ssd swap - high usage of Terabytes Written

[jeanlain]

None of the possible causes of excessive writes appears specific to Apple Silicon, except for Rosetta2, which may increase RAM usage hence swap.
The fact the most people reporting here use M1 Macs could be due to the fact that this is the AS Mac sub-forum and that, for some reason, the issue was first reported on an M1 Mac (and attributed to the M1 without clear evidence), which initiated a chain reaction.

I believe that this potential issue should not condition the choice between an Intel and an M1 Mac, unless some must make heavy use of ram-hungry Intel apps, maybe.

 

[majormike]

You don't get it. Swapping the SSD is effectively the same as using an external.


Non Sequitur comparison as your complaint is basically 'how do I replace the soles of my shoes rather then getting another pair of shoes'? Also you dodged the question of where are the 3 years coming from?

I would ask what were you doing to have it run hot enough to fry your RAM like a egg? My 2008 iMac has the same memory it had as does the 2013 one I am typing this on. Never mind the whole reason you get a Mac Pro is for the upgradability.

Somebody on Twitter said that their friend's M1 failed after somewhat 600 TB of written Data, which scared the **** out of me.

The 2008 Mac Pro had that issue, Ram running too hot, it wasn't an issue in my 2012 12-Core Mac Pro, which is still running fine till this day with the same ram but you can lose the lottery sometimes, that's what I'm saying.

And regarding external storage being equal to internal I have to disagree, even with TB3, you're gonna have a higher latency, less read and write speeds and generally less durability due to it not being sufficiently cooled in an external enclosure.

 

[osplo]

The real biggest issue is most people are freaking out on the raw TB and not the percentages.

Lonehorn shows 1% in about 2 months - that is 200 months (16.6 years) for 100%.

Dan Moren's numbers are even better as 1% in 3 months is 300 months or 25 years.
His 2017 iMac's 14% in 42 months (3.5 years) produces the same life span - 300 months (42*100/14).

These are examples of people freaking out over the raw TB without following though on the math. If the tools they are using can be trusted than the percentages produced must also be trustworthy.

Well said.

By now MILLIONS of M1 machines have been sold. Just a few of them showed abnormal behaviour. Every macOS or Safari release seem to be diminishing the issue more and more.

So, for those who ask, I would buy an M1 machine without hesitation. I did, actually. Can't be happier.

 

[Forti]

Well said.

By now MILLIONS of M1 machines have been sold. Just a few of them showed abnormal behaviour. Every macOS or Safari release seem to be diminishing the issue more and more.

So, for those who ask, I would buy an M1 machine without hesitation. I did, actually. Can't be happier.

oh c'mon, are you for real?

By now MILLIONS of M1 machines have been sold. Just a few of them showed abnormal behaviour.

Most people don't read tech news. and definitely they don't read MacRumors forum. And definitely x2 they don't even know how to check TBW and what is it.

Every macOS or Safari release seem to be diminishing the issue more and more.

Just one simple: no, it doesn't.


and why do people care about this issue? Because something somehow and for reason we simple don't know yet - is writing an equivalent of 3-4 4k videos into my SSD every single day.

and even leaving the computer in sleep mode for 12-16 hours end up with ~~40 GB written. 40gb - you know how long the text should be to take that amount of space?


Dont get me offensive or something - I just hate when someone is telling me "this is not an issue, this is normal". No, it's not normal

ps - by doing simple math my mac m1 should last for 2-5 years. My MacBook intel from 2018 has today 47.5TBW. My Mac mini almost 9.
Both machines were used the same way, same task, same power-on time every single day, and the intel one is running bigSur for couple of months now.

 

[Maximara]

ps - by doing simple math my mac m1 should last for 2-5 years. My MacBook intel from 2018 has today 47.5TBW. My Mac mini almost 9.
Both machines were used the same way, same task, same power-on time every single day, and the intel one is running bigSur for couple of months now.

Those raw TB tell us nothing. Give the percentages and time used.

 

[Forti]

Those raw TB tell us nothing. Give the percentages and time used.

Am I talking to the wall? It doesn't matter?

ofc it does matter... . I do wan't to know what and why is writing me 300-500 GB per day on my disk with just browsing websites basically. I'm not working for nothing more than just text.

 

[Maximara]

Somebody on Twitter said that their friend's M1 failed after somewhat 600 TB of written Data, which scared the **** out of me.

Which is a sample of size of one. As I said when this example first came up they could have been due to the controller going walkabout and I also asked what the workload was. 'I'm sorry I can't hear you over the crickets. What was that workload again?'

 

[Maximara]

Am I talking to the wall? It doesn't matter?

Yes it does matter if that is not moving your percentage then something is wonked with the numbers

ofc it does matter... . I do wan't to know what and why is writing me 300-500 GB per day on my disk with just browsing websites basically. I'm not working for nothing more than just text.

