M1 SSD high read and write usage per smartctl

Robert Follis wrote:

Opening Safari with 1 tab spikes Kernel Tasks from 0% to over 5000% and 1 tab uses 1.03GB of memory

Only other app showing any major memory use in BitDefender

Just uninstalling BD to see what happens

Erk - ok.

I don't have an M1 Mac so I don't know if this is relevant, but kernel extensions have caused issues with MacOS in the past. Open a Terminal window and type kextstat. Look for anything that's not "com.apple ..."

But this is just throwing stuff out there. Another thing that may be just something to occupy a few minutes is to reset the SMC - https://support.apple.com/en-us/HT201295

YI other people with 16GB RAM are reporting exactly the same problem - See @sfromgi in this thread "My MBpro M1 16/1tb is showing 6.6TB written in less than 1 month of lite use" That's even worse than me

6.6TB in a month isn't so bad. My own averages out at 5TB/month.

For one, reads don't matter with SSDs, only writes.

I have been investigating this "issue" as well. I checked out 3 different

SMART tools all of which gave different usage data. Also, when comparing

to what Activity Monitor is saying for the number of bytes, it doesn't jive

with these tools reports. As a matter of fact, one day, Activity Monitor

said there were 40 GB of writes and smartctl actually reported that

TBW went down by 10 GB!

Another example today, I did a photo processing session and Activity Monitor

showed nearly 70 GB of writes, but only 500 MB swap was used so it is not

a RAM swapping issue and based on Activity Monitor, the apps that were used never

did more than a couple GB of writes. After checking the Smart tools I had,

one said about 30 GB was written another said only about 10 GB was written

and a third said 40 GB was written.

Also, some of the SMART tools themselves do not have numbers

that make sense to give the totals they are getting, i.e. sectors

written times sector size do not add up to total bytes written.

Personally, I believe the following:

1) The SSDs in the M1 machines are not totally documented in terms of

SMART status available and formats of the data and how they should be

read and interpreted..

2) While Activity Monitor has been around a long time, I believe there

are reporting issues with it as well.

3) Until there are reliable tools to measure the usage, no one really

knows how many writes (or reads) there really are being done.

Just looking at your data, units/sectors written being reported is 24 MB.

The M1 SSD sector size is 4096 (if that is even being reported correctly.

The math says the total number of writes should be ~120 GB, which is

awfully close to a difference factor of 1000. So, is the units read correct?

Is the total calculation incorrect? Is the sector size/unit size even correct?

Is Big Sur going berserk with writes? Is anything correct?

I've enabled "Prevent computer from sleeping automatically when the display is off" so the M1 Mac mini should never go into sleep. What's really odd is, Swap Used reported by Activity Monitor was 0 bytes prior to display automatically turned off after set period. After waking up the display, Swap Used has increased to 526MB.

As a result, KernelTask has written large amount of data during the time while the display is off. Every time the display is automatically turned off, the Swap Used is increased! By the end of the day, Swap Used is a few GB, and KernelTask has written close to 200GB of data (per day) and that seems excessive.

I am on macOS 11.3 and the M1 Mac mini is 8/256GB model. Something is odd during the sleep time, perhaps some background process is leaking memory causing excessive swap usage.

--

I've tested further to discover this:

1. Turn screen saver on after 1 minute
2. Turn off sleep in power mode
3. Swap Used is 0 bytes after a clean reboot
4. Let the screen saver starts after 1 minute
5. Wait for another minute

Now Swap Used is 636MB.

My smartctl report on my MBP M1 states 58.TB written (328,696,111 write commands). I have only had it since the 9th of January (Power on hours @ 239) so I'm particularly concerned about this potential issue.

I found an interesting piece on the DriveDx site. It seems the

"Raw" value that all these apps use to determine total bytes written

can be cloaked in mystery especially with a brand new drive. It

stated:

"Each health indicator (S.M.A.R.T. attribute) has a raw value aka raw data. Raw measured values (provided by a sensor or a counter) are stored in this field. Sometimes different parts (high word, low word, etc) of raw value contain different kind of information.

The drive manufacturer defines the meaning of this value (but often corresponds to counts or a physical unit). The exact meaning of raw value often considered as trade secret. These values may significantly vary between different manufacturers and drive models and should NOT be compared with other devices or other vendors."

So, it would appear that unless Apple has released the precise details of this

value, perhaps all the readings are bogus. Could it mean the values are

significantly lower or even higher? It seems no one really knows for sure at

this point. Actually, reading through that entire page, without precise manufacturer

definitions nearly every value that may be read in SMART status could be ?????.

The explanation:

https://binaryfruit.com/drivedx/help#ui-health-indicators-raw-value

There is one last thing that is at this point a total unknown.

That is who makes the SSD chips for Apple and what are

their lifetime TBW.

From some quick research, for say a 512 GB NVMe drive, I found

TBW lifetimes range from 300 TBW to as high as 1200 TBW.

There are also a wide range of things that contribute to that

such as over-provisioning (how much "extra" SSD space is

in the Apple SSDs), wear leveling (the SSD deals with this and

APFS also helps it), how robust is the error detection and correction,

etc.

So, without the actual TBW lifetime, one is not sure to just

be "concerned" or to be "panicked".

