Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Can you elaborate on this? Why would the size of the chip have any determination on changes to the memory controller.

Asked another way, how can you determine from a technical standpoint what changes were made to the memory controller hardware from just the chips that talk to it?
The wonders of the powers of 2.

You have 2 RAM chips on the SoC. 8x 16-bit CH is then split into 2x 64-bit CH from the memory controller to each RAM chip.

8 GB - 2x4 GB with each getting RAM chip a 64-bit connection
16 GB - 2x8 GB with each RAM chip getting a 64-bit connection
24 GB - 2x16 GB with each RAM chip getting a 64-bit connection (some sort of hardware/software block so you can not address the full 32 GB only 24 GB)

Or...
24 GB - 1x16 GB + 1x8 GB with each RAM chip getting a 64-bit connection
 
Can you elaborate on this? Why would the size of the chip have any determination on changes to the memory controller.

Asked another way, how can you determine from a technical standpoint what changes were made to the memory controller hardware from just the chips that talk to it?
On the M1 line we got the following memory configurations: 8GB, 16GB, 32GB, 64GB and 128GB. All of these are a function of doubling the density of each NAND chip or adding more channels (dual channel). In the case of the MacBook line, it's one chip per channel.

In the case of the M2, we have 8GB and 16GB. But the odd one out is 24GB. 24GB can't be made by simple doubling of chip densities as that would mean the initial base storage is 12GB, which we know it's not. So, how to get 24GB? Well, there are two ways:
  • Lopsided memory (1x 8GB NAND chip and 1x 16GB NAND chip) 16+8 = 24
  • Binned down memory (2x 16GB) 16+16 = 32 but with a software limiter or hardware binning to 24GB only.
Either way this looks bad as Apple intentionally gimped memory capacity on the M2.
 
On the M1 line we got the following memory configurations: 8GB, 16GB, 32GB, 64GB and 128GB. All of these are a function of doubling the density of each NAND chip or adding more channels (dual channel). In the case of the MacBook line, it's one chip per channel.

In the case of the M2, we have 8GB and 16GB. But the odd one out is 24GB. 24GB can't be made by simple doubling of chip densities as that would mean the initial base storage is 12GB, which we know it's not. So, how to get 24GB? Well, there are two ways:
  • Lopsided memory (1x 8GB NAND chip and 1x 16GB NAND chip) 16+8 = 24
  • Binned down memory (2x 16GB) 16+16 = 32 but with a software limiter or hardware binning to 24GB only.
Either way this looks bad as Apple intentionally gimped memory capacity on the M2.
When I saw 24 GB of RAM for M2 these old nVidia video cards came to mind. That led to my 192-bit comment but even without doing the math on bandwidth, my gut again said it was 128-bit given the 100 GB/s bandwidth which is half of Pro/Max chips running 256-bit/200 GB/s etc.
 

Attachments

  • GF11620Memory.png
    GF11620Memory.png
    59.7 KB · Views: 87
  • GK104Memory.png
    GK104Memory.png
    64.6 KB · Views: 77
When I saw 24 GB of RAM for M2 these old nVidia video cards came to mind. That led to my 192-bit comment but even without doing the math on bandwidth, my gut again said it was 128-bit given the 100 GB/s bandwidth which is half of Pro/Max chips running 256-bit/200 GB/s etc.
Agreed, the 100GB/s bandwidth gave it away with the memory controller. I remember those sneaky nVidia memory configuration, heck they still happen.
 
Agreed, the 100GB/s bandwidth gave it away with the memory controller. I remember those sneaky nVidia memory configuration, heck they still happen.
I just had a flashback to my GTX 970 and 3.5 GB.

I had to get a refresher for anything newer than that but from my recollection it was awkward and not sneaky this time around.

RTX 3060 - 192-bit memory bus - 12 GB of VRAM
RTX 3060 Ti - 256-bit memory bus - 8 GB of VRAM

There were no 6 GB RTX 3060 desktop cards. There is a mobile RTX 3060 with 6 GB though.
 
On the M1 line we got the following memory configurations: 8GB, 16GB, 32GB, 64GB and 128GB. All of these are a function of doubling the density of each NAND chip or adding more channels (dual channel). In the case of the MacBook line, it's one chip per channel.

