M1 Mac Users Report Excessive SSD Wear

[toke lahti]

According to a quote from an (admittedly anonymous) Apple engineer that I reproduced in my last comment in this thread, the numbers are not accurate. If that is correct--and I suspect that even an anonymous Apple employee would not mislead in a case like this, where the truth will eventually come out--then it will be just a matter of time before additional evidence bears this out. Right now, in spite of the endless but understandable speculation by M1 users and potential M1 buyers, this thread is almost entirely populated by hyperbole.

Correct me, if I'm wrong, but Apple does not manufacture the ssd or its controller?
And the data is coming from the controller, not from the OS?
Or is it with M1 that the ssd controller is in M1 chip and Apple can have messed up the standard NVMe SMART data?

 

[flopticalcube]

As I understand it, the controller is now part of the M1 SoC.

 

[jdsingle]

As per my above post, at the wear rate reported, more like 10-20 years.

As per the lack of me replying to your quote, I didn't read through 15 pages to look for your insight into this.

 

[Maximara]

As per the lack of me replying to your quote, I didn't read through 15 pages to look for your insight into this.

Post #353. Method: looked at name throAU and used the find command of the browser on each page before that point. Time to find: 1 minute (if that long).

 

[bushman4]

Rule of thumb for the future: Never buy the first model of something new as there are many kinks and improvements to be had

 

[jdsingle]

Post #353. Method: looked at name throAU and used the find command of the browser on each page before that point. Time to find: 1 minute (if that long).

I’m seriously struggling to follow your logic here. I read the article. I made a post. For me to have read post 353 per your comment implies I had foresight that this post was made or that I am able to read through 353 posts in one minute. Did you even read?

 

[Maximara]

I’m seriously struggling to follow your logic here. I read the article. I made a post. For me to have read post 353 per your comment implies I had foresight that this post was made or that I am able to read through 353 posts in one minute. Did you even read?

I think you don't understand the point. I did not "read through 353 posts in one minute"; never claimed I did.

I spelled out exactly how I went about finding throAU's post: "used the find command of the browser on each page before that point". It is akin to using Google to find a passage in a book or a lyric on genius - just a lot saner because the material is much smaller and you are using the browser' Find command (under Edit in Chrome) rather than being at the mercy of the search algorithm and metadata.

 

[Maximara]

Rule of thumb for the future: Never buy the first model of something new as there are many kinks and improvements to be had

The problem with that is given the consumer and quarterly profits mindset if everybody did this odds are there wouldn't be a second product of that line.

The thing here is there are just too many questions to go Chicken Little here - what is causing it, are the numbers even real (as the TB*100/percentage = TBW crosscheck produces some really gonzo numbers with what we have been given), and how long the SSD will live.

 

[SeanGold]

So just to update, I reached out to Apple again tonight and didn't get much of an answer once again, but I did get a rep that came onto my computer to look at what I'm talking about. He collected my system data, and we took screenshots and he says they'll send it to the technicians and they should get back to me on Saturday. Hopefully he conveyed the context in his notes well enough, and maybe I'll finally get an answer as to what's going on with this issue. He was on while my machine wrote about 8GB's in about 6 seconds, and then proceeded to write another 50GBs or so in a few minutes while the machine was doing nothing and we were talking on the phone.

 

[Phantom iCloud tabs]

So just to update, I reached out to Apple again tonight and didn't get much of an answer once again, but I did get a rep that came onto my computer to look at what I'm talking about. He collected my system data, and we took screenshots and he says they'll send it to the technicians and they should get back to me on Saturday. Hopefully he conveyed the context in his notes well enough, and maybe I'll finally get an answer as to what's going on with this issue. He was on while my machine wrote about 8GB's in about 6 seconds, and then proceeded to write another 50GBs or so in a few minutes while the machine was doing nothing and we were talking on the phone.

They'll get to the bottom of this, but my instinct here (and that's all it is) is that there's no way these numbers are accurate, and that there's a big bug in the drive/data write reporting.

 

[Maximara]

They'll get to the bottom of this, but my instinct here (and that's all it is) is that there's no way these numbers are accurate, and that there's a big bug in the drive/data write reporting.

I fully agree. The most basic crosscheck (TB*100/percentage = TBW) produces some really gonzo numbers

For example, the 170TB written with smartctl reporting 2% used reported by Fomalhaut produces a totally off the wall impossible 6827 (170*100/2.49) TBW for his drive. If they go with basic math where 2% x 50 = 100% then we get 8500 TBW (170TB x 50) which is insane. There is just no way those numbers are real. It is basic math so either the measurements are messed up or the percentage is garbage.

