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

 

[Eug]

rumored "compact" Mac Pro

Cube: Fail*
Trash can: Fail**
Third time’s a charm?***


* I own two of these.
** I was looking to buy one of these used just for fun but they are still stupidly pricey.
*** I’m not counting the Mac mini.

 

[gdansk]

I think they can do it this time. M1 efficiency with the new focus on function over form should avoid some of the problems. Though I hope they keep some full size PCIe slots somehow.

 

[Heartbreaker]

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.

Doubt that. To quote the very accurate Gurman:
“Apple engineers are currently developing a new ‌Mac Pro‌ that looks like the current design at about half the size. It’s unclear if that Mac will replace the current ‌Mac Pro‌ or if it’s an additional model. Apple’s chip designs could help the company reduce the size of its computers due to increased power efficiency, but the current ‌Mac Pro‌ is large, in part, to fit components like additional storage drives and graphics chips,” reports Bloomberg.


The Compact Mac Pro is the new Apple Silicon Mac Pro, not an inferior model. I expect it will come in 2-M1M and 4-M1M versions, as Gurman indicated before for Mac Pro.

He says it's unclear if the new one is a replacement or if they are sold together, but I would expect the Big Mac Pro chassis will end with the Intel Mac Pro, whenever that happens. They may overlapp a bit, but I doubt the big Mac Pro will be converted to ARM.

Look at the current Mac Pro. It's mostly Card cages. There are 8 slots, 4 of them are double wide. That's like a 12 card slot cage, much of it there for GPUs, which are probably off the menu, so MUCH less need for Slots. You could cut almost half the height and still have 4 slots.

 

[LightningZ71]

Assuming that Apple believes their GPU is a sufficient replacement for any of their add in cards, what remains that requires actual PCIe slots that can't be served by TB4 connections?

 

[moinmoin]

Though I hope they keep some full size PCIe slots somehow.

You already know that they will keep it small enough to not have to excuse themselves for it to lack full size PCIe slots.

 

[moinmoin]

This management aims to keep the second group of P cores unloaded as much as possible

I'd guess the group of P cores can be power gated as a whole (or close to that) if not in use, so this behavior makes perfect sense.

 

[gdansk]

Assuming that Apple believes their GPU is a sufficient replacement for any of their add in cards, what remains that requires actual PCIe slots that can't be served by TB4 connections?

I had hoped at some point Apple would make upgrade GPUs as they seem to develop quite a lot more quickly than CPUs, especially with these new coherent MCM GPUs on the horizon. I guess TB4 would work too.

 

[Heartbreaker]

Assuming that Apple believes their GPU is a sufficient replacement for any of their add in cards, what remains that requires actual PCIe slots that can't be served by TB4 connections?

A few slots to keep people from screaming that they are going back to the closed Trashcan.

I quick bit of google shows me that after GPUs, the Mac Pro PCIe slots seem to be most used for hosting fast SSD drives.

 

[LightningZ71]

TB4 is essentially 32GB/s (should read 32Gb/s) of PCIe bandwidth. That's enough for most high speed storage solutions. For GPUs in external enclosures usd for compute, it should typically be sufficient as well, though, I don't see a lot of demand there beyond what the Max can provide normally.

 

[Doug S]

Cube: Fail*
Trash can: Fail**
Third time’s a charm?***


* I own two of these.
** I was looking to buy one of these used just for fun but they are still stupidly pricey.
*** I’m not counting the Mac mini.

Mac Mini: success
Macbook Air: huge success

Not everything "small" has been a fail - and I'd argue the Trash Can's issues weren't because it was too small - if it had the same volume in a more traditional form factor it would have been fine. Its odd shape was an issue when Apple wanted to upgrade it and standard components wouldn't fit. While I don't think anyone outside of Apple knows for sure my hunch is the case was designed first by Jony Ive because he decided to break the model of boring PC towers by making one that's round, and the engineers were told to make the components fit.

That, along with other form over function fails like the butterfly keyboard and Macbooks with only one port are what happens when Steve Jobs isn't around to tell Jony Ive "no". Apple is better off now that he's gone. Design is important, but it has to take engineering realities into account. Ironically with an M1 Max based design making it round would be very easy to manage lol

My guess for the more compact Mac Pro is that it would be a similar form factor to the Mini, but 2x larger in all dimensions. That gives them enough room for a board holding two M1 Max and all the LPDDR5, and sufficient airflow for near silent cooling. There's no reason to make it a tower, because I don't anticipate the compact version will have any internal expandability at all.

