Become a MacRumors Supporter for $25/year with no ads, private forums, and more!

8GB RAM is a disappointment?

Jouls

macrumors member
Aug 8, 2020
49
28
16gb is 200. 32gb will be 800 more. 64gb 1200 more than that. Can you say ouch? My 699 mini with 32gb ram would be 1699. With 64gb it’s 2899.
I checked shopping for the Intel mini. It‘s $ 600 more for 32GB and $ 1000 more for 64GB. But it’s still a lot.
 
Comment

mr_roboto

macrumors member
Sep 30, 2020
83
89
Referencing the section I bolded above, you miss one critical difference between the Intel/AMD use of system RAM for the iGPU and Apple's approach. Both Intel and AMD have to access the RAM via the system I/O, which is slower than the processor. Apple's approach places the RAM on the same die as the processor, which means that all reads and writes to the RAM are happening at the core speed of the processing cores.
Ok, so let's start with this one. This is an idea that seems to be floating around, and it's easy to see why people repeat it. It's an example of how, if you aren't an expert in a technical field, it's very easy to arrive at wildly wrong conclusions starting from a grain of truth.

Since it's gonna take a lot of words to explain properly, the TLDR for those who don't want to read a wall of text is: There is no reason to believe Apple has a performance advantage due to how they've packaged LPDDR4 memory in M1.


The long version:

Apple has not actually put DDR4 RAM on the same die as the processor. High density DRAM and logic are so different that the TSMC factories which make Apple's M1 SoC logic die literally cannot manufacture high density DRAM.

What Apple has actually done is to assemble DRAM into the same package as the SoC die. Each Apple M1 package has two LPDDR4 DRAM packages soldered to it. This style of SoC packaging, where DRAM packages are soldered to the SoC package, is called package-on-package, or PoP (boring name, I know). PoP has been around for quite a while in the phone and tablet world, and is used mostly because it saves board space.

Apple doesn't gain any performance from PoP. The memory is standard LPDDR4 memory, running at standard LPDDR4 speeds. Intel has processors which can use standard LPDDR4 too. Here's one that Apple's using in some of the remaining Intel MacBook Pro 13" models:


This 10th gen i7 supports 128-bit LPDDR4 running at 3733 Mbps, and the 11th gen Intel processors in its class have bumped LPDDR4 speed up to 4266 Mbps. 128-bit x 4266 Mbps is exactly the same DRAM interface width and speed as Apple M1.
 
Comment

Eric Idle

macrumors regular
Jan 4, 2020
104
52
It's 2020, 8GB of RAM is not enough if you have a lot of browser tabs (20+) and any other application open.

You are incorrect. I have 30+tabs open in Safari, 30+tabs open in Firefox and only 8gb RAM. Further, my memory pressure is reported as green by activity monitor. This is in a 2020 MacBook Air, Intel. So you are clearly wrong.
 
  • Like
Reactions: Sanpete
Comment

Stefdar

macrumors member
Feb 4, 2012
56
70
Guys, I live in Athens - Greece and the BTO Macbook Pro M1 I ordered with 16 gigs of RAM is going to take another 2 weeks to arrive, so yesterday I picked up a Macbook Pro M1 8/512. Everyone here can go on forever about the 8GB and 16GB difference but I am telling you, I am keeping this machine! 8GB in a M1 Mac is like 16GB on an Intel Mac. I am a graphic designer and I usually have a lot of programs open. Right now I have all the programs that you see in the screenshot open, I am working on a rather big psb in Photoshop native beta, two big vector files in intel Illustrator, one file in inDesign, I am watching a 4K movie in AppleTV, one huge company credentials file in keynote, etc, and the Macbook flies, no slowdown what so ever. I don't know (and don't care), if it's because of the M1 pool thing or the (amazing) speed of the M1, I do know that 8 GB in this machine acts like 16 on my previous Macbook Pro 13" high end with 16GB of RAM and quadcore i7 - 1TB SSD.
 

