy load, but it would fit into that price tier in between the Mac mini and the Mac Studio (which costs twice as much as the Mac mini), and it would be quiet enough for most prosumers. Furthermore, there would be several levels of the M1 Pro Mac mini to meet different price points, using binned variants, not to mention the option of 16 GB RAM. (You can't get less than 32 GB RAM in the Mac Studio.)

 

[Ottonomous]

This is from the teardown.

View attachment 58898

View attachment 58899

View attachment 58901

View attachment 58900

BTW, basically everything (except RAM) is modular on the machine. It was mentioned that the SSD is removable, but so are all the ports, etc.

View attachment 58902

View attachment 58903

That concession, modularity, alongside affordability (relative to predecessors) suggests that they're appealing more and more to the mainstream. They're contemplating a future where they're a cheap and powerful alternative to AMD/Intel prebuilts/custom PCs, with the added benefit of interoperability between their mobile devices.

 

[Doug S]

That concession, modularity, alongside affordability (relative to predecessors) suggests that they're appealing more and more to the mainstream. They're contemplating a future where they're a cheap and powerful alternative to AMD/Intel prebuilts/custom PCs, with the added benefit of interoperability between their mobile devices.

Is it a concession, or is Apple simply designing in a way easiest for them to build systems? The higher end you go the more options you have or may want to have in the future. Maybe next year's model offers 40GbaseT or 100Gbit fiber as an option, similar to how the Mini offers an upgrade to 10GbaseT. Since the Studio has two SoCs it has two SSD controllers, and probably a larger range of storage options. Modularity makes sense to reduce the number of SKUs of the board that differ only in installed NAND.

There's really no way you can make a concession to modularity to "appeal to the mainstream" in devices that have soldered DRAM. That DRAM may be soldered for a good reason, but if you can't upgrade/replace the RAM no one is going to care that fixing a broken ethernet jack is easy to do! Sorry you can't replace the tires on this car yourself, but swapping out the radiator is a breeze!

 

[Eug]

Some Power and performance numbers. Kind of looking like Apple went really conservative on the Mac Studio. Maybe some OS/firmware updates will get more out of the M1 Max & Ultra?

In Geekbench 5, the CPU is peaking at about 72 Watts.
In 3DMark Wildlife Extreme, the GPU is peaking at about 84 Watts. (In Geekbench 5 Metal, it was below 55 Watts.)





Apple's power utilization graph mentions GPU power usage maxing out around 110 Watts, so it appears these benchmarks cannot max out the GPU.

 

[Roland00Address]

This is from the teardown.
BTW, basically everything (except RAM) is modular on the machine. It was mentioned that the SSD is removable, but so are all the ports, etc.

Thinking about this later and it makes complete and utter sense for the ssd to be not soldered in like it is on the phones and laptops. Having a modular design provides many benefits such as easy repairability but arguably more importantly allows Apple to figure out exactly how much of X they want like they want 2 TB ssd or 4 TB ssd and not get locked in where a customer wants the 2 TB but only the 4 TB is in stock with the warehouse.

The flexibility providing by the modularity helps Apple manage the Supply Chain but also put pressure on suppliers since ssds are just raw components unlike a M1 or M1 Ultra.

-----

There is some downsides to this which is size, but size is not a problem with the mac studio and arguably the mac laptops but it is a big deal with the phones.

 

[DrMrLordX]

Mac Studio appears to be Apple's answer to those wanting machines where lack of noise is of paramount importance.

They could have just watercooled the thing for that price.

 

[Eug]

They could have just watercooled the thing for that price.

Heh. I think the watercooled G5 disaster put the final nail in that coffin.

 

[DrMrLordX]

Heh. I think the watercooled G5 disaster put the final nail in that coffin.

Possibly, though watercooling has come a long way since then.

 

[eek2121]

Let's not forget that M1 is Apple Silicon v1.0.

They might be willing to push things a bit more in M2 or M3, but we are never going to see the crazy power budgets that Intel and AMD consider OK. Apple does not want to build Macs that sound like a jet taking off from a carrier when pushed hard. But perhaps allowing the fan to become audible when pushed hard is OK

There is probably some driver optimization still to be done on the GPU side since Apple Silicon GPUs have so many more cores than than the iPhone/iPad SoCs that came before, and that's before we consider the potential issues around making them work well across SoC dies. We've seen Intel, AMD and Nvidia get some pretty good sized GPU performance gains from driver fixes after release, especially of a brand new generation of GPU. Apple is likely to do the same.

More likely, the transistors are so tightly packed that they can't hit the higher voltages necessary to increase frequencies.

Also, the highest total power consumption I've seen the SoC use is 97W, which isn't much below where the 5950X sits, so it isn't like they have drastically lower power consumption than competitors.

What Apple has mastered (beyond a node shrink) is scalability, and they are able to do that thanks to a mobile first design. As I sit here playing a lightweight game on my 5950x, the CPU power consumption is 100W and all 16 cores are pegged at 4.6 GHz, yet only 1 core is fully loaded, and another is 70%. AMD targets core readiness and high clock speeds above all else. Apple's chips are mobile by design. AMD's chips are server/workstation chips scaled down. 2 completely different approaches. We'll see what Zen 4 brings to the table, however. Either way, I need to buy a Mac for a new contract I just signed, so we will see what the offerings are when the time comes.

