ฝัง
- เผยแพร่เมื่อ 29 ก.ย. 2024
- #SSD #Archive #Testing
Long term SSD data storage test. How long can an SSD store data while unpowered? I'm doing my own basic test to store and check data after 1 year and 2 years left unpowered.
Link to WD White Paper on SSD and Hard Drive data retention: documents.west...
Due to suggestions, I did end up doing a read test of the data. I unfortunately did not do a read test when disks were new, but I do have the 1 year test results here:
102458 Total MB
683 Total Files
150 MB Avg File Size
1 Year Check:
SSD 1 (Worn): 207 seconds total ~ 495 MB/sec avg read speed
SSD 2 (Fresh): 225 seconds total ~ 456 MB/sec avg read speed
While I don't have the original read speed results, I'd say there is really no degradation in performance so far with those rates.
I thought I lost my usb drive. 7 years later I found it under my car seat. Might have slip out of pocket while driving. Data still intact after 7 years of brutal environment inside car, hot, cold.
I have had a USB flash drive in my car for years and it is about 15 years old (4GB) but its for mp3´s for the car stereo and since i drive very little its seldom even powered up and i have had no problems and here in Stockholm Sweden the outside temperature can be over 30 degres C in the summer (much more in the car) and -20-25 in the winter in the worst cases and also quite humid.
USD driver is HDD not SDD 🤦
@@privateassman8839It's still based off NAND tho
@@privateassman8839 🤦hdd is hard disk drive which is the thing with the spinning disks
@@privateassman8839 It's flash memory, i.e. the same tech as an SSD. So not a hard drive, not sure why you are so confidently wrong?
I lost a Sandisk USB pendrive, 16gb, somewhere around the house, i moved 3 times during that time, i find it with the last move i made, it was stuck under a flap of a box of random stuff! Ow, how long was this pendrive lost? About 10 years, yeah, crazy, i know! And did it still had anything on it? Yes, yes it conserved all the data wonderfully, it had some pictures and music on it. It's insane that a usb pendrive can hold data for that long!
During my last move I came across a box of my old floppy disks. (5.25 inch and 3.5 inch) Although these were all in perfect physical condition (being kept in a sealed box the entire time of their storage) They had been sitting for the best part of 40 years. Not a single one of those disks could be read anymore. Some of them I managed to format to make them useable again but this was just out of curiosity. Most of them couldn't even be low level formatted any more. Seems the magnetic coating on those disks was not too reliable.
Not the same kind of memory.
@@KenFullman I had some Atari floppies in storage for 3 years. Not climate controlled. I have been able to read several with no problems. A few had 1 or 2 bad sectors.
@@KenFullman technically the only way that should’ve happened is if they were exposed to any some sort of magnetic field while they were in storage
Yup, the data is just not there yet. But with experiments like in this video we will get there.
Really nice to see someone testing this. Hope to see findings in a year.
It looks like you plan to re-use these for the 3 year test? Wouldn't the controller refresh those cells as soon as it got power again? I'd hope they would be capable of detecting degradation and refreshing cells, how else would they prevent drain while powered on? If this wasn't a normal maintenance tasks, they would lose data over time no matter their power state.
Thanks for the comment. I had the same question. I guess he just wants to check 2 years not 3. Othwerwise his test would require waaaaaay more SSDs ^^.
He actually answered to a similar comment from @mayonaiseking.
"When you plug it in, it doesn't "reset" or "refresh" the cells. SSD's have to write at the page or block level. This would take a lot of time to "refresh" the cells, at least as much time as it takes to write to the entire SSD. The SSD does not know how log it's been unpowered. The only thing that would trigger it would be the wear leveling algorithm or garbage collection. But those usually require some form of file addition or deletion."
If you never turn on the SSD it becomes Quantum Storage.
The oldest SSD I have checked was a 128MB thumb from 2005, lost since then in my junk drawer, until I FINALLY cleaned it out (almost 100% useless stuff). It was filled with pictures, videos, and documents I thought I had lost forever. It was fine. BTW, back then, that 128MB thumb drive sold for $18 at Target, which was a real bargain at the time.
$18 was pretty good tbh.
Thats a flash drive not an SSD lol. Very similar, SSDs use flash memory, but theyre a bit more complex on the inside than a thumb drive.