In the case of the M2, we have 8GB and 16GB. But the odd one out is 24GB. 24GB can't be made by simple doubling of chip densities as that would mean the initial base storage is 12GB, which we know it's not. So, how to get 24GB? Well, there are two ways:
  • Lopsided memory (1x 8GB NAND chip and 1x 16GB NAND chip) 16+8 = 24
  • Binned down memory (2x 16GB) 16+16 = 32 but with a software limiter or hardware binning to 24GB only.
Either way this looks bad as Apple intentionally gimped memory capacity on the M2.
Thanks for the explanation, follow up question:

1.What’s the downside of a lopsided configuration *if* everything is fast enough?

2. My initial question came about from you saying the memory controllers being “enhanced” was BS. Regardless if the capacity of the chips attached to it, how does any of what is mentioned above mean that nothing has chanced on the controller itself?

I don’t know enough about controllers, but the way I read your post seemed to indicate that no enhancements to the controller itself could have been done, but you got there by working backwards from merely capacity? Isn’t there more “under the hood” on a controller that can be revised?
 
Thanks for the explanation, follow up question:

1.What’s the downside of a lopsided configuration *if* everything is fast enough?

2. My initial question came about from you saying the memory controllers being “enhanced” was BS. Regardless if the capacity of the chips attached to it, how does any of what is mentioned above mean that nothing has chanced on the controller itself?

I don’t know enough about controllers, but the way I read your post seemed to indicate that no enhancements to the controller itself could have been done, but you got there by working backwards from merely capacity? Isn’t there more “under the hood” on a controller that can be revised?
  1. Downsides to lopsided configuration is that in theory (or at least in x86 dual channel) you loose memory connectivity speed due to the imbalance of NAND chip capacities. Not sure if that's the case with M1/M2 chips are this is the first time seeing it and we have no actual machines to run benchmarks.
  2. Apple mentioned in the presentation that with the second generation 5nm node, they could pack more transistors. That allowed them to updated the memory controller which then allowed for the up to 24GB configuration. They enhanced it to allow faster LPDDR5 but kept the same bit rate. Which means, theoretically, the M1 can have more than 16GB. Therefore, it is best to assume that Apple meant the faster data rate on memory rather than more RAM capacity.
Enhancements come from several factors, to type of connection, frequency of the bus, the bit rate, the amount of memory channel lanes and capacity per lane of addressable memory. Apple just mentioned "enhanced" controller, what that means from the presentation is, from an engineering standpoint, the updated frequency and connection. Definitely not the addressable memory. That's where I call BS. Then, they limit it? Further BS.
 
  • Like
Reactions: Eidorian
The wonders of the powers of 2.

You have 2 RAM chips on the SoC. 8x 16-bit CH is then split into 2x 64-bit CH from the memory controller to each RAM chip.

8 GB - 2x4 GB with each getting RAM chip a 64-bit connection
16 GB - 2x8 GB with each RAM chip getting a 64-bit connection
24 GB - 2x16 GB with each RAM chip getting a 64-bit connection (some sort of hardware/software block so you can not address the full 32 GB only 24 GB)

Or...
24 GB - 1x16 GB + 1x8 GB with each RAM chip getting a 64-bit connection
According to AnandTech, LPDDR allows for multiples other than 2. So it is likely two 12GB modules.
 
Well you learn something new everyday.

Given that Apple is still using just two stacks of memory, it looks like the company is finally taking advantage of LPDDR’s support for non-power-of-two die sizes (e.g. 12Gb dies), which allows them to get 12GB of memory into a single package without any further shenanigans.

Edit:

Micron has 96 Gb (12 GB) density LPDDR5.
 
Last edited:
Cool, looks like we got to the bottom of it, but we’ll have to wait for tear downs to verify.

Thanks for the lesson Jav
 
I was thinking of perhaps finally updating to a maxed out Macbook Pro 16 for 4.2k but now I'm unsure whether I should wait for a M2 macbook pro? What do you guys think? When would it come?
 
