M1 Owners: Is 16GB the same as Intel or has it better memory management?

[NJRonbo]

I really didn't want to buy the recently announced Mac Mini.

I had hoped a refresh of the Intel version was around the corner and that there would be a pro model of the Mini introduced this year that allowed more than 16GB ram. That doesn't look like it's going to happen in 2021.

So, I caved in and pre-ordered myself a 16GB with 1TB drive. It arrives in two weeks.

For most people, 16GB would be fine. Me, I use computers differently than most. I have over two dozen apps open at all times, mostly in the background for instant access and information. I am attaching a screenshot of how memory is currently being used on my 64GB 16" MBP and you can see, I use a lot of it.

How does 16GB fare on a M1? Is it basically the same as Intel or is there better memory management that makes 16GB the new 32GB (for example)?

 

[CheesePuff]

RAM is RAM, and it will not behave differently on the M1 vs Intel in how RAM is managed. However, since the M1 uses Unified Memory, it doesn't need to allocate 1.5 GB right off the top solely to the iGPU, and when it comes to swap on the storage drive its faster then the read/write of the Flash memory of the models it replaced

 

[casperes1996]

Bear with me as I try and illustrate something with code.

typedef struct Point {
    UInt32 x;
    UInt32 y;
} Point;

int main(int argc, char** argv) {
    Point *p = malloc(sizeof(Point)*100);
    printf("How much space is heap allocated?");
    return 0;
}

It does not matter if you compile and run that on x86, ARM or any other architecture, it will always be a minimum of 6.4Kb allocated. Two 32 bit unsigned integers in a Point, and 100 points. There is no way for the CPU architecture to change the laws of physics on that.

The only thing is that for a memory use-case like you describe where most of the memory is not under active use; The apps are sitting open and thus in memory, but you're not interacting with them and they may be idle and the CPU thus never needing that memory, you'd likely be able to page it to disk without really noticing. For one thing the SSD is super fast, and for another the M1 has a very wide OOO buffer for memory and disk operations to be queued way ahead of time while it works on other instructions.

 

[NJRonbo]

I appreciate the responses!

So, basically, looking at the screenshot I provided above, there is no way I can get away with that kind of app consumption on the 16GB M1 Mini...right?

I'll just have to make the Mini a "light" version of the 16"MBP

 

[UBS28]

ARM smartphones even ship with 16 GB RAM memory these days. Macrumors are smoking some good stuff if they believe 8GB of data loaded in RAM is 2 GB on ARM.

Also it is kinda of funny if you buy an ARM laptop that has half the memory of some ARM smartphones.

When I order a 16” M2X MBP later this year, it will definetly not be with 8GB of RAM.

 

[casperes1996]

I appreciate the responses!

So, basically, looking at the screenshot I provided above, there is no way I can get away with that kind of app consumption on the 16GB M1 Mini...right?

I'll just have to make the Mini a "light" version of the 16"MBP

All I can really say definitively is that you will have paging to disk. Whether or not that will really be a big performance hit, or be fairly invisible to you depends on how speed sensitive the data retrieval is.

 

[leman]

So, basically, looking at the screenshot I provided above, there is no way I can get away with that kind of app consumption on the 16GB M1 Mini...right?

You won’t know without trying it out. First, if you have more RAM the OS will use more of it to aggressively preload data, even if it’s not strictly necessary. So looking at your screenshot, I’d conclude that your Intel machine would do just fine with 32GB, or maybe even with 16GB. Second, M1 seems to be excellent at memory management. It memory compression and swap latency appear to be significantly better than on Intel machines. People have reported that they were able to run quite memory-heavy workloads on these Macs without any performance issues.

If your RAM need boils down to multi-tasking, I’m quite confident that M1 will be able to swap your background apps in and out faster than you will be able to notice.

 

[NJRonbo]

Okay, so I am going to try loading all of this on a 16GB M1 Mac Mini when it arrives in two weeks and I'll let you know how it fares.

I noticed there is a new import tool during initial OS setup that allegedly allows you to wirelessly copy over all your information from another Mac? I know iCloud information is ported over, but is it now possible to do a wireless transfer of all your apps from (for example) a 16" MPB to M1 Mini?

 

[Chozes]

Not 100% sure here but I have had performance issues on my 8GB Mac Mini. I still love it though. Its tiny and silent.

Switiching programs with heavy multitasking causes big and medium pauses. After several seconds things are ok again but responsiveness is rough.