People are always super doom and gloom abkut flash media, SD cards etc over time. Its way overblown. If you dont abuse them and write read too often, theyll likely last a long time. Ive got SD cards that have been in devices with ocassional uae for like 15y and it still works fine.
I've still got a 32MB microSD card. I bet that was advanced at the time.
That's an SLC, 🥇 gold
Crazy to think about the fact that few years ago I paid around the same prize for 128gb thumb drive
I read somewhere that modern NAND flash is almost immortal when run in SLC mode, since the charge level in the floating gate always remains far below spec, leakage at those levels is miniscule, and signal processing is way easier when the controller only has to check for the presence or absence of a charge. Though using SLC flash for bulk long term storage would be pretty silly for most applications due to the price.
By the way, this video was a pretty good reminder to power up some of my old forgotten devices to keep their little machine spirits from starving to death.
You can buy a SLC ssd for about the price of a hard drive up to 2TB. Silicon Power makes some decent SSD's.
@@JessicaFEREM Thats for sure not SLC, you are probably reading some advertisement bs like "SLC Cache" and other stuff they shove into names to make you think its something better. All it means is that your TLC cells are used as a cache in SLC mode meaning that your drive is only fast when empty because they didn't want to include actual cache on the drive.
The fact is that although real SLC is rare and expensive it is worth it to those who need speed and longevity. I would still very strongly advise against storing any important data on just one storage medium type. Ideally the more important it is the more copies there should be as well as a diversity in the type of media. This way if one copy is bad you hopefully still have good ones left. This can be for personal documents that don't get shared with ANYONE or with ones that you do share for various reasons. Also if the documents backed up were scans of any paper copies keep them too. That just gives you one more backup even though there may be more labor in scanning them again. Then if all else fails hopefully you still have the paper copies. Just remember to store them in a cool dry place where insects will not get in and destroy them. Some people may think that multiple backups is overkill but they don't realize that keeping those is a good idea. They are just cynical and think it is more money and while drives are not free they are surprisingly cheap these days. A 1 TB 2.5 inch hard drive was just a bit over $50 for me after taxes.
Even the MLC SSD nowadays is almost impossible to find in ordinary consumer market and only limited to server/data center level
If I could reflash my 512GB MLC to become 128G SLC, I would definitely do that!
What I'm actually surprised by is the fact that those 70TBW SSDs survived 280TBW. That's impressive
70TBW is guaranteed number
@@GewelReal yeah I see now. It would take at least decade of normal use to get there even, so this is reassuring
Crucial is the commercial trademark that can last the most, its really fucking impressive.
@@Usernameblank9891 It’s _crucial_ for drives to store data as long as possible
@@Usernameblank9891Really depends. Their P4 Plus ist really limited compared to size.
Add to the statistics pile... I have an SSD as a boot drive in a rack server. I powered the machine down in the summer of 2016. Not expecting anything, I booted it back up about a month ago (2023) - so it had just sat for 7 years - and it was perfect! Everything fired up like nothing had happened.
For clearance you would have to make sure the hashes are matched, but ig it counts as a test.
I was going to comment on how useless a one year video would be (anyone who has known what an SSD is for more than a decade knows that they can safely expect one to last way way longer) but then it occurred to me: what if all the SSDs and flash drives that we have that have retained data for a decade or more have done so because they are a decade or more in age? What if modern SSDs are much worse? Thank you for this series.
I agree. You would need to have decades of drives to do this test and by then the units would be obsolete. Its a nice thought. I have a box full of old thumbdrives that i will probably never use, but when i go to look at one for making a giveaway content, i have to move data that has been sitting there for years and years (sometimes over 10yrs). I don't see data loss. Furthermore this test would be rendered useless unless people bought the same drives as you. I would for maintenance if just pluging in the drive is good enough, or you should move it to refresh it (ie does plugging it in refresh the state of all bytes to max) this test should be doable in some way.
I did buy SSD in the year 2011, but it didn't survive power surge. It took me a decade to buy another SSD, because data reliability is much more important than speed. However I also have an USB stick, which I bought in the same time and there are some MP3s in it. I tried to listen them this year and while you can still access them, some of them are corrupted. The corrupted data isn't constant. Part of the songs are all right and part of the songs have cracking sounds in it.
