I get them on my M1 - and not when it’s swapping.
Look, I grew up on a sheep farm and every time I see this subject line I want to answer that it depends on how many sheep you have… but then I realise that’s childish and keep quiet. Until now.
17 Likes
I am very happy with 8GB on M1 MacBook air.
I don’t think it is like it used to be with RAM with the M1 chip. It is lightning fast much faster than my iMac Pro with 32GB ram.
The speed increase is thanks to the speed of the SSD and the bandwidth that carries data from RAM to the SSD. So swap is faster. Your RAM is not, I don’t think, inherently faster on its own. (At least, not compared to the jump in speed of swap.)
Which is what makes this question difficult to answer, because, as others have said, the question is really this:
“At what rate does the SSD inside a modern Mac degrade? If I frequently go into swap, but it doesn’t feel slow, am I causing harm to the computer over the long term? If so, how much harm? Will I need to replace the computer sooner than I would otherwise?”
And nobody has a good answer for that beyond the normal answer, which is still that most folks could always benefit from more RAM (unless you truly never use swap), because that’s the most conservatively safe answer one could give.
The problem with the question is that it’s not as specific as one might think, and also, it’s mostly a litmus test for how conservative you are with your computing needs.
The original question would be easier to answer if it was phrased like this: “I frequently use all 32gb of RAM on my Intel Mac. Do I need to get more than 32gb of RAM for my new M-series Mac? If not, can I get away with 16gb?”
That’s still hard to answer, but at least the question is more specifically scoped.
3 Likes
Size of swap isn’t that big of a deal on SSD longevity, what’s being stored there is. The system tries to keep mutable items (user data) in memory while putting immutable items (program code) to the swap file. If memory is needed the program code is only written to the swap file once. After that it can be thrown away since it hasn’t been changed and just read from from the swap with no degradation of the SSD.
Things can get bad when the number of swap outs gets very high. A system that is thrashing is spending more time swapping items in and out than running code.
1 Like
Thank you for clarifying my mistakes!
That’s not exactly true. On a computer you always have the hierarchy of access latencies that would go in a grossly simplified way like this: CPU registers → RAM → disk While in the Intel era Apple has achieved fantastic speeds on the disk to RAM step as to make it unnoticiable with their custom non replaceable SSD, in the Apple Silicon era the memory to CPU bandwith has also gotten bigger. So it’s not like disk got faster while memory did not, it’s just that the slowest element of the chain has become so fast that it’s unnoticeable for us humans unless you run a benchmark tool.
That’s the key question. You could run a SSD disk benchmark every month to see how the performance is degrading --sure it must be! But since Apple has been shipping devices with SSD disks coupled with meager memory banks for quite some time now one could also expect that some people would have already noticed that degradation due to RAM shortage if that was the case. Probably the Apple disk controllers do lots of tricks to make sure that the degradation does not surface during the useful life of the computer.
That depends on the definition of “using”. macOS will make use of all the RAM it sees so after several days of usage almost all users will see their RAM used no matter the RAM they have. So for example if an app requests to allocate, say, 1GB RAM on a 8GB system that’s exactly what it will get assigned but on a 32GB system macOS will say: “Ok, here’s your 1 GB plus another one just in case”. And if all apps are happy with the memory and there is memory left, it will cache disk blocks into RAM until almost everything has been filled. If memory pressure increases because apps begin requesting more RAM space, the OS will start first swapping unused data segments to disk, then executable code segments, and finally it will try to swap whatever it can on least recently used basis. Beyond this the user will probably notice a performance degradation so he/she will close some apps , and then the cycle begins again.
I will have my Bill Gates moment right here, right now and claim that 640kb… er, 16GB should be enough for everyone because these things are now so ridiculously fast and powerful that our cognitive capabilities while multitasking are not that high that we cannot close some apps and reopen them after we finish some of the work we are doing. Yes, there will be specialized use cases when 16GB are not enough. Developers running VMs & containers, XCode, training neural networks, pro creative edition… yes, those cases should benefit or require more RAM but those specialized users should be able to do the math on their working sets and decide if the hefty RAM upgrade price is a “nice to have” or a “must”, usually depending on the amount of $$$ they do with their work.
7 Likes
You know, I do my sheep with more RAM power than I need too. 
I’ve only got 31 breeding ewes this year and we used 6 different rams out of a potential sire pool of 25. 
Of course maintaining genetic diversity in a rare breed is a prime directive for our farm.
6 Likes
I disagree with that. The speed difference between RAM paired with an Intel chip via the motherboard and the speed of RAM which is part of the SOC is night and day.
This is simply from a hardware perspective beyond the fact that macOS and Apple Silicon are designed to work together.
Exactly. CPU-to-RAM transfer is just like moving physical parts around in a building. The faster the CPU can get stuff to and from RAM, the faster it can do whatever it needs to do. And same with the disk. Small doors and long hallways slow things down.
From the “building” perspective, the AS transition basically made a bigger door between the “CPU” room and the “RAM” room, and significantly decreased the length of the hallway between them.
Alright all, I stand corrected. Thank you! No need to jump on my original post any further. 
5 Likes
Prepare to be absolutely shocked.
I don’t think most of these tabs use more than a few kilobytes on disk space, let alone consume substantial amounts of RAM. I can easily have 70 or so tabs open on my 8GB Mac Mini M1 split between Firefox and Safari. From observation, most of these “open” tabs are reloaded from the internet when I access them - they aren’t even populated with content when I click on them. In other words, an “open” tab seems to be little more than a database entry and a “placeholder” tab in the tab bar. Showing the “tab overview” in Safari confirms this.
Likewise, I admit to having “open” about 150 tabs on my 2019 iPad mini - that has only 3GB of RAM. And that iPad is very snappy. It only struggles when I have lots of content in very few most recent tabs (e.g. watching videos or after much scrolling down on sites that dynamically load additional content, e.g. Instagram profiles).
Browser tabs are often RAM-intensive. But having them “open” in a well-optimised browser doesn’t mean that all of them are actually loaded into RAM - or on disk.
My experience with Chrome is that over 100 tabs starts swapping pretty quick. Might just be my experience, or it might be how long they were open.
Tangential but fun quote here about RAM usage giving way to SSD as speeds increase. This rate of NVMe transfer will hopefully be in premium consumer hardware within five years.
Something similar is happening now with SSD/NVMe storage. The generational jumps there have been even larger than in CPU land. We’ve rapidly gone from speeds around ~500MB/sec with Gen 2 drives to ~2.5GB/sec with Gen 3 to ~5GB/sec with Gen 4 and now Gen 5 is promising us a ridiculous ~13GB/sec shortly! That’s the kind of order-of-magnitude leap forward that requires you to rethink your assumptions.
We’re exploring moving both caching and job queuing at Basecamp to NVMe instead of RAM. The latency is now close enough that the advantages of abundant, fast NVMe storage wins. We just bought some 12TB NVMe Gen 4 cards for $2,390. 12TB! Using the new E1/E3 NVMe form factor, we’re now looking at the possibility of a petabyte’s worth of NVMe storage in a single rack server for around $200K. That’s bananas!
1 Like
Modern browsers are sophisticated in terms of memory/CPU management. After a period of time they offload tabs that aren’t visible, saving at least 80% of RAM. Similarly JavaScript calls are limited in the background reducing CPU load. Edge is more aggressive than Chrome by default… I don’t know about Safari (Although I’ve never had RAM problems on my 8Gb machine, despite being tab hoarder!). The effect on SSD will be negligible, similar to the caching functionality browsers already have.
The overall effect on RAM usage is a combination of how many tabs were opened recently (before the timeout) and the content of those tabs.