I was thinking of perhaps finally updating to a maxed out Macbook Pro 16 for 4.2k but now I'm unsure whether I should wait for a M2 macbook pro? What do you guys think? When would it come?
I love my M1 MacBook Air. I did want the MacBook Pro but the problem is all that power would just sit idle.

If you make money with your laptop and can get a tax deduction and save time, go for Pro. Otherwise the MacBook Air is great for the size.
 
I love my M1 MacBook Air. I did want the MacBook Pro but the problem is all that power would just sit idle.

If you make money with your laptop and can get a tax deduction and save time, go for Pro. Otherwise the MacBook Air is great for the size.
Yeah I would need it professionally and it would replace my older apple products that were still intel. It's a big investment either way so wanted to know what the best point is to buy....
Right now there's a 2 month wait for a macbook pro I don't want to get it in August and then M2 comes around and is much faster. Thoughts?
 
With fast (faster?) ssd speeds, might it be wiser to opt for larger storage and let the OS swap as opposed to paying memory upgrade pricing?

Are Apple's bigger SSD options faster?
 
I'm considering the 14" MBP with 16 GB RAM and 8-core M1 Pro chip. Is there a breakdown of performance/comparisons yet?

Also wanting the 14" for the extra ports, but also trying to make the smartest long-term decision. I had my last Mac for over a decade and want to get the same mileage out of whichever one I go with next.
What do you use your machine for? I have that same config in the 16” and do single-user 3d modeling of small architecture projects and I’m very happy with performance. Of course, I could have paid $1500 more and renders would be twice as fast, but at 10 minutes for full quality render, I am happy. And real-time walkthroughs while modeling are very snappy.
 
16 GB is becoming tighter for the modern web. Brave takes up easily 10-12 GB on my 32 GB machine.

The modern Web is indeed far more bloated than ever before, but we're a long way off from 16GB being tight. The 16GB you have today with modern memory management, faster memory, and fast SSD drives isn't at all like the memory of days past that you're thinking of.

Cue the avalanche of Activity Monitor captures showing zero memory free.

I'm not going to get into that. That horse has been beaten dead so many times that nobody's even bothering to correct people anymore. People don't understand what that memory chart is actually showing.

24GB is a really smart bump in RAM. Tests show that 16GB is a very ample amount of RAM on the M-series Macs, but anyone who has anxiety about future proofing can bump up half a step higher.
 
Last edited:
The modern Web is indeed far more bloated than ever before, but we're a long way off from 16GB being tight. The 16GB you have today with modern memory management, faster memory, and fast SSD drives isn't at all like the memory of days past that you're thinking of.

Cue the avalanche of Activity Monitor captures showing zero memory free.

I'm not going to get into that. That horse has been beaten dead so many times that nobody's even bothering to correct people anymore. People don't understand what that memory chart is actually showing.

24GB is a really smart bump in RAM. Tests show that 16GB is a very ample amount of RAM on the M-series Macs, but anyone who has anxiety about future proofing can bump up half a step higher.
I think what many people don’t realize is that Intel macs had dedicated ram on the GPUs, but Apple Silicon macs use the unified ram for graphics memory.

On my intel macs I extremely rarely had yellow memory pressure.

But on my m1 air I get it all the time.

Most of the time I don’t feel the impact, but frequently the mac starts swapping and is visible struggling.

Then I have to quit browsers and other memory heavy apps. Oftentimes I even have to reboot.

And that’s today. Imagine what it’s going to be like in two years when even more apps move over to electron?

And that’s a real concern. Even 1Password, an app that has been a native app for over 15 years has moved to electron.
 
  • Like
Reactions: rmadsen3
On my intel macs I extremely rarely had yellow memory pressure. But on my m1 air I get it all the time.

Likely something else was going on there. If memory is implicated, it wasn't memory alone that's your issue.

I body slammed an 8GB M1 MBP for two weeks running a full developer's workload reserved for my 32GB Intel MBP. I ran it with the sole intention of running the machine into the ground while I was taking it out for a test drive. Instead I was left utterly impressed at how well it handled being pushed to the max for two weeks straight. The memory was red the entire time I was using it and I wouldn't have known there were any problems without a stopwatch.
 
Last edited:
  • Love
Reactions: BusanAA
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.