My 8 core Ryzen 4900HS Asus G14 laptop stays smooth in these circumstances but thats a different category of laptop to the M1 stuff. My current stance is to wait for more powerful Apple hardware coming and use the Mac Mini M1 in the mean time.

 

[jdb8167]

Okay, so I am going to try loading all of this on a 16GB M1 Mac Mini when it arrives in two weeks and I'll let you know how it fares.

I noticed there is a new import tool during initial OS setup that allegedly allows you to wirelessly copy over all your information from another Mac? I know iCloud information is ported over, but is it now possible to do a wireless transfer of all your apps from (for example) a 16" MPB to M1 Mini?

Migration assistant has been around on Macs for many generations of macOS. It copies user data, Library settings, Applications and more. In general, it is pretty reliable. For switching from Intel to the M1, you should go through all the imported apps and check to see if a native Apple Silicon version is available but all the Intel versions should run without too many problems through Rosetta2.

 

[NJRonbo]

Migration assistant has been around on Macs for many generations of macOS. It copies user data, Library settings, Applications and more. In general, it is pretty reliable. For switching from Intel to the M1, you should go through all the imported apps and check to see if a native Apple Silicon version is available but all the Intel versions should run without too many problems through Rosetta2.

JD,

I am aware of the migration. I used to connect cables between a new and old Mac to transfer stuff over.

I thought I had seen a screenshot where it is now wireless whereas you simply put one device next to the other and it imports everything over wirelessly. Am I right about that?

I am not concerned about non-native silicone apps as Rosetta 2 should take care of that compatibility, yes?

 

[Apple Knowledge Navigator]

I feel this topic has been beaten to death already.

RAM is RAM. It doesn't matter whether you're accessing memory from an M1 Mac or ZX Spectrum - the principles are still the same. It's an allocation of memory that can be randomly accessed.

The reason so much fuss has been caused with M1 RAM is because of how it's accessed.

With the unified memory architecture - combined with the sheer speed of the M1's single core performance - it would appear that users are generally quite interested in pushing their system to the limit as real world tasks are so quick.

Put simply; RAM appears to behave differently because the speed differences between page swaps and RAM access are almost imperceptible now.

Think of it as a warehouse being next-door to a retail shop, compared to miles away in a different town. When stock is needed, the former can just walk by and get it, whereas the latter would need to travel longer distances at the cost of longer time away.

Just buy the RAM in account for what you may need in the future, just as you always have.

 

[jdb8167]

JD,

I am aware of the migration. I used to connect cables between a new and old Mac to transfer stuff over.

I thought I had seen a screenshot where it is now wireless whereas you simply put one device next to the other and it imports everything over wirelessly. Am I right about that?

I am not concerned about non-native silicone apps as Rosetta 2 should take care of that compatibility, yes?

I think you are confusing iOS and macOS. Migration assistant is wireless over WiFi but you have to answer a few questions when setting up. It is pretty painless though.

 

[the8thark]

Put simply; RAM appears to behave differently because the speed differences between page swaps and RAM access are almost imperceptible now.

That's the main takeaway here. Swapping between accessing RAM and accessing virtual memory (SSD space used as Memory) is very quick these days. Also compared to the old HDD days, virtual memory these days is much better due to it being on high quality SSD in the M1. Virtual memory will only get better in the future as the quality and speecds of SSD's improve over time.

I would also say how the M1 dynamically adjusts how much RAM is allocated to each application is even faster and more effective too.

 

[JohnE121]

I am totally new to the Mac world and hope I don't get flamed too badly with this post. The video link below is a person comparing the Mac mini to a $10K+ Mac Pro. Exactly 1 minute into the video he states that 48G of RAM in the Mac Pro is comparable to 16G in the Mac mini. Is anyone able to explain this statement? I did a little research on unified ram and the closest explanation I found is that unified RAM does not have to make multiple copies of existing data in RAM for other processes.

 

[Joelist]

The thing is that there are some unusual behaviors seen in M1 with regards to RAM. I noted in my new MBA thread that (for example) the Apple Silicon versions of Office 365 programs (actually Apple Universal) running the same files had smaller sizes in RAM. Plus to be honest Big Sur seems to handle RAM and swap differently both in Intel and Apple Silicon - it appears to be swapping more readily instead of keeping whole applications loaded up in the RAM.