 

[toke lahti]

So just to update, I reached out to Apple again tonight and didn't get much of an answer once again, but I did get a rep that came onto my computer to look at what I'm talking about. He collected my system data, and we took screenshots and he says they'll send it to the technicians and they should get back to me on Saturday. Hopefully he conveyed the context in his notes well enough, and maybe I'll finally get an answer as to what's going on with this issue. He was on while my machine wrote about 8GB's in about 6 seconds, and then proceeded to write another 50GBs or so in a few minutes while the machine was doing nothing and we were talking on the phone.

Btw, what is your display resolution? Do you use many of them?
Could this be such a bug that OS swaps GPU ram to ssd?

 

[TheSynchronizer]

I fully agree. The most basic crosscheck (TB*100/percentage = TBW) produces some really gonzo numbers

For example, the 170TB written with smartctl reporting 2% used reported by Fomalhaut produces a totally off the wall impossible 6827 (170*100/2.49) TBW for his drive. If they go with basic math where 2% x 50 = 100% then we get 8500 TBW (170TB x 50) which is insane. There is just no way those numbers are real. It is basic math so either the measurements are messed up or the percentage is garbage.

According to the SSD endurance test by techreport, a 256GB Samsung 840 SSD exceeded 2.4PB, or 2400TBW+, before it failed. And this was in 2015 - could it be possible that since then advancements in SSD technology have allowed for a much higher TBW ceiling?

Maybe these numbers could be real, as unlikely as that would be, but technology moves fast and I honestly wouldn't be surprised if they are real.

 

[Maximara]

According to the SSD endurance test by techreport, a 256GB Samsung 840 SSD exceeded 2.4PB, or 2400TBW+, before it failed. And this was in 2015 - could it be possible that since then advancements in SSD technology have allowed for a much higher TBW ceiling?

Maybe these numbers could be real, as unlikely as that would be, but technology moves fast and I honestly wouldn't be surprised if they are real.

I was thinking about a stated TBW which is what companies generally warranty their drive for. For example a
2017 Toshiba-OCZ RD400 topped out at a warranted TBW of 592. The Samsung 840 PRO Series is given a 73 TBW (it's across 28GB, 256GB, 512GB so that is clearly an average)

This demonstrates the problem and reason Apple, in all odds will, never give the TBWs for its SSDs. The number is for warranty reasons and doesn't really reflect what the drive's true TBW is.

 

[zimv20]

you're holding it wrong.

 

[AlifTheUnseen]

Don't know whether this has been discussed before or not: In the AnandTech article on the Apple M1, they measured AC wall power consumption while doing several things with the M1 mini, e.g. Single, Multi Core and GPU stuff.

https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested

They reported low two digit Watts for the various tasks. Now I checked (also on AnandTech) that a number of different SSDs use low single digit Watts while writing stuff.

I am no electronics expert... would it be possible to "measure" power consumptions on M1 minis that are in a SSD writing spree? Would the signal-to-noise ratio allow to observe the higher power consumption while GBs of data being written in a very short time to the SSD?

Maybe this might help answering the question whether this is a reporting bug or a more serious ssd killing spree bug, but maybe it's a too faint signal...


BTW: Here is my data on a 2015 iMac on Mojave with 8 GB and a 500 GB Samsung SSD - I have HUNDRED+ Safari tabs open at any time ;-) :
173 Wear_Leveling_Count 0x0032 184 184 100 Old_age Always - 1279933350262 174 Host_Reads_MiB 0x0022 099 099 000 Old_age Always - 109795983 175 Host_Writes_MiB 0x0022 099 099 000 Old_age Always - 95043085 --> 95 TB written, 84% wear level capacity.

 

[Jouls]

173 Wear_Leveling_Count 0x0032 184 184 100 Old_age Always - 1279933350262 174 Host_Reads_MiB 0x0022 099 099 000 Old_age Always - 109795983 175 Host_Writes_MiB 0x0022 099 099 000 Old_age Always - 95043085 --> 95 TB written, 84% wear level capacity.

I have those old age readings as well. Where did you get the wear level capacity from? And does it translate to 16% worn out? I forgot what the data point from the M1s is called.

 

[AlifTheUnseen]

I was confused to no end. One reads everywhere that the 173 wear leveling count starts at 100% and drops. Yet, smartctl showed a value of 184 as quoted above (threshold: 100, worst:184). Indeed, my guess is that this is a different convention: the wear leveling count starts at 200% and the "spare capacity" is fully used when the wear leveling count reaches 100%... so I downloaded DriveDX and it reported the same numbers as smartctl and additionally showed a status of "84% ok" (that I called wear level capacity, as DriveDX did not name it specifically). So, yes, I think my SSD is 16% worn out, as you wrote.