 

[Doug S]

TB4 is essentially 32GB/s of PCIe bandwidth. That's enough for most high speed storage solutions. For GPUs in external enclosures usd for compute, it should typically be sufficient as well, though, I don't see a lot of demand there beyond what the Max can provide normally.

Thunderbolt 4 is 40 Gb/s not GB/sec - or 32 Gb/sec of PCIe bandwidth i.e. basically a single lane. I think adding a few x4 slots for faster SSDs, fibre channel or 100 Gb ethernet would be desirable for the full sized Mac Pro built from the M3. The more compact one will probably not be expandable, which would explain why they are introducing an Ice Lake based Mac Pro next year to keep the people who need more CPU cores as well as those who need fast storage/networking or large amounts of RAM happy before they're ready to handle that with an ARM Mac Pro.

 

[Heartbreaker]

My guess for the more compact Mac Pro is that it would be a similar form factor to the Mini, but 2x larger in all dimensions. That gives them enough room for a board holding two M1 Max and all the LPDDR5, and sufficient airflow for near silent cooling. There's no reason to make it a tower, because I don't anticipate the compact version will have any internal expandability at all.

Your guess contradicts what Gurman said, which was that it's basically the current Mac Pro Design at about half the size.

 

[Doug S]

Your guess contradicts what Gurman said, which was that it's basically the current Mac Pro Design at about half the size.

There's zero reason for it to be so large unless it will support DIMMs and PCIx4 slots, so if he's right it will have those as well.

 

[jpiniero]

There's zero reason for it to be so large unless it will support DIMMs and PCIx4 slots, so if he's right it will have those as well.

Power consumption might be higher than you think. Might be cheaper in cooling to go bigger.

 

[Mopetar]

It would be strange for them not to have DIMMs when the main reason to solder memory in their laptops is space. They're going to have a massive amount of a memory channels available if they are using up to 4x M1 Pro Max dies or something similar to that. This thing is also pretty limited if it doesn't have room for adding PCIe cards for various specialty needs.

 

[Gideon]

It would be strange for them not to have DIMMs when the main reason to solder memory in their laptops is space.

I also believe they will have at least some PCIe (5.0) slots for added flexibility (CXL among others), but I really doubt they have DIMM slots. It's not just about the slots.

They already have 8x LPDDR5 channels per CPU (of which there will be 2-4). For DIMMS they'd need two to four additional DDR4/5 memory-controllers (of which there is no proof on M1 MAX chip anyway).IMO The added routing complexity on package alone would be silly.

Look at it this way. Adding a hypothetical PCIe 5.0 slot would allow:

• Some new future afterburner card
• Custom networking hardware
• A potential future GPU down the line
• or the aforementioned CXL memory/storage (that would allow Terabytes of memory in total).

Those memory-controllers on the other hand would require:

• Tons of designing and validation work.
• Take noticeable die-space
• Add needless complexity to the memory hierarchy
• Add significant complexity to the CPU package and require lots of new pins to the socket.

What's worse, In the end they would only be useful to the selected few who not only need more than 256GB of memory, but to whom the latency of the extra memory is somehow so critical that CXL wouldn't cut it.

Given the usual mac-pro clientele that niche is so extremely tiny it would never be worth the expenses.

 

[Heartbreaker]

I also believe they will have at least some PCIe (5.0) slots for added flexibility (CXL among others), but I really doubt they have DIMM slots. It's not just about the slots.

They already have 8x LPDDR5 channels per CPU (of which there will be 2-4). For DIMMS they'd need two to four additional DDR4/5 memory-controllers (of which there is no proof on M1 MAX chip anyway).IMO The added routing complexity on package alone would be silly.

I agree, they won't add more memory channels to the M1M die.

But they also need some way to connect the dies in 2 way and 4 way configs.

If they need a bunch of extra memory channels, and they need a way to connect the dies, then a central Central I/O die makes sense. This only requires 1 High speed connection on the M1M die, and they can add memory channels to the I/O Die.

While this requires that they Fab and additional die, it seems to have the most advantages when considering how to both connect the dies, and add a lot more memory channels.

 

[Doug S]

It would be strange for them not to have DIMMs when the main reason to solder memory in their laptops is space. They're going to have a massive amount of a memory channels available if they are using up to 4x M1 Pro Max dies or something similar to that. This thing is also pretty limited if it doesn't have room for adding PCIe cards for various specialty needs.

I don't know if space is the main reason to solder memory in laptops. Apple cares a LOT about power draw, and that's reduced by soldering the RAM and even moreso by using LPDDR instead of standard DRAM. They could have had DIMM slots in the Mac Mini but don't.

