Discussion Apple Silicon SoC thread

Page 71 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Eug

Lifer
Mar 11, 2000
23,583
996
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:

Gideon

Golden Member
Nov 27, 2007
1,608
3,573
136
32/64GB of HBM looks indeed a bit problematic. Then again i don't see a problem of having two memory pools if they use 2.5D packaging anyway
 

dr1337

Senior member
May 25, 2020
309
503
106
Well, that was the point of my question. Are you suggesting they will have dedicated GPU memory and separate system memory?
I think there are a bazillion different ways apple could go about it. 4 stacks of HBM2E at 16gb each seems like an easier and cleaner solution than having 8 LPDDR5 dies spread out around the place. However I can also totally see them keeping two of the lpddr5 chips on package and just putting the rest on the mainboard. or switching to two stacks of HBM on package with ddr flanking it on the mobo. Either way I don't think cost is a major concern for apple for their high end parts, and if the leaked specs are true I expect their memory solution to be very beefy whichever route they go.
 
  • Like
Reactions: Gideon

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
For the performance and packaging goals that Apple typically goes for, it would be a surprise to me if they went with more than two packages per CPU, and, given the expanded ram bandwidth requirements that that segment of the market will have, they're going to need to maximize throughput per package. They can either use LPDDR5 stacks in place of the LPDDR4X ones (not exactly new to the industry as it's been in use in mobile phones for almost a year now), or they can use HBM2E stacks. This will allow them to have devices with 32-64GB per CPU (assuming the already existing setup is retained).

What could be more interesting is what they will do with storage. We're already seeing them really lean on the SSDs hard for swap and still achieving solid performance. The metrics are showing that they are burning through SSD life at alarming rates in some situations. What will they do to mitigate this in professional environments where this will be more pronounced? Will they move to custom SSD implementations? They'll need something special there, as, assuming that they don't go to devices that allow RAM upgrades, even with devices that have 256GB of ram (4 processor chips with two packages each) will sometimes run into situations where they need more memory.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
I agree. What about memory technology? I think M1 is 2 channel LPDDRx. This 2xM1 chip needs access to up to 64GB memory. What about 4 channels of LPDDR5, with about 3x the bandwidth of M1.

Will this be on TSMC 5nm or 4nm?

Yeah I've been wondering about the memory too. The driving factor will be what memory config they need to allow sufficient bandwidth for the 4 chiplet Mac Pro to compete with the top end AMD and Nvidia discrete solutions. We already know they're not supporting third party graphics, so there's no reason to include a PCIe slot so even the Mac Pro should be pretty compact.

They could essentially dedicate all the on package memory to graphics, since will require separate DDR5 controllers for DIMMs for the Mac Pro at least. The rest may have it soldered to the board if it isn't exclusively using on package memory. Can they get sufficient bandwidth if they had for instance 4x LPDDR5 controllers per chiplet? That would be 16x LPDDR5 controllers for the high end Mac Pro.

It will very likely be made on N5P - they're talking about a summer release so N4 is out of the question since it won't begin volume production until sometime this fall.
 

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
What could be more interesting is what they will do with storage. We're already seeing them really lean on the SSDs hard for swap and still achieving solid performance. The metrics are showing that they are burning through SSD life at alarming rates in some situations. What will they do to mitigate this in professional environments where this will be more pronounced? Will they move to custom SSD implementations? They'll need something special there, as, assuming that they don't go to devices that allow RAM upgrades, even with devices that have 256GB of ram (4 processor chips with two packages each) will sometimes run into situations where they need more memory.


The Mac Pro will obviously support third party storage, whether that's via m.2 slots or they also include 2.5" or 3.5" slots we'll see. I think we see Apple's SSD as the "boot drive" and then customers can expand from there.

You're crazy if you think they will only have on package RAM for the Mac Pro. The current model maxes out at 1.5 TB, there's not a chance in hell they reduce the amount of memory possible all the way down to 256GB in the new one. It will have DDR5 DIMM slots, I guarantee it.
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
2021 24" iMac M1 internals:

csm_EzcaVbZVUAQBtD2_1c397835d8.jpg

2017 27" iMac Core i5 internals:

pYHRZpMG6v3p6uIL.huge.jpeg

The headphone jack is on the side, because the length of an audio plug is longer than the iMac's thickness.

Note though that the 2021 iMac has an external power supply. The 2017 iMac has an internal power supply.
 
