RAM

[ivnj]

I just found this article talking about the new M2 Macs.

Apple unveils M2 Pro and M2 Max: next-generation chips for next-level workflows

Supercharging MacBook Pro and Mac mini, M2 Pro and M2 Max feature a more powerful CPU and GPU, up to 96GB of unified memory, and power efficiency.

www.apple.com

It says

"Supercharging MacBook Pro and Mac mini, M2 Pro and M2 Max feature a more powerful CPU and GPU, up to 96GB of unified memory, and industry-leading power efficiency"

That is a lot. I am just curious what program could possibly use 96GB of RAM? Are editing programs and games really getting that hungry?

 

[t0mat0]

Reading footnote 2 - looks like OTOY Octane X 2 was a Mac application they highlighters for starters. “systems were tested using a scene that requires over 40GB of graphics memory when rendered.”

with unified memory architecture as far as I can tell a chunk of the total GB of memory can be effectively GPU memory.

Hopefully reviews and commenters can better highlight what more memory can do considering the high price to add more and the up to 96GB a MBP can go to now.


The press release covers powering visual effects, training machine learning models, stitching together gigapixel images - I‘m pretty sure iJustine used Metashape for her Max Pro review - whether it can fully utilise RAM and is native I don’t know - would be an interesting compare vs a Mac Pro.

MacBook Pro

MacBook Pro laptop with M4, M4 Pro and M4 Max chips. Built for Apple Intelligence. Up to 24 hours of battery life. Liquid Retina XDR display.

www.apple.com

 

[estabya]

There are lots of workloads from photo/video/audio editing tasks, various virtualization environments, mathematics/physics uses, or even intensive coding/compilation tasks that require immense amounts of memory.

Basically, if you're the target market for a MBP with 96GB of memory, you'll know it 😂

 

[chill991]

I ordered 64gb, but did think about 96gb. Passed since price was getting crazy. My workflow has me running multiple VM's. My current MBP has 32GB so doubling the memory will be very welcome.

 

[Ethosik]

Adobe eats everything you throw at it. I literally have a 720 project in Adobe that fills up my 128GB of RAM test Windows system at one time. I have a lot of plugins and effects on that video, and its an older one which is why its 720. But the project runs just fine on 8GB of RAM too. But if you absolutely need that level of cache for some Adobe apps, then even 128 or 256 or higher is needed.

 

[ivnj]

I ordered 64gb, but did think about 96gb. Passed since price was getting crazy. My workflow has me running multiple VM's. My current MBP has 32GB so doubling the memory will be very welcome.

How much was the 64GB and how much is 96GB? And is the M2 mac mini soldered like the M1? Or is the ram upgradable in the new M2 mini?

 

[theorist9]

How much was the 64GB and how much is 96GB? And is the M2 mac mini soldered like the M1? Or is the ram upgradable in the new M2 mini?

All Apple Silicon RAM is currently soldered. [There's been speculation that the upcoming Mac Pro may have upgradeable RAM, but that's not known.]

Here are the current RAM limitations:

M2 (Mini, Air, and 13" MBP): 24 GB
M2 Pro (Mini and 14"/16" MBP): 32 GB
M2 Max with 30-core GPU (14"/16" MBP): 64 GB
M2 Max with 38-core GPU (14"/16" MBP): 96 GB

The fact that the M2 goes to 24 GB instead of 16 GB, and the M2 Max with 38-core GPU goes to 96 GB instead of 64 GB, means Apple offers high-capacity RAM modules on those models. As to why they were offered only on those models, perhaps Apple thought it was at the low end and the high end that the extra capacity was most needed. And as to why they weren't offered on all models, perhaps those modules were in limited supply, or perhaps Apple wanted to limit the number of configurations they offered to simplify their supply chain.

You can check the prices for your country on Apple's website.

 

[fakestrawberryflavor]

There is a common misinterpretation that you can run less ram on Mac as opposed to a PC. Sure, MacOS is efficient at resource management but, with the introduction of the unified memory concept, you actually likely need MORE ram than you did before. The reason is because video memory which would have its own physical memory in a PC or even an x86 based Mac, now doesn’t exist. It’s shared with the CPU memory. All memory shares the same pool. If you are running lots of programs that need ‘regular memory’ and graphics workloads that use ‘video memory’ it’s all taking space up in the unified memory together.

 

[apostolosdt]

The above posts are very informative regarding apps on the pro side. I use two Mac Pros, all with 64GB RAM, for writing, photo editing, and symbolic math work. So, from that viewpoint, my experience indicates that the 64GB are still not enough for Adobe software, very helpful with virtualization (I run Windows 10 and Ubuntu on Parallels), but most of all indispensable for peace of mind: I don't have to worry when I open yet another app while lots are already running.