Attachments

  • Screen Shot 2020-11-21 at 5.32.58 μμ.png
    Screen Shot 2020-11-21 at 5.32.58 μμ.png
    780.4 KB · Views: 52
  • Screen Shot 2020-11-21 at 5.33.11 μμ.png
    Screen Shot 2020-11-21 at 5.33.11 μμ.png
    1.5 MB · Views: 49
Comment

hagjohn

macrumors 6502a
Aug 27, 2006
985
1,852
Pennsylvania
I bought a M1 Mini with 16MB of RAM (1TB SSD), just to be sure. In the past, I've gone the route of regretting not getting extra RAM and not being able to afford to change it. I know the M1 is different and much better than intel/AMD but Id rather spend the extra $200 than regret not getting it.
 
  • Like
Reactions: KayEm6419 and phl92
Comment

phl92

macrumors regular
Oct 28, 2020
109
28
Under Jobs this would have never happened. Yes, under "mediocre" devices like the Laptops of the last decade, you need options for people (browser only users, professionals)... but with that M1 I think they could have gone with one spec only just like iPhones..., letting us choose storage, screen size.. basta :D
 
Comment

phoenix-mac-user

macrumors member
Sep 21, 2016
86
74
I am giving up. I am trying to make the M1 Pro slow down, and I can't. Pay attention to my swap size...
I have been doing the same thing, watching my swap size and it has handled everything just as well as the 2018 i5 mac mini with 32GB of memory I replaced.

I am so used to having Memory anxiety as a long term Mac user that I still question my decision but it feels like i made the right move.
 
Comment

thettareddast

macrumors regular
Aug 29, 2016
130
106
Maybe Apple should do to Mac what they do to iPhone: just completely hise the RAM count, so only the ultra-nerds will complain.

(Partially kidding)

it seems that the active ram count is almost always hovering around 6gb, it says to me (in layman terms) that the OS limits the active ram use to ~6gb, leaving 2gb for critical burst use. The rest that “normally” goes to ram is stored in swap.

so its like a subclassification of ram address.

(this is the furthest thing from a technical explaination, so could be total bollocks, but from observation it seems how little or how much i apps i run, ram is always ~6gb)
 
  • Like
Reactions: Stefdar
Comment

richpjr

macrumors 68040
May 9, 2006
3,242
1,696
I am giving up. I am trying to make the M1 Pro slow down, and I can't. Pay attention to my swap size...
I wonder what constantly swapping to an SSD does to it's lifespan. Clearly the more you read and write to an SSD, the shorter the lifespan. But will it make a noticeable difference in the lifespan or will the computer be obsolete by the time that happened.
 
  • Like
Reactions: Sanpete
Comment

Stefdar

macrumors member
Feb 4, 2012
56
70
I wonder what constantly swapping to an SSD does to it's lifespan. Clearly the more you read and write to an SSD, the shorter the lifespan. But will it make a noticeable difference in the lifespan or will the computer be obsolete by the time that happened.
I had to run almost all my programs to reach that point. On my average working day I will probably not even have a swapfile worth mentioning. After all, the same thing happens when you save files, so the SSD's lifespan is what it is, you are always going to write files on it non stop, and I am sure there are safety "protocols" in place to prevent damage.
 
Comment

xWhiplash

macrumors 68030
Oct 21, 2009
2,801
1,714
I wonder what constantly swapping to an SSD does to it's lifespan. Clearly the more you read and write to an SSD, the shorter the lifespan. But will it make a noticeable difference in the lifespan or will the computer be obsolete by the time that happened.
High quality SSDs are usually 600-1000+ Terabytes written. Swapping a few GB a day doesn't really have that significant impact. For the SSD to last 5 years with the low end TBW, you would need to write over 330 GB to the drive DAILY. And it would still last 5 years. You will probably upgrade your system before the SSD dies due to writes.
 
