What we know so far is that some people are reporting completely unreasonable, dangerous levels of SSD write volume, which on some configurations would be expected to prematurely wear down SSDs (which are not replaceable) well before the expected lifetime of the device. We don't know yet what triggers this; it looks like some kind of bug or software combination, but it's too early to have good leads.
The worst example so far is David's, which, if scaled proportionally (by Flash cell wear) from his 2TB SSD to a base 256GB model, would be reaching 100% lifetime usage within less than a year. This calculation is based on an assumption of equal (proportional) overprovisioning for those two models, and an assumption that whatever triggers this behavior doesn't care about total SSD size; these are sensible assumptions to make at this stage but not verified.
https://twitter.com/marcan42/status/1361160838854316032
This is clearly a software problem, and fixable with an update. It's not a hardware defect, it's the kernel (we think it's a VM* issue) hammering the SSD way too much.
There is reason to be concerned, and to make sure Apple fixes this if it is a real bug. There is no reason to be alarmed and panic about these machines. If this is a real issue they have to fix it; they aren't stupid, Apple knows full well that the SSDs in some M1 macs dropping like flies within a year would be a PR disaster for them. What we need to do know is gather data and try to find a way to reproduce this.
* VM means Virtual Memory, not Virtual Machine, in this context, for all you non-kernel folks.
Most users will be able to get a decade worth of usage out of it.
Nothing to see here, move along.
That is two big assumptions: assuming linear degradation and assuming the mac is usable until it reaches 0%.
I would think that it starts degrades faster as it wears, and that by the time it is at least 90% on a 256gb or even 512gb SSD it will be effectivly unusable.
I thought i already accounted for that by stating "most users will get a decade of use". That's roughly 66% of linear degradation to 0%.
> and that by the time it is at least 90% on a 256gb or even 512gb SSD it will be effectivly unusable.
As i understand it, the "lifetime" is a reflection of the spare sectors, meaning once the spare sectors run out you'll start seeing errors instead of relocated sectors. It will probably continue for some time after that.
Knowing Apple, if it becomes a problem anytime within the first 5-6 years, they'll replace it for free. I had a 2008 iMac with a manufacturing problem on the Seagate 1TB drive it shipped with, and 5 years after purchase (2013), i was able to get the drive replaced for free despite my machine showing no signs of the error. 5 years after purchase, and a full 3 years after most consumer laws stop protecting you.
Other than that, the only reason this is a "problem" is because SSDs has an indicator that tells you when they'll expire. Spinning rust doesn't have that, but some spinning rust will also expire after 5-6 years, and most will have expired by the time a decade has passed (assuming daily usage).
This is the figure that storage manufacturers use in their warranties to decide if a device has been excessively used - NOT the amount of writes.
Even if you just look at the 2TB model itself, 3% in 2 months means the thing will reach 100% in 5 years which is... not great.
So there is reason to be concerned here.
The product configuration as shipped inflicts permanent damage on itself, it's very difficult to see this as anything but a manufacturing defect.
What "2-4 years wear"?
It's 1% of usage in 2 months. That's par for the course for any SSD in any laptop with regular usage, and it should last long before the laptop is updated in 6 or so years...
Heck, 50% "percentage used" would take 8 years with this rate...
Windows 10: Data Units Read: 11 332 875 [5,80 TB] Data Units Written: 8 173 369 [4,18 TB]
Linux: Data Units Read: 5 837 296 [2,98 TB] Data Units Written: 4 891 744 [2,50 TB]
There's a separate SMART attribute for remaining spare, and that is supposed to stay at 100% until you reach 100% lifetime, but the actual failure of cells can come sooner or later.
This is the worst example so far, 3% on a 2TB SSD.
https://twitter.com/david_rysk/status/1361155414994407424
Care to share the actual write volume that corresponds to 1% used for you on your 256GB drive, so we can validate whether the linear scaling by drive size assumption is reasonable?
SMART/Health Information (NVMe Log 0x02)
Critical Warning: 0x00
Temperature: 32 Celsius
Available Spare: 100%
Available Spare Threshold: 99%
Percentage Used: 1%
Data Units Read: 42,513,003 [21.7 TB]
Data Units Written: 35,670,814 [18.2 TB]
Host Read Commands: 444,309,142
Host Write Commands: 237,126,099
Controller Busy Time: 0
Power Cycles: 213
Power On Hours: 199
Unsafe Shutdowns: 13
Media and Data Integrity Errors: 0
Error Information Log Entries: 0So 18TB / 256GB = 70 drive writes is 1%.
For the 2T drive, we had 150TB = ~50 drive writes being 3%.
So, it seems this isn't linear after all. But it's also not constant; 150TB would put you at 8% used on your 256GB model with this scaling (instead of the ~23% if it were linear), which is still not insignificant, and would still get you to 100% within a couple years. Though there is significant rounding error with the "1%" figure.
I saw someone else mention 3% on Twitter too and asked for their drive size, so I hope that can give me another more accurate data point.
SSD capacity * ~600 / Bytes written
I get that its a much higher wear than we would traditionally expect, but given I have 8GB of RAM that seems near inexhaustible., maybe it just a different approach to memory management, permitted by the blazingly fast storage we now have access to? I simply couldn't run all this stuff on my 16GB 2015 machine.
I cant run figma though - that thing eats 4GB for a smallish project, so it almost always complains about my available memory when im using figma.
2TB model: 5000TBW
256GB model: 2000TBW
That means that, at the worst known rate that David had (150TB in 2 months), a 256GB drive would reach lifetime writes in ~2 years. That's still way too fast, so this is still something Apple needs to adjust if it is happening to people (and given what he told me about the write rates he was seeing while using and switching apps, it really does seem like a bug, not normal app swapping).
FWIW, the "app nap" thing is what iOS does, but macOS can't do that because Mac apps aren't really designed that way. macOS just does traditional swap plus compression, as far as I know.
I'm aware it will be a form of virtual memory management - but given im currently (and actively) running safari, chrome, firefox, affinity designer, affinity photo, slack, mail, a handful of terminals, a couple of ios apps, an iphone 11 simulator, 2 android emulators, 4 instances of vscode recompiling code as I type, and a bunch of utility apps theres definitely more than just a traditional swap going on here. Its an 8GB machine, and it really feels like it has unlimited RAM (until i open figma, as mentioned before - theres something about that app that the m1 really doesnt like).
Of course, if it is a bug, ill be happy to have it fixed :)
Also, I just realised you are the OP :) And the guy working on asahi (awesome!). Would love to know more if you get to the bottom of this behaviour - intentional or otherwise!