Is that very day or just for the day you looked at it? Also Chrome has been pointed out as a possible candidate, others have pointed to Time Machine, still others have pointed to old Intel software. Nobody really knows and Apple can't say jack because if they jump the gun they could be in a court room faster then you can say "defamation".

 

[Sysmet]

New Mac mini M1 16/256 came with 11.1 Big Sur preinstalled. Made clean install of 11.2.3 from USB after first boot.

First few hours — written ~100GB: mainly bird process and some kernel_task, cloudd, launchd and Mail.
Second day — written 10GB after office work with a few active Rosetta 2 apps and CrossOver x86 apps.
After 20h sleep, 5GB written.
Third day (current) — written 4GB during last 2 hours with light office work and surfing.

Everything works like a charm. Fantastic machine, don't hesitate.

PS: I don't use Time Machine, I use only Safari as a browser (fastest browsing experience I ever had), I haven't stopped indexing or any other processes. Also I'm not an insanely heavy user like some folks with 500 opened tabs.

 

[k-hawinkler]

New Mac mini M1 16/256 came with 11.1 Big Sur preinstalled. Made clean install of 11.2.3 from USB after first boot.

First few hours — written ~100GB: mainly bird process and some kernel_task, cloudd, launchd and Mail.
Second day — written 10GB after office work with a few active Rosetta 2 apps and CrossOver x86 apps.
After 20h sleep, 5GB written.
Third day (current) — written 4GB during last 2 hours with light office work and surfing.

Everything works like a charm. Fantastic machine, don't hesitate.

PS: I don't use Time Machine, I use only Safari as a browser (fastest browsing experience I ever had), I haven't stopped indexing or any other processes. Also I'm not an insanely heavy user like some folks with 500 opened tabs.

That sounds about right.

 

[Forti]

Yes it does matter if that is not moving your percentage then something is wonked with the numbers

Is that very day or just for the day you looked at it? Also Chrome has been pointed out as a possible candidate, others have pointed to Time Machine, still others have pointed to old Intel software. Nobody really knows and Apple can't say jack because if they jump the gun they could be in a court room faster then you can say "defamation".

Im not using chrome. Only safari.
Im not using time machine either. Every single day I post a lot of info about the on 1-15 Pages.

 

[Maximara]

Im not using chrome. Only safari.
Im not using time machine either. Every single day I post a lot of info about the on 1-15 Pages.

And yet other people are saying they are getting as low as 10GB a day. Again if this was a fault with Apple everyone using these tools or looking at activity monitor would see it - and not everyone is seeing it.

 

[Wando64]

Somebody on Twitter said that their friend's M1 failed after somewhat 600 TB of written Data, which scared the **** out of me.

Thankfully they are still under warranty, so not much of an issue really.

In any case, the reason there is a warranty is because technology sometimes fail prematurely.
I am sure if you look hard enough there are cases of SSD failing even quicker and with fewer TBW.

 

[Fishrrman]

"It's not a bug, it's a feature..."

How many have considered as to whether all this disk swapping/writing might be an inherent part of Apple's new m-series design?

That (even though they won't talk about it) it could have been intentionally done, perhaps as an integral requirement of the "direct memory access"...?

 

[JMacHack]

"It's not a bug, it's a feature..."

How many have considered as to whether all this disk swapping/writing might be an inherent part of Apple's new m-series design?

That (even though they won't talk about it) it could have been intentionally done, perhaps as an integral requirement of the "direct memory access"...?

I think that fast swapping is indeed a “feature” I’m not convinced that terabytes of swapping over such a short period is intentional though. Assuming it is a bug, I’d guess that Apple is still trying to figure out why it’s happening, since it seems to happen sporadically and there’s zero consensus here as to what’s causing it.

Maybe it’s architecture related, Apple’s been moving the iOS and MacOS codebase ever closer and we know that iOS is aggressive in clearing the RAM.

 

[rui no onna]

Do we know exactly what the smartctl percentages actually mean?

Does 1% mean that *all* the SSD cells have, on average, been written to for 1% of their guaranteed number cycles (e.g. 30 times for an SSD that has 3000 write cycles per cell)

Or does it mean that 1% of the cells have now exceeded their guaranteed number of writes?

The first of these would be less alarming because you could (as you have done) estimate the point at which the entire disk (assuming some kind of wear-leveling is used) would be about to reach its guaranteed write limit, above which we would expect to see cells marked as permanently failed. The disk would be "on borrowed time" but it doesn't mean that anything has failed yet, and we could still have the full capacity and have time to make backups.