I was rather shocked to see the _manufacturer_ only claims a year! I'm certainly not going SSD any time soon!
I've got an old Kingston HyperX 2.5 that I still use to this day on random machines for a donor boot drive, etc. It's wearing down according to the S.M.A.R.T report, but still well within usable range even after ... 11 years.
But it does have some issues it appears in some part of it from time to time, and having (I think) identified the problem area of the memory in software side of things; I tend to just format around the sectors via byte based formatting.
Works like a charm after that, for the most part.
Meanwhile, some western digital m.2's I bought not even a couple years ago in total between them all have failing life spans despite being generally well taken care of. One of them outright failed I think in a bad enough way that it corrupted all sorts of boot files that to the best of my ability at least cannot be fixed without serious risk to the rest of the data on the device.
Sure, failures happen. But the other 2 similar m.2 drives I have aside from that one that failed; also have failing life spans already too.
Not a good look for WD from my angle.
Current machine being used has 2 P1600X optane m.2 drives. Low storage capacity, but so far absolutely zero issues. A little slower than the absolute fastest stuff out there now I admit; but stable as stable by definition gets as far as I am seeing so far. I've had these for about half a year so far. The WD's already started showing degradation by this point.
I love it. I might spring for the 500$ 960 GB U.2 version instead. (on sale right now.) If it's just as reliable as its m.2 smaller version, then I am quite happy to pay that much, and more; to go with a Raid 60? setup perhaps?.
Anyways. Aside from all of that; Samsung evo 2.5 drives seem to do well as far as I care so far as well. I've had a few in the past, but they got sold with other machines I built for people who didn't want to spend a lot on a new drive. Zero'd the data on those drives for them, did a test or two for reliability insurances, and didn't see any reason to not pass them on for a lower cost. Can't say for certain how long those are lasting those customers in the past, but based on how well my kingston one is doing and how well my other samsung 2.5's have done over the past decade prior to each's own sale as second hand; I suspect they are probably still working to this day. Aside from catastrophe somehow. I suspect my new 2TB one will probably last a decade before replacing it becomes a concern at all. Smart data on all my 2.5's of samsung brand in the past basically still had 97+% life span left prior to selling as used.
What else... oh; haven't booted up my other newest m.2 from samsung yet; cause the pro drives apparently need a firmware update right away prior to real use or you suffer a ridiculous drop in lifespan for the drive. no thanks.
So, I'm considering getting another optane drive for booting for that machine, and fix the samsung drive that way first before letting it get real use. Slight cost to the wallet, that mostly can be justified by just how stable those optane drives seem to be as boot drives.
Seriously. I can't begin to explain just how nice and stable this setup has been with them.
Does this also apply to USB thumb flash drives ? I have several 256MB USB flash drives that I purchased over 10 years ago and they all still hold the original data and pass all low level surface checks. I see from your video that SLC flash is more reliable than modern MLC flash memory. I think you should be writing to them 8 hours a day for a year before they start to degrade and not once then storing them away for year.
I just found a 12-year old USB drive recently.. a rabbit-themed Chinese New Year exclusive Kingston USB drive with a whopping 8GB storage, which was top of the line back in the day. Coincidentally this year it's the Year of the Rabbit as well, so it has come full circle. Worked fine. My files and backups from 12 years ago are still in there, and when viewed based on detailed properties they're all still showing that they're created or last modified 12 years ago!
Crazy how humans managed to create such micro architecture and it still managed to surpass over 4 times it's manufactured limit. Kinda mind boggling. Also great experiment, and an informative video about digital data storage to boot. Love to see more!
Think that's crazy back in 2015 TechReport had a Samsun 840 PRO reaching over 2.4 PB written without error which is 33 times its rated limit of 73 TBW, and in 2017 a German outlet had a 850 Pro 250 GB last a whopping 9.1 PB before dying which is 60 times its rated limit of 150 TBW.
Since SSD's, they are dancing to prevent parallel architecture. Every storage chip should be part of a "striping RAID" by default, pulling 100% of the available bus bandwidth for data transfer. The whole storage market became a collective scam.