If they have one form factor for all ARM Mac Pros, they'd need to make it big enough to accommodate ones that are expandable even if some lower end models are not. So maybe we see an ARM Mac Pro next year with two M1 Max"+" SoCs and no expandability in an entirely empty case that looks ridiculous, but the later high end models with DIMMs and x4 slots actually need that volume.

 

[Mopetar]

I'm pretty sure that using DIMMs on their notebooks would require making them thicker. One could always argue the real reason is that it also means they can charge a premium for soldering more memory on themselves and there's probably some truth to that.

However, the Mac Pro is so niche and so high-end they can already charge through the nose and just let people wanting to save a bit toss in their own RAM. Some will probably pay the Apple markup just to not have to deal with the support themselves anyways.

They won't sell a really cut down Mac Pro in a smaller case. They'll slap that configuration into an iMac Pro and sell you that instead. Depending on what you're doing you probably want that anyways.

People who want a cheap server are probably going with a bunch of Minis like they have been for a while now. Unless Apple makes a new server model that's meant for people who really want to scale, the economics of a huge number of minis makes more sense and you can probably cram more of those into less space anyways.

 

[Doug S]

I'm pretty sure that using DIMMs on their notebooks would require making them thicker.

Depends on how packed in the internal components are. SO-DIMM slots can be flush to the board and horizonal, so they are basically only the thickness of the board plus the thickness of the SO-DIMM (and they can specify single sided only to further minimize the thickness)

That doesn't have to add to the thickness of the laptop, it just comes down to whether there is sufficient internal "free" volume to make the board larger to accommodate a couple SO-DIMM slots. There isn't room for that in the Macbook Pro, but that wasn't a design goal so they fully used the space they had with dual rather large fans. I imagine they could have figured it out if they'd been designing with SO-DIMMs in mind, and perhaps made the fans a bit smaller or only had one.

 

[Doug S]

I've been wondering for a while what the maximum amount of memory you could get in an LPDDR5 package, Andrei's article about Samsung's introduction of LPDDR5X has answered that at least for LPDDR5X - 64 GB.

So a future Pro Max die that currently can handle up to 64 GB would be able to go to 256 GB with LPDDR5X (and go from 400 GBps to 533 GBps and consume 20% less power as a bonus!)

A Mac Pro built with four such dies could be configured with up to 1 TB of memory. I've always been of the opinion Apple will have to support DIMMs in the Mac Pro as a max memory config of 256 GB is just not going to be enough. But at 1 TB that view gets pretty iffy as there can't be much of a customer base for DRAM configurations above that size.

 

[Heartbreaker]

I've been wondering for a while what the maximum amount of memory you could get in an LPDDR5 package, Andrei's article about Samsung's introduction of LPDDR5X has answered that at least for LPDDR5X - 64 GB.

Theoretically. We will see what actually gets delivered.

 

[Doug S]

Theoretically. We will see what actually gets delivered.

True, it may not make the price list unless/until they feel they have a market for it. But if Apple called up either Samsung or Micron wanting such parts I'm sure either would be willing to supply them.

 

[Eug]

Dammit. My Mac Pro 8-core Xeon X5365 machine is giving me headaches. For now I've replaced it with a 16 GB Core m3-7Y32 12" MacBook. Remarkably, it's working fine driving a 30" Apple display and a bunch of USB devices all through its single USB-C port, but this is not an ideal solution for the long term of course. (I'm using it for VPN and business productivity applications, not content creation.)

The Mac minis with the new case design can't come soon enough.

 

[Gideon]

HW unboxed review:

A pretty good take, I agree mostly.

The only part I subjectively find arguable is the claim that windows laptops running essentially without power-limits (say 70W) "to complete tasks as fast as you can" beating M1 is somehow a good option.

Yes technically you can do that, but why then does Intel get such a bad rap from a lot of people (them included) with their 250W CPUs?

Maybe I'm just getting too old, but I detest loud and hot laptops (especially when working with others in the same room). Dumping that much heat out really makes working with the laptop itself unbearable very fast and is particularly bad during summer.

I think that apple is way too conservative with having no turbo at all. Something like Intel PL1 is still very useful, even though it should be way less aggressive with voltage and probably shorter. I still much prefer their take to the increasingly more aggressive turbos Intel (and even AMD more and more) have in their 15W let alone 25-45W laptops.

PL2 and AMD's 1.35x TDP boost are perfectly fine on a desktop. But I've yet to see a laptop (not talking about those clunky desktop replacements) where this long-time boosting behavior doesn't make coding for extended periods a miserable experience. This is the area where the turboless ARM chips thrive (it's rather niche I admit). 5nm level SKUs will surely help on the x86 side as well (and possibly little-cores), but where can I buy a laptop them today?