 

[Rob2020]

I have had the same system you’ve ordered since Christmas (16GB, 1TB) and while I can’t argue with the basic logic that a byte is a byte, I can offer this personal observation. I have used this Mac Mini fairly heavily: Word, Excel, PowerPoint, Affinity Photo, Luminar 4 and AI and DaVinci Resolve, plus a host of smaller apps. I’m not literally running all those at one time, but I run several at a time with lots of Safari windows open all the time, plus mail, reminders, iMessage, etc. I’ve rendered moderately complex one hour videos in Resolve while working with Word and Affinity or browsing and with a few others open, but not being used. It never seems to struggle, and as I look back at my RAM usage over the time I’ve had it, it usually Has about 6GB free and it has never used swapfile space. No one can tell you ahead of time whether your particular usage will overwhelm this system, but I am certainly finding this sufficient for my moderate needs. I can’t explain the technical side of things, but this system seems extraordinarily efficient in how it uses it’s resources. I’m not saying it doesn’t have limits, but I suspect you will be pleasantly surprised.

 

[NJRonbo]

I feel this topic has been beaten to death already. RAM is RAM. It doesn't matter whether you're accessing memory from an M1 Mac or ZX Spectrum - the principles are still the same. It's an allocation of memory that can be randomly accessed.

I am sorry that this topic has frustrated an individual.

Actually, this hasn't been beaten to death when more and more individuals, as clearly shown in this thread, are posting the kind of real-life experiences I was looking for on how memory is handled differently by the M1 devices and Big Sur.

 

[leman]

RAM is RAM. It doesn't matter whether you're accessing memory from an M1 Mac or ZX Spectrum - the principles are still the same. It's an allocation of memory that can be randomly accessed.

This is technically accurate but not necessarily helpful RAM is RAM — but certain properties of a specific implementation can make significant practical differences. You say so yourself - page swaps appear to be extremely fast on M1 hardware, so you can actually get away with less RAM. You only really need enough RAM to supply the active task. This is not necessarily the case with Intel machines, which appear less efficient with RAM juggling.

Just buy the RAM in account for what you may need in the future, just as you always have.

This again is a good advice, but the benchmark might have changed. Users normally operate under a common set of assumptions that are based on some sort of "common knowledge". Say, it is a common knowledge among Resolve users that to do 4K video edits you need 32GB of RAM and they have some popular blog posts to show the difference. But now M1 users are doing the tests and they find out that they can do comparable work, with the same (or better) performance, while only using 8GB of RAM. That's the gist of it, its "buy what you need", but "what you need" might or might not have changed.

That's the main takeaway here. Swapping between accessing RAM and accessing virtual memory (SSD space used as Memory) is very quick these days.

The interesting bit is that swapping appears to be faster on M1 than on Intel machines using the same SSD. Maybe integration of the SSD controller into the SoC has dramatically improved the latency (haven't seen any benchmarks on this). Maybe Apple has hardware-accelerated memory compression. Maybe page management itself is faster on M1. Who knows. But they certainly did something and memory management does feel different.

The video link below is a person comparing the Mac mini to a $10K+ Mac Pro. Exactly 1 minute into the video he states that 48G of RAM in the Mac Pro is comparable to 16G in the Mac mini. Is anyone able to explain this statement?

I would assume that it means that he can do a certain type of work on a 16GB M1 Mini that he would usually need 48GB of RAM on his Intel machine.

I did a little research on unified ram and the closest explanation I found is that unified RAM does not have to make multiple copies of existing data in RAM for other processes.

Machines with dedicated GPUs might need to keep around extra copies but if your processing is CPU-driven, there should not be any difference between Apple's unified RAM and RAM as we know it. From the CPU standpoint, RAM is already "unified" (unless you are using a multi-CPU machine).

 

[giffut]

I am pretty sure that the upcoming models will allow higher RAM ceilings. I do treat the first generation ARM macs as is and will care only for the basic setups, so we have room to invest in future models. We just came back to macos after the illfated 2016ff series hardware, which was not very reliable, performance limited and very expensive for our budget.

I stopped using huge internal SSD sizes and externalize storage like NAS and attached devices, anyway, as you can´t swap internal storage when it fails. And that is a too expensive lesson I don´t want to learn firsthand.

We have a macbook air and mac mini with 8GB RAM/256GB SSD, they perform perfectly well. When you strictly close apps you don´t need at the moment, you have plenty of storage available at any time - that at least is our experience. This includes apps like Final Cut, Davinci Resolve, Lightroom Classic and Ableton.