 

[sam_dean]

There is a common misinterpretation that you can run less ram on Mac as opposed to a PC. Sure, MacOS is efficient at resource management but, with the introduction of the unified memory concept, you actually likely need MORE ram than you did before. The reason is because video memory which would have its own physical memory in a PC or even an x86 based Mac, now doesn’t exist. It’s shared with the CPU memory. All memory shares the same pool. If you are running lots of programs that need ‘regular memory’ and graphics workloads that use ‘video memory’ it’s all taking space up in the unified memory together.

This is why I'd only buy a iMac M2 Pro with 32GB as my Intel iMac has 32GB.

 

[ivnj]

Yes very informative. I was just curious. I remember ram coming in K. Then MB. Now GB. Someday in the future everyone will complain that even 1TB of ram is not enough and is sluggish.

 

[leman]

There is a common misinterpretation that you can run less ram on Mac as opposed to a PC. Sure, MacOS is efficient at resource management but, with the introduction of the unified memory concept, you actually likely need MORE ram than you did before. The reason is because video memory which would have its own physical memory in a PC or even an x86 based Mac, now doesn’t exist. It’s shared with the CPU memory. All memory shares the same pool. If you are running lots of programs that need ‘regular memory’ and graphics workloads that use ‘video memory’ it’s all taking space up in the unified memory together.

This is a great example of fallacious thinking. You do not get significantly more useable RAM on a traditional PC with a dGPU. Much of the GPU resources will be mirrored in the system RAM. The GPU RAM does not exist to "free up" the system RAM. It exists to improve the performance of the GPU.

The "you can run less RAM on a Mac" thing you mention is definitely an overblown statement, having it's roots in practical experiments conducted by some reviewers that illustrate that Apple Silicon behaves better than Windows in memory-starved situations. In other words, once you ran out of RAM you will still probably be able to use your Apple Silicon Mac more or less effectively, which is not the case for a Windows PC. This is most likely the effect of larger RAM pages, built-in hardware memory compression as well as optimised swapping protocols on Apple Silicon. Of course, neither of these things will magically turn 16GB into 32GB.

 

[Carlson-online]

But apple, why can't I just have an M2 Pro chip with 24GB RAM ?! Happy to pay +180 for it, but +360 for it is too much, my load usasge is ~ 20GB

 

[Alex Cai]

How much was the 64GB and how much is 96GB? And is the M2 mac mini soldered like the M1? Or is the ram upgradable in the new M2 mini?

8-16 is enough for daily tasks, 16+ is for pro workflows.
All applesilicon don’t have upgradable ram since it’s literally a part of the soc

 

[jdb8167]

All applesilicon don’t have upgradable ram since it’s literally a part of the soc

RAM is in the same chip carrier as the SoC but is not part of the chip. Still makes it hard if not impossible to allow user upgrades.

 

[Digitalguy]

This is a great example of fallacious thinking. You do not get significantly more useable RAM on a traditional PC with a dGPU. Much of the GPU resources will be mirrored in the system RAM. The GPU RAM does not exist to "free up" the system RAM. It exists to improve the performance of the GPU.

The "you can run less RAM on a Mac" thing you mention is definitely an overblown statement, having it's roots in practical experiments conducted by some reviewers that illustrate that Apple Silicon behaves better than Windows in memory-starved situations. In other words, once you ran out of RAM you will still probably be able to use your Apple Silicon Mac more or less effectively, which is not the case for a Windows PC. This is most likely the effect of larger RAM pages, built-in hardware memory compression as well as optimised swapping protocols on Apple Silicon. Of course, neither of these things will magically turn 16GB into 32GB.

You may want to quote your sources for the experiments you mention, especially to see how much difference there is... and also what systems were used, speed of SSDs for paging etc.

 

[leman]

You may want to quote your sources for the experiments you mention, especially to see how much difference there is... and also what systems were used, speed of SSDs for paging etc.

It was an impressionistic stuff done by some YouTubers, don’t remember who. Nothing rigorous or quantifiable really, but the was clearly a noticeable difference in how the two systems behaved under similar load. Unless it was manipulated of course.

 

[Digitalguy]

There is a common misinterpretation that you can run less ram on Mac as opposed to a PC. Sure, MacOS is efficient at resource management but, with the introduction of the unified memory concept, you actually likely need MORE ram than you did before. The reason is because video memory which would have its own physical memory in a PC or even an x86 based Mac, now doesn’t exist. It’s shared with the CPU memory. All memory shares the same pool. If you are running lots of programs that need ‘regular memory’ and graphics workloads that use ‘video memory’ it’s all taking space up in the unified memory together.

Great point I have pointed out the hugh BS that some youtubers say on their channels like "8GB of unified memory is as good as 16 on Intel Macs", while it makes no difference at all and if anything sharing memory means less memory for the CPU, although not much of a difference in that case either. A bigger difference with Intel Macs is faster paging so you don't see the effects of running out of memory, but that's due to SSD speed, not to RAM, and to better Apple Silicon CPU ability to compress stuff in memory compared to Intel CPUs, but again that's not due to RAM but to a more capable CPU without the user noticing as much the difference.

