On Saturday, 8 March 2025 22:48:29 Greenwich Mean Time Dale wrote:
> Frank Steinmetzger wrote:
> > Am Thu, Feb 20, 2025 at 11:40:48PM +0100 schrieb Frank Steinmetzger:
> >> Another thing to consider: don’t put it into the safe for a year without
> >> powering it up. As was explained in a previous mail, QLC uses sixteen
> >> different levels of charge inside one single flash cell. The chance of a
> >> bit flip increases the longer the SSD is powerless and charge slowly
> >> (very slowly) dissipates. It’s hard to find exact numbers, and it’s more
> >> of a statistical question. Could be a research topic for a slow Sunday.
> >> ;-) Also, don’t you live in a hot area?
> > 
> > I knew I’ve seen this data in the past, but couldn’t remeber where and in
> > what context. I just stumbled upon the relevant info again.
> > 
> > If you search for "jedec temperature and data retention" you find this
> > PDF:
> > https://www.jedec.org/sites/default/files/Alvin_Cox%20%5BCompatibility%20M
> > ode%5D_0.pdf
> > 
> > It contains establised standard values for flash durability. On slide 28
> > it
> > says:
> > Client-class SSDs should retain their data w/o power for 1 year if used
> > 8 hours/day at 40 °C and kept at or below 30 °C when off.
> > There is also a table for other temperatures; the time is cut in half for
> > 5 degrees more.
> > 
> > Mind you, that PDF is 15 years old, TLS had just been released to the
> > public one year earlier according to
> > https://en.wikipedia.org/wiki/Multi-level_cell#Triple-level_cell
> > 
> > -- Grüße | Greetings | Salut | Qapla’ Please do not share anything
> > from, with or about me on any social network. Some are so convinced,
> > they don’t even know anymore of what.
> 
> That is confusing.  Data should be OK for one year without power but
> only if powered 8 hours per day.  Me scratches my head.  Maybe it means
> should be OK for a year without power and not sure what to think about
> rest. 

As I understand it these are two different performance requirements, which 
were stipulated in this proposed standard back in 2010.

The first, "Active Use (power on)", describes the level of performance on a 
typical client workload and operating environment, 8hrs/day at 40°C.  I don't 
know if this is controller temperature, NAND flash cell temperature, device 
sensor temperature, PC case temperature, or room temperature.  The actual 
standard document would explain this, as well as the method for measuring it.  
Meanwhile, I note my NVMe SSD's sensor tells me the drive is idling at 45.8°C 
as I write this.  o_O

The second, "Retention Use (power off)", sets a limit of 1 year @30°C.

The FFR and UBER thresholds apply to *both* of the above.


> I suspect that things have changed a lot in that time frame despite the
> confusion.  They may have improved things significantly since then. 

Hmm ... I wouldn't take this to the bank.  Standards are typically formulated 
following consultation with industry to establish the art of the possible, but 
without cutting too much slack on manufacturers so as to promote innovation 
and improvement.  The manufacturers seek to minimise overall costs, while 
keeping their head above the thresholds stated in the standard.  So there's a 
tension right there, which hopefully delivers innovation and ideally lowers 
costs over time - but it does not necessarily improve performance.  Check how 
quality of USB flash drives has gone south over the years, following the 
trajectory of their price.

The move from SLC to QLC has been effective in reducing costs, but at some 
point the limits of physics will constrain performance.  It would be 
interesting to dig out the latest JEDEC standard and compare thresholds to see 
how they may have improved with more modern manufacturing methods.


> Still, if the one year part, less the other part, is accurate then but
> better now, then my powering up even once a month should be OK.  I only
> need to worry about if something happens to me and no one touches it for
> many months, a year or longer. 

Given the temperature of the device would probably be below 30°C for at least 
half of the year in day time and most nights, I would think your data will be 
fine.  :-)

I have a 11 year old Toshiba OCZ SSD, which gave me a scare a week ago.  It 
reported an I/O error, dmesg showed it being inaccessible - I can't remember 
the exact error.  I thought it was game up with this drive, but surprisingly 
following my reseating the SATA cable and a reboot later it has worked without 
further complaints.  With reports of OCZ drives failing at an alarming rate 
more than a decade ago, I am happy it has lasted this long.