Appreciate the time and effort you are taking to conduct these tests! Looking for to the next results.
I have a couple of 60GB slc ssd drives of 13 years old, not powered on for at least 10 years. Data was still fine and one drive still worked perfectly. One has become really slow, like 100KB/s write speed. These were used in a server and cost 750 euros a piece. It ran an intensive database application.
I use 2x 8TB qlc ssd drives for storage and backup every 1-3 months for 9 months now, so far no issues. The backup is powered off and stored at a differend location.
Check the SMART info on that slow one. When SSDs have used up all their hot swap cells they tend to drop the write speed way down to prompt people to back it up before it dies. Some SSDs when they fail will block writing completely. Some SSDs will self brick and make themselves unreadable so you data is inaccessible.
@@greggv8 that drive didn't have diagnostics I could read, maybe a proprietary tool would be able to but I'm not sure.
@@imqqmi From my experience the data in those flash drives outlasts the functionality of the controllers. All the broken SSD's I encountered had a dead flash controller.
For craps and giggles you could just wipe the slow drive and try to write then see if the performance returns to normal as some have experienced with some models of Samsung drives.
@@MrKillswitch88 would be worth a try because many SSDs slow a lot when they're over 50% full
Nice experiment. But wouldn't you test two years without power like this about three times?
After one year, you power the first set up, so they are basically reset. You check the same set again after (initial experiment start + 3 years), which is only two years after they last have been powered on.
And that is also the time frame for the second set.
Same goes if the three years would start after the first year, which makes four years.
could you elaborate please?
How do you actually factor in the reset of power levels during your tests?
Or are they not reset, because no data is written to them?
When you plug it in, it doesn't "reset" or "refresh" the cells. SSD's have to write at the page or block level. This would take a lot of time to "refresh" the cells, at least as much time as it takes to write to the entire SSD. The SSD does not know how log it's been unpowered. The only thing that would trigger it would be the wear leveling algorithm or garbage collection. But those usually require some form of file addition or deletion.
@@htwingnut Thank you for the answer it makes sense.
So basically your test is to check data integrity but if you take a brand new SSD, it should still work in 5 years even without power ? You would just lose your data ? The SSD would be still functional ? (if we assume actual material decay is negligible)
I've had laptops sitting around for years with SSDs and they booted just fine with all data intact many years later. About the only times I've ever heard about this happening was with MUCH lower quality flash chips.
No doubt. But the conundrum is that older SSD's were SLC or MLC which are more robust with more room for error. Modern SSD's in the last 5-8 years are primarily TLC or QLC which wear down more quickly and a small amount of data leakage can result in data corruption.
@@htwingnut I agree, also modern ssd's are now having to handle gigabytes and terabytes of writes instead of megabytes so a much greater density of bits increases the potential bit rate errors.
I data integrity to speed. Since I only use my rig for office and databases I am using a hdd. Has anyone tried a modem hdd? They are so underrated. I am currently booting my win 10 os in well under a minute
@@addydiesel6627 Yeah, I prefer hdd for storing docs, data, apps etc.. I only use a small ssd for booting the OS and a 2GB ram drive for my browser cache and temp files.
@@addydiesel6627what is a modem HDD?
It'll take about 70 years minimum to lose data from non-use.
Yes, I heard that too! ... They tested an SSD which was unpowered since 1953, and the data was totally intact! 👍🤣
@@marcse7encitation needed please
@@marcse7enI agree with bloodyiron here I wanna see that.
wow, succinct, no bs, no flashy garbo, just a simple, concise, and effective presentation. thank you sir.
Excellent research, was really wondering how much an ssd would last un-powered. I messed up my company's laptop ssd and I'm on parental leave for about 2 years. So it seems that time is not on my side. I highly doubt even the cheapo ssd's will fail to retain data even after 2 years of being un-powered or even minimal data loss. Anyway great video!
Even a cheap USB flash drive will keep data for 10 years - an SSD will keep the data well beyond even that. SSD/HDD manufactures are very conservative when they state the data retention dates.
@@lophilip Got any proof of that?
Cheapo USB flash drives invariably use cheap flash memories paired with cheap flash controllers. Multiple low standards, multiple points of failure. Which basically means that failure is almost guaranteed given enough time.