Comment

IowaLynn

Contributor
Feb 22, 2015
1,783
417
An SDD has built-in spares. Not sure if TRIM handles that or not. Those tiny SDXC cards are more prone to fail. All of which, backup and redundancy so you are prepared yet hoping never comes to that.
 
Comment

xWhiplash

macrumors 68030
Oct 21, 2009
2,801
1,714
An SDD has built-in spares. Not sure if TRIM handles that or not. Those tiny SDXC cards are more prone to fail. All of which, backup and redundancy so you are prepared yet hoping never comes to that.
Its always good to backup your data, even with spinning hard drives. I have had a brand new spinning hard drive 1 month old die on me before :(
 
Comment

Sanpete

macrumors 68030
Nov 17, 2016
2,778
1,212
Utah
Ok, so let's start with this one. This is an idea that seems to be floating around, and it's easy to see why people repeat it. It's an example of how, if you aren't an expert in a technical field, it's very easy to arrive at wildly wrong conclusions starting from a grain of truth.

Since it's gonna take a lot of words to explain properly, the TLDR for those who don't want to read a wall of text is: There is no reason to believe Apple has a performance advantage due to how they've packaged LPDDR4 memory in M1.


The long version:

Apple has not actually put DDR4 RAM on the same die as the processor. High density DRAM and logic are so different that the TSMC factories which make Apple's M1 SoC logic die literally cannot manufacture high density DRAM.

What Apple has actually done is to assemble DRAM into the same package as the SoC die. Each Apple M1 package has two LPDDR4 DRAM packages soldered to it. This style of SoC packaging, where DRAM packages are soldered to the SoC package, is called package-on-package, or PoP (boring name, I know). PoP has been around for quite a while in the phone and tablet world, and is used mostly because it saves board space.

Apple doesn't gain any performance from PoP. The memory is standard LPDDR4 memory, running at standard LPDDR4 speeds. Intel has processors which can use standard LPDDR4 too. Here's one that Apple's using in some of the remaining Intel MacBook Pro 13" models:


This 10th gen i7 supports 128-bit LPDDR4 running at 3733 Mbps, and the 11th gen Intel processors in its class have bumped LPDDR4 speed up to 4266 Mbps. 128-bit x 4266 Mbps is exactly the same DRAM interface width and speed as Apple M1.
OK, that's a helpful clarification for those who think the RAM is on the same die as the processor or that how it's placed is a reason for an M1 advantage over Intel in RAM usage.

I think you had in mind last night to explain why @djjeff was right. He denied a basic claim at issue here, that "M1 Macs need less RAM to perform equivalently to, if not better than, their Intel-based counterparts." That's found in the article linked to below, which he dismissed without argument as nonsense. Do you disagree with the section of the article on M1 RAM usage or that claim?

 
  • Like
Reactions: MEJHarrison
Comment

mr_roboto

macrumors member
Sep 30, 2020
83
89
I think you had in mind last night to explain why @djjeff was right. He denied a basic claim at issue here, that "M1 Macs need less RAM to perform equivalently to, if not better than, their Intel-based counterparts." That's found in the article linked to below, which he dismissed without argument as nonsense. Do you disagree with the section of the article on M1 RAM usage or that claim?

I did have that in mind, yes. Unfortunately it takes a lot of time to write what I hope are good explainer posts, and it was late, so I didn't continue. I'll try to do so as I can.

I do disagree with John Gruber's article. It was disappointing to read. I don't read Daring Fireball a lot, but I have the impression that Gruber has an app software engineering background. He should know better, but then, he might not have great understanding of the stack at layers below where he works.

(For what it's worth, my position in the engineer ecosystem is that I write Verilog - the language used to design chips. This isn't as low a level as you can go, but it is the last one where there's anything resembling a high level language, and it's underneath all the software. I have significant low-level software background, too, so I'm pretty conversant with the internal architecture of operating systems, although I'm not a real domain expert there. And I'm definitely inexperienced at writing higher level code like a Mac application. Everyone has their specialty!)

So let's dissect Gruber's article.