 

[Digitalguy]

It was an impressionistic stuff done by some YouTubers, don’t remember who. Nothing rigorous or quantifiable really, but the was clearly a noticeable difference in how the two systems behaved under similar load. Unless it was manipulated of course.

So, in essence nothing verifiable. My guess? What really makes a difference is SSD speed, something that applies to Windows too. And the way the OS manages to use the CPU to compress workloads in RAM, so basically software. My guess is MacOS is more efficient than Window at that. And Apple Silicon is one of the best CPU so that helps. But Windows could improve memory compression and high end x86 CPUs are as good as Apple Silicon at this point (although less power efficient).
Having said that, saying that it's due to Unified Memory it's simply not true, as it's more a matter of SSDs and software optimization.

 

[leman]

So, in essence nothing verifiable. My guess? What really makes a difference is SSD speed, something that applies to Windows too. And the way the OS manages to use the CPU to compress workloads in RAM, so basically software. My guess is MacOS is more efficient than Window at that. And Apple Silicon is one of the best CPU so that helps. But Windows could improve memory compression and high end x86 CPUs are as good as Apple Silicon at this point (although less power efficient).
Having said that, saying that it's due to Unified Memory it's simply not true, as it's more a matter of SSDs and software optimization.

It’s definitely not due to unified memory. Intel has had unified memory since 2012 or so, and Intel Macs don’t show the same behavior as the M1 series. It’s not the SSD either as the access latency of newer SSDs are not any better than what we had before.

The “8GB is same as 16GB” is obviously total nonsense, but the experience still suggests that Apple Silicon seems to be more responsive under high memory pressure. As I wrote before, I think that it boils down to the larger memory pages (less overhead when swapping), hardware memory compression (more mileage from physical RAM) as well as optimizations of SSD paging. Hardware memory compression alone means that you can get a few more GBs of effective RAM with only a minor slowdown if you combine it with a competent thread scheduler.

 

[falainber]

This is a great example of fallacious thinking. You do not get significantly more useable RAM on a traditional PC with a dGPU. Much of the GPU resources will be mirrored in the system RAM. The GPU RAM does not exist to "free up" the system RAM. It exists to improve the performance of the GPU.

The "you can run less RAM on a Mac" thing you mention is definitely an overblown statement, having it's roots in practical experiments conducted by some reviewers that illustrate that Apple Silicon behaves better than Windows in memory-starved situations. In other words, once you ran out of RAM you will still probably be able to use your Apple Silicon Mac more or less effectively, which is not the case for a Windows PC. This is most likely the effect of larger RAM pages, built-in hardware memory compression as well as optimised swapping protocols on Apple Silicon. Of course, neither of these things will magically turn 16GB into 32GB.

Why would GPU resources be mirrored in system RAM? These resources are needed only by GPU. It's possible that the program could use system RAM temporarily to load/prepare GPU resources, but once they have been sent to GPU what's the point of keeping them permanently in main RAM? Sloppy programming?

 

[leman]

Why would GPU resources be mirrored in system RAM? These resources are needed only by GPU. It's possible that the program could use system RAM temporarily to load/prepare GPU resources, but once they have been sent to GPU what's the point of keeping them permanently in main RAM? Sloppy programming?

GPU memory is limited, so you need to swap things in and out as needed. Also, many data buffers are synchronised between the CPU and the GPU, so you need to keep then in both memory pools. This was particularly bad with OpenGL where the driver had to keep pretty much everything in the system RAM, it is possible that it got better with newer APIs where memory management is more low-level, but this does not preclude the necessity of caching the data in the system RAM.

 

[Appletoni]

I just found this article talking about the new M2 Macs.

Apple unveils M2 Pro and M2 Max: next-generation chips for next-level workflows

Supercharging MacBook Pro and Mac mini, M2 Pro and M2 Max feature a more powerful CPU and GPU, up to 96GB of unified memory, and power efficiency.

www.apple.com

It says

"Supercharging MacBook Pro and Mac mini, M2 Pro and M2 Max feature a more powerful CPU and GPU, up to 96GB of unified memory, and industry-leading power efficiency"

That is a lot. I am just curious what program could possibly use 96GB of RAM? Are editing programs and games really getting that hungry?

Fritz 18 GUI when using 7-piece syzygy endgame tablebases.
ChessBase 17 program when doing Mega 2023 database stuff.
Stockfish chess engine.
...

 

[Appletoni]

GPU memory is limited, so you need to swap things in and out as needed. Also, many data buffers are synchronised between the CPU and the GPU, so you need to keep then in both memory pools. This was particularly bad with OpenGL where the driver had to keep pretty much everything in the system RAM, it is possible that it got better with newer APIs where memory management is more low-level, but this does not preclude the necessity of caching the data in the system RAM.

Swap is not possible because the 8 TB SSD on my MacBook Pro 16-inch M1 MAX is full.
That's why Apple should best offer 16 TB SSD and 128 better 256 GB RAM.