I fully agree with your explanation. The way the SMART raw values should be decoded can be different by device/manufacturer. For instance, if you look at Crucial web site: https://www.crucial.com/support/articles-faq-ssd/ssds-and-smart-data they are explaining how the SMART raw 202 attribute should be decoded and, in this case, it corresponds with the definition. Coming back to Macbook NVME drives, if I use smartctl on my Macbook of 2017 (that I'm using 8 hours on each working day), it reports only 1% used and 17 TB written and this does not obviously correspond to the real use I did.

Open up activity monitor and go to the disk tab. Look for the kernel_task process and that will give you an idea of what is being written to the SSD. Then click apple button in menu bar, "about this mac" and then click "system report" click on "Software" and in the main panel check the "Time Since Boot". This will give you an idea of how intensively the disk is being written to.

In my case it is nearly six days since boot and the kernel_task process has written just over 5TB (Two of those days were the weekend when I probably used my macbook pro a little less).

Now, I have had my MacBook 75 days, so if I now check the output of sudo smartctl --all /dev/disk0 :

Percentage Used: 2%
Data Units Written: 160,257,271 [82.0 TB]

So 82TB over 75 days is ~1.1 TB per day and seems to tally with what the activity monitor is reporting.

So I think the issue is real.

One thing to note, many NVMe SSDs can have significantly large

amounts of built in DRAM. With proper management, many of the temp

writes (caches, swap, etc.) may never even reach the actual NAND cells

of the SSD.

The DRAM in the SSD is used to speed the writes up as a direct write

to the NAND cells is comparatively slow.

As one last additional note, it seems that heavy use of Intel apps

can be a culprit of excessive writes. With some investigating, I

found that one photo processing app I used grew to a point on

my M1 MacBook Air so much that it was using 10 GB of swap

and as such, a lot of writes were constantly taking place and was

basically continually swapping. I have the same app on an Intel iMac

and I duplicated the same session exactly and the app never used

more than 2GB. It seems there may be an issue with Rosetta 2

not properly managing memory allocation (a leak?) to a point

where it does not return memory requested and then released

by an application.

Also, with the M1 being so fast and efficient and the blazing speed

of the SSD, heavy memory swapping is not even noticed like

it would in the old days with massive slow down and spinning

beachballs.

Unfortunately issues with SMART reporting have been around since the beginning since drive manufacturer's don't always provide the necessary information to properly interpret the real values. It is crazy that such a useful feature is not properly documented by the drive manufacturer. Does DriveDx or smartcl provide the old style table layout of the attributes to show the "Value", Worst", and "Threshold" reporting? If so, then that is the best way to monitor the health of the drive as the drive will automatically decrement the "Value" and "Worst" as the attribute ages. Computing the "RAW Value" depends upon knowing how that value is stored.

I've been monitoring our organization's Intel Macs more closely after we received a worn out SSD after having a 2017 MBPro Logic Board replaced by Apple (the SSD failed within 90 days of the repair and had used up all its blocks -- there is no way the user wrote that much data to the SSD in 90 days). Ever since I've checked each SSD that Apple has replaced and I've even compared them to other SSDs from laptops that have not been repaired. I've noticed that the SMART values being reported are eerily similar to each other and even to the values reported by the OP here. The interesting thing is the USB-C Macs all have SSDs identified as Apple (AP) whereas the older non-USB Macs had the SSDs identified by the real manufacturer such as Samsung (SM), Toshiba, or SanDisk, etc.

If you have enough of a hardware fleet around, might want to deploy some tools to dump the smart data across the fleet. Much like figuring out which part of ACPI a vendor is using, figuring out where to look for the over-provisioning and error data that matters is ha;f the “fun”.

I have written 3TB in 7 days of use. I have the 16GB Mem version. If the TBW is 300, the Mac will be dead in 2 years, as the SSD is not replaceable. Less if I use it as I want to use it (as a SW dev, I want to be constantly rebuilding my apps, starting / stopping services, running local databases etc.)

i have started using My PC more, to save the Mac self destructing. i try to keep fewer apps open at the same time to reduce paging, but this makes the Mac more painful to use. I try to shut down the mac when not in use, to avoid it writing during sleep. I have uninstalled whatsapp, and now only use this on the phone.

The posts saying that the TB written reading might not be accurate sounds like wishful thinking, but lets hope. Using the PC has reminded me how bad the MacOS finder is compared with file explorer, but how good Outlook mac is compared with windows version (no unified inbox etc).

M1 SSD wear is looking like a serious problem. My very clean M1 MBP 8/256 using fairly standard business apps has used 21% of SSD life in 5 months writing 258TB !! to disk. This means the SSD will die at just under 2 years which is unacceptable. 

Also lots of runaway kernel tasks using over 5000% CPU! Safari appears to cause a lot of this

How do I get this addressed?

Apple is going to have huge warranty problems around the two-year mark

Do you have a antivirus, cleaner, optimizing, etc. apps

installed?

Also, are all apps up to date as well as any hardware drivers?

What sort of apps are in your startup items (both user and system)?

What web browser are you using and what sort of plugins

are being used?

Any or all of the above can cause runaway CPU use.

Also, no one really knows what the TBW lifetime is for the

Apple SSDs. Also, like anything else, once you hit TBW limit,

it may start to show issues or could last for lots and lots of TBs

longer.

The other fact about TBW ratings for any SSD, is they are just

a hedge for manufacturer warranty coverage. Much like the milage

and warranty for a car. If the warranty is for 50,000 miles it just

doesn't shut down at 50,001 and cease to work.