* He starts off with the only quantitative data he's got - he links to an Apple engineer's tweet about the performance of retaining and releasing NSObject. The same one Azl just linked.

* Gruber goes on to clarify, in a way which might still be muddy to people without a technical background, that the numbers in that tweet aren't a measure of more efficient memory use, just a faster implementation of the mechanism used to track memory use in Apple's software. (I wrote a post about this earlier in the thread.) Accelerating retain/release is important, and a really big clue as to why M1 is so fast, but does nothing to reduce the amount of memory a program needs.

* Gruber starts segueing into our main topic of "You can get more done with less RAM" by saying that the NSObject retain/release performance shows that Apple co-designs hardware and software. And I agree, they do have a tight feedback loop between those two parts of their company.

So what we're now hoping for is that Gruber will go on to introduce some kind of evidence that this co-design has resulted in lower memory consumption. But that evidence never arrives. At best he just repeats anecdotes of people being excited that they replaced an Intel Mac with lots of memory with a M1 Mac that has less, and the M1 Mac was still faster. But this doesn't show that M1 has memory-use-reducing magic! It just means that it's fast, and that in that particular load the penalty of not having a lot of RAM wasn't enough to overcome M1's performance advantage. (Note that the penalty can be nothing at all. A lot of people seem to have formed unrealistic ideas of how much RAM they need to do specific things.)

* He mentions that Android uses garbage collection and iOS uses reference counting, and that's one big reason why an iPhone needs less RAM than an Android phone. This is true! Of the two memory management techniques, GC tends to be more wasteful of RAM. But it's a non sequitur here. We're only concerned with macOS on Intel versus macOS on M1, and the frameworks aren't going to count references any harder on M1 than they do on Intel. Apple has given us no reason to believe that memory management at this level is any different on M1.

* Gruber claims that because I/O is faster on M1 than Intel Macs, when you run out of memory and use swap, it's faster. The claim that I/O is faster seems to be true... as long as you're not looking at the upper end of the Intel Mac lineup. Apple has been shipping SSDs comparable to M1 SSD performance for years, just not in entry level models like the Air.

Furthermore, we're still looking at ~3.4 GB/s performance in the very biggest 2TB M1 SSDs (you won't get that with the baseline 256GB SSD, since SSD performance usually scales up as you add more flash memory). M1 can talk to DRAM at about 58 GB/s, according to the testing done by technical reviewers. (It's probably worse than these throughput numbers suggest, too. What really matters to a system under heavy swap load is disk access latency, and there's good reasons to expect that the ratio between DRAM and flash latency is even worse than the throughput ratio. But it's hard to find latency measurements of Mac SSDs, so I can't quantify this for y'all.)

So improved swapping performance isn't really a satisfactory answer. Disk speed is still way worse than DRAM speed, and equivalent disk speed is available on Intel models.

* Gruber mentions that he has a lot of stuff open on his 16GB review unit, and everything's smooth. But that's not an impossible result for Intel Macs either. In both cases, even if you've gone well past the limits of your available RAM, as long as what we low-level guys like to call the "working set" (the subset of virtual memory which is actually being read or written frequently enough to matter) fits into the physical RAM you have, you'll be doing just fine. You might notice the occasional minor delay here and there, such as when you switch to an app that hasn't been foregrounded in a long time, so it has to swap things in to begin responding quickly again, but overall the system can reach a stable state where it's not actively swapping all the time even though there's gigabytes of data in the swap file.

This is what the "Memory Pressure" graph is all about. It's Apple's attempt at measuring, in a rough sense, whether the system is under the kind of memory starvation and memory access pattern that produces high continuous swap activity, and therefore makes your computer feel slow.

A common feature of the screenshots people make where they try to prove that M1 needs less RAM is that, if Activity Monitor is visible and on the memory tab, sure they've got lots of swap used, but memory pressure is in the green. That means dynamic requirements are low, so the system isn't actively moving lots of data in and out of swap all the time.
 
Comment
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.