  • Love
Reactions: eek2121

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
The Mac Pro will obviously support third party storage, whether that's via m.2 slots or they also include 2.5" or 3.5" slots we'll see. I think we see Apple's SSD as the "boot drive" and then customers can expand from there.

You're crazy if you think they will only have on package RAM for the Mac Pro. The current model maxes out at 1.5 TB, there's not a chance in hell they reduce the amount of memory possible all the way down to 256GB in the new one. It will have DDR5 DIMM slots, I guarantee it.

I'm not saying that I fully believe it myself, but, going with multiple tiers of memory is going to be quite complex. I see them pushing form factor over function. Other than the boot drive(s), any other storage could be externally attached. TB4 has a bunch of bandwidth, so I'm not too down on that as a choice.

As for additional RAM, I don't know what to tell you. I don't think that they will move away from the tightly integrated memory setup that they currently use for the M1as it seems that it's a significant factor in how well their system performs. I can see them possibly adding a second controller through some sort of less integrated interface, such as a CXL hub or something similar. However, I wouldn't put it past apple to do something proprietary with SSD, in a similar fashion to the PS5, or maybe a special wide SSD that's PCIe 4.0 x8 or equivalent.
 
  • Like
Reactions: Gideon

Gideon

Golden Member
Nov 27, 2007
1,608
3,573
136
You're crazy if you think they will only have on package RAM for the Mac Pro. The current model maxes out at 1.5 TB, there's not a chance in hell they reduce the amount of memory possible all the way down to 256GB in the new one. It will have DDR5 DIMM slots, I guarantee it.
I hope they are competent enough and instead use on-package memory coupled with a CXL DDR5 solution (hopefully evenwith a HBM "Cache" layered on the module as well). Something like this:
The third type is perhaps the one we’re most interested in today: memory buffers. Using CXL.memory, a memory buffer can be installed over a CXL link and the attached memory can be directly pooled with the system memory. This allows for either increased memory bandwidth, or increased memory expansion, to the order of thousands of gigabytes.

At least if I were Apple that's what I would aim for. Considering how all-out they went with their CPU designs and as they have not shunned from expensive memory before ...
Even if it's a rather limited and super expensive halo product a HBM2 + CXL expansion solution would be a winner for mindshare alone.
 
Last edited:

Eug

Lifer
Mar 11, 2000
23,583
996
126
Even if it's a rather limited and super expensive halo product a HBM2 + CXL expansion solution would be a winner for mindshare alone.
98% of people who buy Mac Pros probably wouldn't even know what that means. And for those who do, many wouldn't care as long as it does what it needs to do.

IMO such "mindshare" would not be to be a meaningful goal for Apple.
 
  • Like
Reactions: Magic Carpet

Gideon

Golden Member
Nov 27, 2007
1,608
3,573
136
98% of people who buy Mac Pros probably wouldn't even know what that means. And for those who do, many wouldn't care as long as it does what it needs to do.

IMO such "mindshare" would not be to be a meaningful goal for Apple.
This might be due to my bad English being a non-native speaker.

I didn't mean mindshare as "look at the cool tech we are using" but rather mindshare from the performance it would have with 32 Firestorm (+ 8 Icestorm) cores.

The peak bandwidth from a single CXL DDR5 memory buffer would be 128 GB/s (and that would be on top of what on-package memory provides, which with HBM2 would be bonkers). For reference Threadripper 3990X has only about 95.37 GB/s and 3995WX has about 200 GB/s.

All in all it would be surprisingly competitive with top-of-the line Threarippers, I could definitely see it winning some benchmarks with HBM and a CXL memory buffer and offer similar amounts of total memory
 
  • Like
Reactions: Eug

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
I don't even think that HBM is even needed here. Just replacing the LPDDR4X modules in the M1 package with LPDDR5 at the maximum current JEDEC spec, and multiplying that across up to four packages gives a crazy amount of RAM bandwidth. Those ICs are already available on the market, consume less power than HBM2E, and are cheaper per GB from what little I can scrape together. Any CXL attached memory solution need not use anything so crazy as HBM2E, and could, instead, just use a wide array of ECC DDR4 modules if desired.
 

eek2121

Platinum Member
Aug 2, 2005
2,904
3,906
136
The Mac Pro will obviously support third party storage, whether that's via m.2 slots or they also include 2.5" or 3.5" slots we'll see. I think we see Apple's SSD as the "boot drive" and then customers can expand from there.