 

[soresu]

More likely, the transistors are so tightly packed that they can't hit the higher voltages necessary to increase frequencies.

Also, the highest total power consumption I've seen the SoC use is 97W, which isn't much below where the 5950X sits, so it isn't like they have drastically lower power consumption than competitors.

What Apple has mastered (beyond a node shrink) is scalability, and they are able to do that thanks to a mobile first design. As I sit here playing a lightweight game on my 5950x, the CPU power consumption is 100W and all 16 cores are pegged at 4.6 GHz, yet only 1 core is fully loaded, and another is 70%. AMD targets core readiness and high clock speeds above all else. Apple's chips are mobile by design. AMD's chips are server/workstation chips scaled down. 2 completely different approaches. We'll see what Zen 4 brings to the table, however. Either way, I need to buy a Mac for a new contract I just signed, so we will see what the offerings are when the time comes.

There's more to it than just what AMD does to the hardware itself.

Your CPU utilisation is also configurable in Windows by percentage.

I do this purely because the Noctua HSF I installed has an abominably annoying sound when the CPU gets even slightly under load.

I can run a 4K AV1 clip with GPU HW decoding disabled (so all CPU work) and it does not even breach 20W on my 3950X in Ryzen Master.

Mobile hardware makers are nothing if not highly aggressive with power scaling, even more so when they have absolute control over software as Apple does - I would not expect that to change just because the device was intended to be wall powered.

Ryzen n950X are high end mainstream desktop chips, designed for power use as you say - a better comparison is to SoCs designed to be more power conscious like the 5xxx or 6xxx APU series.

 

[Thala]

Also, the highest total power consumption I've seen the SoC use is 97W, which isn't much below where the 5950X sits, so it isn't like they have drastically lower power consumption than competitors.

Excuse my ignorance, but the 5950X does not have a GPU of the RTX 380 performance level mind you? So using SoC power numbers is dubious - you might want to check the RTX380 idle power for reference. More importantly the 5950X does not even have the CPU performance of the Apple M1 Ultra, in fact my 5950X is significantly slower, while consuming 130-140 W.
So the conclusion that the M1Ultra is in a completely different league with respect to power efficiency is not an understatement.

 

[Doug S]

The flexibility providing by the modularity helps Apple manage the Supply Chain but also put pressure on suppliers since ssds are just raw components unlike a M1 or M1 Ultra.

It fits well with Apple's existing yearly purchases of raw NAND for iPhone/iPad. They were already buying a massive amount of NAND, they just needed to up the order a bit for Apple Silicon Macs.

 

[Doug S]

More likely, the transistors are so tightly packed that they can't hit the higher voltages necessary to increase frequencies.

Perhaps, but "not a lot of room" != "no room at all". Apple doesn't bin on frequency, so the frequency they select has to pass pretty much 100% of parts, because they don't have anywhere to put functional M1 chips that can't run at 3.2 GHz. They want to only disqualify parts that have defects they can't work around i.e. selling with fewer cores. There's clearly got to be room to increase frequency by at least 10-15% just knowing the typical frequency binning curve. And that's even before you start using high performance cells which TSMC indicates can gain more performance (10-20% according to their figures)

It is interesting how much engineering margin there is in the power supply sizing and cooling design in the Studio. Maybe that's just engineers being allowed to over-engineer without the usual penny pinching making them cut corners, or maybe there is a surprise coming with M2. I mean, as it stands there's no reason why the Ultra variant needs the heatsink to be made from copper. There's no reason the power supply needs to output 370 watts, unless they've allocated half the power budget for USB-C PD.

 

[Eug]

It is interesting how much engineering margin there is in the power supply sizing and cooling design in the Studio. Maybe that's just engineers being allowed to over-engineer without the usual penny pinching making them cut corners, or maybe there is a surprise coming with M2. I mean, as it stands there's no reason why the Ultra variant needs the heatsink to be made from copper. There's no reason the power supply needs to output 370 watts, unless they've allocated half the power budget for USB-C PD.

Mac Studio has 4 full-fledged Thunderbolt 4 ports on the M1 Max, and 6 full-fledged Thunderbolt 4 ports (utilizing 6 independent Thunderbolt controllers) on the M1 Ultra. The power supply is built for the M1 Ultra, but is the same one on the M1 Max model as well. And there are an additional 2 USB-A ports.

I'm not sure how much power each USB-C port on the Mac Studio can output, and how much they can output in aggregate, but the interface theoretically can support up to 100 Watts per port I believe.

 

[repoman27]

The total power budget for all of the ports on the Mac Studio is unlikely to be much greater than 125 W. The Type-C ports aren't designed to provide power for charging connected laptops, but Apple probably does want them all to be able to support bus-powered devices without issue. That would translate to (15 W Vbus + 1.5 W Vconn) * 6 = 99 W. The USB Type-A ports only need to provide 4.5 W, but Apple probably provisions them to support their proprietary 12 W divider mode, so that's another 2 * 12.48 = 24.96 W. HDMI accessory power is only something like 0.25 W, and max SD Card power is 1.8 W. All in, that's near as makes no difference 126 W.