 

[Maximara]

I was confused to no end. One reads everywhere that the 173 wear leveling count starts at 100% and drops. Yet, smartctl showed a value of 184 as quoted above (threshold: 100, worst:184). Indeed, my guess is that this is a different convention: the wear leveling count starts at 200% and the "spare capacity" is fully used when the wear leveling count reaches 100%... so I downloaded DriveDX and it reported the same numbers as smartctl and additionally showed a status of "84% ok" (that I called wear level capacity, as DriveDX did not name it specifically). So, yes, I think my SSD is 16% worn out, as you wrote.

Well those numbers do produce a more realistic 593.75 (95*100/16) TBW.

 

[Maximara]

To everyone moaning and groaning about Apple using soldered RAM and SSDs read The scourge of fully soldered and non-upgradeable laptops.

"First and possibly foremost is manufacturing efficiency, which includes both quality control and cost reduction. Every additional removable piece, especially including a SODIMM slot, introduces more cost and another potential fail point. Plus, an actual RAM socket requires an actual human being be there to plug a RAM chip into every laptop that goes down the assembly line, further adding to cost.

More pertinent, though, is the fact soldered RAM can be placed just about wherever engineers decide. With proper research and development, this can lead to streamlined mainboard design as well as increased thermal efficiency. It also means there's no need to include a bulky connector or an access door, and all told, these benefits combine to let laptop designers shave millimeters off case thickness"

 

[Jouls]

I was confused to no end. One reads everywhere that the 173 wear leveling count starts at 100% and drops. Yet, smartctl showed a value of 184 as quoted above (threshold: 100, worst:184). Indeed, my guess is that this is a different convention: the wear leveling count starts at 200% and the "spare capacity" is fully used when the wear leveling count reaches 100%... so I downloaded DriveDX and it reported the same numbers as smartctl and additionally showed a status of "84% ok" (that I called wear level capacity, as DriveDX did not name it specifically). So, yes, I think my SSD is 16% worn out, as you wrote.

Thank you. So my 16" MBP from 2015 looks very good with 21 TBW and 94% health.

 

[larkhon]

To everyone moaning and groaning about Apple using soldered RAM and SSDs read The scourge of fully soldered and non-upgradeable laptops.

"First and possibly foremost is manufacturing efficiency, which includes both quality control and cost reduction. Every additional removable piece, especially including a SODIMM slot, introduces more cost and another potential fail point. Plus, an actual RAM socket requires an actual human being be there to plug a RAM chip into every laptop that goes down the assembly line, further adding to cost.

More pertinent, though, is the fact soldered RAM can be placed just about wherever engineers decide. With proper research and development, this can lead to streamlined mainboard design as well as increased thermal efficiency. It also means there's no need to include a bulky connector or an access door, and all told, these benefits combine to let laptop designers shave millimeters off case thickness"

That’s why it is soldered on the Mac Mini... the article also concludes that a M2 SSD instead of a soldered one isn’t going to add much bulk.

 

[AndyMacAndMic]

To everyone moaning and groaning about Apple using soldered RAM and SSDs read The scourge of fully soldered and non-upgradeable laptops.

"First and possibly foremost is manufacturing efficiency, which includes both quality control and cost reduction. Every additional removable piece, especially including a SODIMM slot, introduces more cost and another potential fail point. Plus, an actual RAM socket requires an actual human being be there to plug a RAM chip into every laptop that goes down the assembly line, further adding to cost.

More pertinent, though, is the fact soldered RAM can be placed just about wherever engineers decide. With proper research and development, this can lead to streamlined mainboard design as well as increased thermal efficiency. It also means there's no need to include a bulky connector or an access door, and all told, these benefits combine to let laptop designers shave millimeters off case thickness"

Ok, you found a couple of advantages, but this does not take away the disadvantages of using soldered RAM and SSDs. Also there are loads of very old Macs and other PC's with replacable RAM and HDD's/SSD's still functioning, so even the pro's you found are maybe not that important.

The 'moaning and groaning' as you call it still has very valid points in my opinion.

 

[SactoGuy18]

I think the YouTube video from Max Tech may be correct. The issue is that because the test checking SSD usage is running through Rosetta 2, it may be giving extremely erroneous results. They may need to rewrite the test so it is a true "Apple Silicon" native app and see what happens.

 

[AlifTheUnseen]

I think the YouTube video from Max Tech may be correct. The issue is that because the test checking SSD usage is running through Rosetta 2, it may be giving extremely erroneous results. They may need to rewrite the test so it is a true "Apple Silicon" native app and see what happens.

Maybe, but then, if it is a 'reporting bug' coming from non-native execution, why are some seeing the bug and others do not - on the same hardware and os? It's Rosetta2-executed code on ALL machines...