You're crazy if you think they will only have on package RAM for the Mac Pro. The current model maxes out at 1.5 TB, there's not a chance in hell they reduce the amount of memory possible all the way down to 256GB in the new one. It will have DDR5 DIMM slots, I guarantee it.

Never under estimate Apple's ability to dumb things down. I suspect they'll use store statistics to drive design decisions. If nobody is purchasing an iMac with 512gb+ of RAM, they won't feel a need to support such a config.

Quite frankly, I fail to see the use case in such a config. Mac Pros aren't servers (though admittedly you can buy a rack kit for them). Even if they were used as a server, I can't see a need for a Mac to use that much RAM. Remember, Apple isn't in the server market. They killed the server oriented version of macOS long ago. Most of the projects I can think of that would use that much RAM aren't available on a Mac. The only use case that I can think of is virtualization, but Macs have limited value there, since they can't run Windows or other x86 operating systems.
 

ashFTW

Senior member
Sep 21, 2020
303
225
96
Assuming Apple releases new version of Intel (Sapphire Rapids X) based Mac Pro, they might also support DDR5, PCIe 5, and CXL 1.1 on the Apple Silicon version. If they go with some IceLake version (also with max 40 cores) instead, then CXL support is unlikely.
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
  • Like
Reactions: Tlh97 and B-Riz

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
So far, eGPU support for M1 Macs isn't an announced coming feature. Presently, if you plug in an eGPU enclosure, the mac will recognize that it is plugged in, and even enumerate all the subdevices. However, there is no software stack at present that enables the eGPU functionality. Apple presently is publicly saying that they aren't going to support it in the near term, and that their M1 chips are good enough that there isn't a point to doing it anyway.

Physically, there's no real reason why it can't be supported as far as I know. The TB3 and 4 specs certainly allow it, and we can only assume that apple's implementation of those ports is fully spec compliant. It just requires that drivers be written and allowed into the OS that support the functionality.
 
  • Like
Reactions: Tlh97 and B-Riz

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
I hope they are competent enough and instead use on-package memory coupled with a CXL DDR5 solution (hopefully evenwith a HBM "Cache" layered on the module as well). Something like this:


At least if I were Apple that's what I would aim for. Considering how all-out they went with their CPU designs and as they have not shunned from expensive memory before ...
Even if it's a rather limited and super expensive halo product a HBM2 + CXL expansion solution would be a winner for mindshare alone.


What does running the accesses across the PCIe bus do to memory latency? How about the granularity of read/write sizes?
 

defferoo

Member
Sep 28, 2015
47
45
91
This is something I've been wondering about, but the design of the current Mac Pro (and any other desktop honestly) is the antithesis of what the M1 is. The Mac Pro is highly modular with user replaceable RAM, storage, graphics, accelerator cards. The M1 essentially takes all of that minus the storage and expansion cards and puts it on a single package. How will Apple build a new Mac Pro with these two opposite philosophies? I guess this is why there is so much discussion going on about it.

I don't see them taking away user expandable storage or RAM on the Mac Pro given the target audience. GPU options could be much more limited than before if they go with the integrated route without support for graphics cards. If they do this and want to have expandable memory, they have to forgo the unified memory model of the M1 and potentially have on-package memory act as a large L3 cache? or will it be dedicated to the GPU with expandable memory being main memory? My prediction is if they don't support graphics cards, they might stack HBM2 on package and have DDR4/5 for main memory.
 
  • Like
Reactions: B-Riz

Doug S

Platinum Member
Feb 8, 2020
2,201
3,405
136
This is something I've been wondering about, but the design of the current Mac Pro (and any other desktop honestly) is the antithesis of what the M1 is. The Mac Pro is highly modular with user replaceable RAM, storage, graphics, accelerator cards. The M1 essentially takes all of that minus the storage and expansion cards and puts it on a single package. How will Apple build a new Mac Pro with these two opposite philosophies? I guess this is why there is so much discussion going on about it.

I don't see them taking away user expandable storage or RAM on the Mac Pro given the target audience. GPU options could be much more limited than before if they go with the integrated route without support for graphics cards. If they do this and want to have expandable memory, they have to forgo the unified memory model of the M1 and potentially have on-package memory act as a large L3 cache? or will it be dedicated to the GPU with expandable memory being main memory? My prediction is if they don't support graphics cards, they might stack HBM2 on package and have DDR4/5 for main memory.