However, the PSU on the iMac pro was also 370 W. Before watching the Max Tech teardown of the Mac Studio and seeing how much engineering effort was put into this product, I had the distinct feeling that this was perhaps not what Apple originally intended to build. It felt like they wanted to do an updated iMac Pro with a mini-LED display, but either the panel wasn't ready yet or there were other supply chain constraints preventing that from happening. So they went with the Mac mini on steroids Studio concept instead. However, both systems were designed to utilize as many common components as possible, hence we end up with a 370 W PSU in a nearly triple-height Mac mini with an SoC that doesn't draw much more than 125 W.

Also, as far as binning goes, Apple probably bins for leakage more than frequency. It doesn't matter if the M1s destined for Mac minis require a little more voltage to hit the base clocks. The M1s for the MacBooks need to get there at lower power, even if it means disabling a GPU Core. And the iPads get the super low leakage parts.

 

[Eug]

The total power budget for all of the ports on the Mac Studio is unlikely to be much greater than 125 W. The Type-C ports aren't designed to provide power for charging connected laptops, but Apple probably does want them all to be able to support bus-powered devices without issue. That would translate to (15 W Vbus + 1.5 W Vconn) * 6 = 99 W. The USB Type-A ports only need to provide 4.5 W, but Apple probably provisions them to support their proprietary 12 W divider mode, so that's another 2 * 12.48 = 24.96 W. HDMI accessory power is only something like 0.25 W, and max SD Card power is 1.8 W. All in, that's near as makes no difference 126 W.

However, the PSU on the iMac pro was also 370 W. Before watching the Max Tech teardown of the Mac Studio and seeing how much engineering effort was put into this product, I had the distinct feeling that this was perhaps not what Apple originally intended to build. It felt like they wanted to do an updated iMac Pro with a mini-LED display, but either the panel wasn't ready yet or there were other supply chain constraints preventing that from happening. So they went with the Mac mini on steroids Studio concept instead. However, both systems were designed to utilize as many common components as possible, hence we end up with a 370 W PSU in a nearly triple-height Mac mini with an SoC that doesn't draw much more than 125 W.

Also, as far as binning goes, Apple probably bins for leakage more than frequency. It doesn't matter if the M1s destined for Mac minis require a little more voltage to hit the base clocks. The M1s for the MacBooks need to get there at lower power, even if it means disabling a GPU Core. And the iPads get the super low leakage parts.

The MacBook Pros also get CPU cores disabled in some models.

 

[Doug S]

So they went with the Mac mini on steroids Studio concept instead. However, both systems were designed to utilize as many common components as possible, hence we end up with a 370 W PSU in a nearly triple-height Mac mini with an SoC that doesn't draw much more than 125 W.

That's really a stretch. Even if what you suggest was true, using the same PSU makes no difference in the ease of bringing a different system to market. That's literally the easiest thing they can do, switching to a lower wattage power supply if 370W is more than they need.

Its ridiculous though just based on the size of the Studio's internals. How in the heck were those giant heatsinks going to fit inside their most recent iMac form factor? If they were going to stick an M1 Ultra in there it wouldn't use the big heavy copper heatsink, it would probably have had something like phase change cooling heat pipes on the package to keep it very low profile and used the entire backside of the display to dissipate excess heat. I could see Apple having done something different and use thin copper fins instead of plastic for the rear of the iMac Pro display. Whatever they did, it would be nothing like the cooling solution they did with the Studio.

Seems pretty clear that they had already concluded integrating a midrange / entry "pro" level desktop with the display is a dumb idea. Not sure where you get the idea the Studio was a rush job. It seems exceptionally well engineered to me, not something slapped together at the last minute as you imply. Why would having a Mini LED display make an all-in-one Studio more attractive? The more expensive the display the worse the idea of integration is. A lot of professionals already own a quality display, and don't want to pay for something they don't need - or to have one that doesn't match if they are one of the many with more than one display.

You get that's why they sell the stand separately, right? Because they know a lot of professionals also have some type of VESA mounting solution (swing arm or whatever) for their monitor(s) which the iMac design also screws up since you can't remove a stand if part of the Mac is integrated within it and if you want a different display height it won't match your other ones unless you set the iMac on a book or something.

 

[jeanlain]

Apple's power utilization graph mentions GPU power usage maxing out around 110 Watts, so it appears these benchmarks cannot max out the GPU.

It is not common for an app, game or such, to use a GPU at its max rated power. Furmark is famous for maxing GPU power. Most other apps don't.
84 W consumed by a GPU that can draw 110 W max is already quite high.

It's the same for CPU power BTW. Primer95 uses much more power than most other apps that still show 100% CPU utilisation.

 

[igor_kavinski]

https://twitter.com/x/status/1505708249135296515

Only Apple would do something wasteful like this.