They are very fast, battery runtime is double the best pc laptop parts, although we still use one very successfully - a Dell latitude 2in1 5290. The macbook air is more finiky than the mac mini, though. We already went through a DFU firmware restore.

At the moment, for our workloads, they are very workable, especially regarding the kids and remote schooling.

 

[Kung gu]

ARM smartphones even ship with 16 GB RAM memory these days. Macrumors are smoking some good stuff if they believe 8GB of data loaded in RAM is 2 GB on ARM.

Also it is kinda of funny if you buy an ARM laptop that has half the memory of some ARM smartphones.

When I order a 16” M2X MBP later this year, it will definetly not be with 8GB of RAM.

The 16" intel model comes with 16gb of RAM. So u can't even get 8gb. lol

 

[armoured]

I am aware of the migration. I used to connect cables between a new and old Mac to transfer stuff over.

I thought I had seen a screenshot where it is now wireless whereas you simply put one device next to the other and it imports everything over wirelessly. Am I right about that?

Yes, migration assistant will work over ethernet cables (need a dongle on the M1 macs), wirelessly, or from a disk.

Reports are pretty positive generally that migration assistant works well over ethernet/wifi in the most recent macos/M1 macs. But note, usually ethernet/wifi are a bit slower, and some wifi networks are less stable; if it 'hangs' in the process you end up with a partial migration and uncertain how much was done (and so you probably end up starting from scratch).

What I've always preferred to do is run migration assistant either from an external disk - complete time machine backup or a full clone of the source computer's system disk (I use carbon copy cloner). Hook the external drive up to your new computer, point migration assistant to the source disk, and run it. In my experience much more predictable and reliable but YMMV. (This was with previous macos versions and older computers).

[There are complications and work needed if your time machine backup includes multiple disks and especially if the source disk is too big for the new computer]

 

[Moi Ici]

I have a 12” MacBook with 16GB memory and 512 ssd. I just got a MacBook (M1) pro 16GB and 512 ssd. Now I know the MacBook (2015) is ancient and pretty underpowered compared to the M1. BUT, I could hardly edit photos on the 12”, constant beach balls & freezes (with every other app closed). I could hardly import my photos off a memory card (had to breakdown the imports to 2 to 3 at a time. This was painfully slow. The M1 is just blazing and I can have open (I tested) a lot of other apps at the same time. So, as a non computer person, just a big fan with said interests, the M1 does some amazing magic.

 

[NJRonbo]

Yes, migration assistant will work over ethernet cables (need a dongle on the M1 macs), wirelessly, or from a disk.

Reports are pretty positive generally that migration assistant works well over ethernet/wifi in the most recent macos/M1 macs. But note, usually ethernet/wifi are a bit slower, and some wifi networks are less stable; if it 'hangs' in the process you end up with a partial migration and uncertain how much was done (and so you probably end up starting from scratch).

What I've always preferred to do is run migration assistant either from an external disk - complete time machine backup or a full clone of the source computer's system disk (I use carbon copy cloner). Hook the external drive up to your new computer, point migration assistant to the source disk, and run it. In my experience much more predictable and reliable but YMMV. (This was with previous macos versions and older computers).

[There are complications and work needed if your time machine backup includes multiple disks and especially if the source disk is too big for the new computer]

Thank You!!!

I KNEW I saw a screenshot somewhere of a wireless migration during the new OS setup. I would just probably use a USB-C cable between both the Mini and my laptop to transfer the data.

 

[mds1256]

Bear with me as I try and illustrate something with code.

typedef struct Point {
    UInt32 x;
    UInt32 y;
} Point;

int main(int argc, char** argv) {
    Point *p = malloc(sizeof(Point)*100);
    printf("How much space is heap allocated?");
    return 0;
}

It does not matter if you compile and run that on x86, ARM or any other architecture, it will always be a minimum of 6.4Kb allocated. Two 32 bit unsigned integers in a Point, and 100 points. There is no way for the CPU architecture to change the laws of physics on that.

The only thing is that for a memory use-case like you describe where most of the memory is not under active use; The apps are sitting open and thus in memory, but you're not interacting with them and they may be idle and the CPU thus never needing that memory, you'd likely be able to page it to disk without really noticing. For one thing the SSD is super fast, and for another the M1 has a very wide OOO buffer for memory and disk operations to be queued way ahead of time while it works on other instructions.

This.... this comment needs cross posting to reddit as there are hundreds of questions about memory management.

Top marks.