Why scrap the unified memory model? They would need to snoop between the SLC of each chiplet, though dropping the SLC on the chiplets and adding some sort of "super SLC" on the package is possible. Given how small the 32MB SLC is on the M1, they could do a 512MB "SSLC" before the die size started getting larger than the 10+2 chip is - I estimate that at around 225 mm^2, if the only change is +6 big cores -2 little cores and +24 GPU cores...though I expect it to use A15 CPU and GPU cores which may be larger.
 

Roland00Address

Platinum Member
Dec 17, 2008
2,196
260
126
Why can't you have extra ram that is in a separate pile as the unified memory?

The difference between memory and cache is not inherent stuff it is just how many layers before the cpu with the cache / memory / storage structure and what is the latency for accessing. Hell some cpus put extra ram on the PCI Express Bus.

Pretty much once you go desktop either with mac mini, or a larger "mac" you are fine with having extra pci express traces where you want to limit those stuff if you are optimizing for battery life.
 

Eug

Lifer
Mar 11, 2000
23,583
996
126
I think someone asked this already but I'm not sure we got an answer:

Why 8+2 and not 8+4? I was reading parts of this thread from last year again and just about all of us just assumed it was going to be 8+4 (although a few people tossed around the idea of a 6+4 part).

But now that I think of it, the only machine where this would matter would be the MacBook Pro. 8+4 offers little advantage for the Mac mini and iMac. It would offer some advantage for the MacBook Pro, but less so because the higher end MacBook Pros are going have more robust battery support. Truly the biggest advantage of an X+4 configuration is with the low power machines like the iPad Pro, MacBook Air, and 13" MacBook Pro.

Plus, for a Mac laptop, perhaps Apple just didn't want to go with more than 10 cores, and in the context of Pro laptop with a big battery, 8+2 makes more sense than 6+4.
 

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Why 8+2 and not 8+4? I was reading parts of this thread from last year again and just about all of us just assumed it was going to be 8+4 (although a few people tossed around the idea of a 6+4 part).

Personally I think it makes sense and phones with a gazllion small cores are a bit dumb really. When the user is doing something it's the big cores anyway and there isn't that much background action to warrant more than 2 cores, IMHO. Unless I underestimate the amount of background actions. (Ok andorid phones small A55 cores are so patehtic they might actually need 4 or more of them)
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,222
5,224
136
I think someone asked this already but I'm not sure we got an answer:

Why 8+2 and not 8+4? I was reading parts of this thread from last year again and just about all of us just assumed it was going to be 8+4 (although a few people tossed around the idea of a 6+4 part).

But now that I think of it, the only machine where this would matter would be the MacBook Pro. 8+4 offers little advantage for the Mac mini and iMac. It would offer some advantage for the MacBook Pro, but less so because the higher end MacBook Pros are going have more robust battery support. Truly the biggest advantage of an X+4 configuration is with the low power machines like the iPad Pro, MacBook Air, and 13" MacBook Pro.

Plus, for a Mac laptop, perhaps Apple just didn't want to go with more than 10 cores, and in the context of Pro laptop with a big battery, 8+2 makes more sense than 6+4.

I am betting the rumor is wrong and they have 4 lower power cores.

These are just pure rumor right now. IMO 8+4 makes more sense. If the other laptops can make use of 4 low power cores to save battery life, then so can higher end MBP. Long battery life is a significant bonus regardless of your laptop size.

Plus those low power cores are VERY tiny, taking up marginal die space.
 
  • Like
Reactions: Tlh97 and Gideon

Eug

Lifer
Mar 11, 2000
23,583
996
126
I am betting the rumor is wrong and they have 4 lower power cores.

These are just pure rumor right now. IMO 8+4 makes more sense. If the other laptops can make use of 4 low power cores to save battery life, then so can higher end MBP. Long battery life is a significant bonus regardless of your laptop size.

Plus those low power cores are VERY tiny, taking up marginal die space.
I wonder if the higher end chips will be the same clock speed. It seems Apple is being very generous with how many GPU cores they can "Chop" before declaring a chip unusable. I think the higher end chips could have a somewhat higher clock speed despite having way more cores.

How would that be reflected in the efficiency cores? Could Apple allow a comparatively even higher efficiency core clock speed, or even a partial clock speed "turbo boost" for efficiency cores in a better cooled part, while still maintaining reasonable efficiency (for a machine with a much bigger battery)?

Sorry, this is just n00b spitballin', trying to match the design to the rumour.