The second would imply that some cells have already exceeded their limit, and have either failed, or could fail imminently. Presumably, if this were the case, we would eventually start to see the available disk capacity decrease, as failed cells were removed from the total available.

Average most likely. Wear leveling is one thing I expect modern SSD controllers have gotten better at implementing.

I saw a SMART value for spare used so one would assume if they've already killed some cells, then they'd replace those with spares and that SMART value would change.

 

[Forti]

And yet other people are saying they are getting as low as 10GB a day. Again if this was a fault with Apple everyone using these tools or looking at activity monitor would see it - and not everyone is seeing it.

Do I really have to write it every couple pages?

We don't know what is causing this issue. We dont know if this is because of m1 architecture, rosetta 2, or just a fault from developers (which I as a developer highly doubts).

I would go with too agressive RAM swap OR rosetta 2 issues.
Some people are using native apps, some don't.

Currently I'm using:
- safari (native)
- Xcode (native)
- iOS simulator (native)
- visual studio code (native)
- terminal (90-99% native, sometimes rosetta 2)
- node.js v15 (native support)
- Apple Music (native, right?)
- Spark email client (rosetta 2) - this the only one app running under rosetta currently. (written 41.5MB today)

those are my main apps. I'm not using chrome, Firefox etc.

So please tell me - which one of this could write 300-600GB daily?

I would say that node.js can do this - but from activity monitor I can't see it Below screenshot from activity monitor. I run node.js project, loaded it, playing with it a bit, doing some changes etc. I can't see how would it write 300-600 GB daily.

ps - Maybe the graphics Is using ssd as a memory? instead of ram (that would be stupid, I guess) hehe

 

[Maximara]

Do we know exactly what the smartctl percentages actually mean?

Does 1% mean that *all* the SSD cells have, on average, been written to for 1% of their guaranteed number cycles (e.g. 30 times for an SSD that has 3000 write cycles per cell)

Or does it mean that 1% of the cells have now exceeded their guaranteed number of writes?

The first of these would be less alarming because you could (as you have done) estimate the point at which the entire disk (assuming some kind of wear-leveling is used) would be about to reach its guaranteed write limit, above which we would expect to see cells marked as permanently failed. The disk would be "on borrowed time" but it doesn't mean that anything has failed yet, and we could still have the full capacity and have time to make backups.

The second would imply that some cells have already exceeded their limit, and have either failed, or could fail imminently. Presumably, if this were the case, we would eventually start to see the available disk capacity decrease, as failed cells were removed from the total available.

"Percentage Used: Contains a vendor specific estimate of the percentage of NVM subsystem life used based on the actual usage and the manufacturer’s prediction of NVM life. A value of 100 indicates that the estimated endurance of the NVM in the NVM subsystem has been consumed, but may not indicate an NVM subsystem failure. The value is allowed to exceed 100. Percentages greater than 254 shall be represented as 255. This value shall be updated once per power-on hour (when the controller is not in a sleep state)." - SMART Attribute Details

This makes it clear what "Percentage Used" means.

 

[wirtandi]

Anyone got any news about the possible fix in 11.3? From what ive seen, those who are on 11.3 seem to report the issue has been fixed

 

[daverdfw]

Anyone got any news about the possible fix in 11.3? From what ive seen, those who are on 11.3 seem to report the issue has been fixed

I am on 11.3b4 and after 4 days 4 hr uptime Kernel_task has written 1.55TB . So not fixed for me.

 

[wirtandi]

I am on 11.3b4 and after 4 days 4 hr uptime Kernel_task has written 1.55TB . So not fixed for me.

Thanks for that. Such a shame though.....I thought 11.3 has finally fixed it

 

[OnawaAfrica]

This makes me wonder how a potential SSD Repair Program would work, considering it’s all an SoC. When my 2015 MBPr’s battery was replaced, the entire bottom case including the keyboard needed to be replaced. It almost felt like a new laptop (RIP again). Would they have to replace the entire M1 SoC? Would they still be able to replace it with the M1 or would it be some newer variant? 🤔

The SSD is not integrated into the SoC. the SoC only has the CPU itself + the Ram. that's what makes it a SoC

 

[OnawaAfrica]

I have a Samsung SSD 860 EVO 500GB with total Power-on time of 1 month & 25 days. it indicates a write if 3.1TB and read of 2.9 TB and the SSD is still very healthy

 

[Sysmet]

Those who have problems with excessive data write should try to make 11.2.3 clean install from USB drive and use Mac for one day without any third party software and turned off Time Machine. Then 1-by-1, add something and log results.

You will figure out what is causing it.

This is software that I'm currently using, plus some Windows programs with CrossOver. No problems so far.

 

[pistonpilot]

We could do without the salty language. Let’s keep this forum on point and respectful.

You're correct, it needs pepper. Screw Apple Doubly for screwing us.