I've had USB flash drives corrupt or lose data after just a few uses. All it takes is one failed cache write, one power failure, one operating system bluescreen - and data integrity is at risk. Good for moving files to another machine, otherwise untrustworthy.
At least proper SSD drives (even external USB SSD drives) tend to have better silicon and better engineering. Along with failsafes built into the USB/PCIe/NVMe/M.2 protocols they use - and sometimes even onboard and offboard capacitors which can provide cache store and graceful shutdown in emergencies.
Had an ASUS tablet plus keyboard transformer like device not switched for almost 1.5 years and yep it lost it os drive and cannot boot. Flash soldered so haven't taken off yet to see what on them. Two other ASUS tablets, a win8 and android were still good with roughly the same power cying. All my ignored kindle readers one fire are all OK, but the batteries really don't like holding charge.
How long to flash bios chips last? I've got some 20 year old boards that are still going strong. It's not the same, but it's kinda the same.
Ancient phones that have been collecting dust in a drawer will still power up no problem (on USB power) after years and years. That flash seems to survive just fine.
Very cool. Just one thing I'd recommend: Write the data in a raw format to the SSD. I don't know how to do this in Windows but on Unix systems, for example on Linux, you can copy the bytes coming from /dev/urandom onto the SSD directly (without filesystem or partition table) and just calculate the hash of that.
For example:
sudo dd if=/dev/urandom of=/dev/sdc bs=2M oflag=sync status=progress
sudo sha256sum /dev/sdc # this will print the SHA256 hash
Then later you can just calculate that one hash again and determine if anything changed much more easily, without a filesystem. And it would also be more accurate since you're checking every single byte.
You're right, he should test for absolute changes. Filesystems and the like often have error correction which could interfere with these tests.
@@HappyGick most filesystems don't have data checksums (ZFS does, but FAT certainly doesn't). the real issue is that all SSD firmware actually uses a redundant encoding for blocks, and you can't count on being able to read data before it's been corrected.
so the hashing is probably pointless: the firmware is already detecting and correcting corruption, and won't return incorrect blocks (will return an error instead).
@@markhahn0 Then these tests would measure how recoverable an SSD is after it's been unused for years. There's no way to measure absolute changes.
A question I have not seen answered yet is, "how long do I need to power the drive on for to refresh the cell charges/data integrity?" Is it a case of just powering the drive on for a few seconds, a few days, or do the actual files need to be rewritten?
Or does permanent damage occur to the drive's data retention ability when left powered off?
What I've read is that it's advisable to either copy the data to a fresh drive when in doubt, or to run a full diagnostic on it so it's basically forced to address all the bits at least once. I that won't refresh the trapped charge I don't know what will.
ssd drivers doing it automatically via controller - manual copying is only for mechanical driers. @@Krawurxus
I had a CF card sitting there in it's container for around 20 years untouched. When I hooked it up to see what was on it... I still had all my photos. I had an SD card that went missing for me somehow. I found it well over 1 year later in the wash machine. It was still perfect and I am currently using it again now. If it can handle hundreds of washes in super hot water and then cold water... it's very durable.!
I found someone's cheap SD card in the dirt and snow on the street and it works perfectly clean. Although such cards very sensitive during writing, broken few by interruption and many fake on market from China.
@@fontenbleau Once in my life I bought a cheap card, thought they're all the same. it died 1/2 way through a trip. I said NEVER AGAIN. Since my CF card was SanDisk and it was always good. The one in the wash machine was a higher end SanDisk card. I bought the CF Card back in 2000-ish. It was $125 for a 125 Mega Bite Capacity! Now (because of that) I don't mind spending a bit for a card, they're all cheaper than that was!!
@@fontenbleauthat's funny you're mentioning China because I have a silicon power 32 gb micro SD card made in china that I bought in 2012 and I never had any data corruption and it's still working fine. It's currently in my Samsung galaxy a21s, that I use as a secondary phone
@@galaxiedance3135it's easy to spot cheap cards, they're for most part, not branded. SanDisk quality has also gone downhill recently and I prefer using Kingston instead
I never really thought about this before I have just transferred over 3tb of movies from two hdds to my second nvme 4tb drive in my PC and i would have just left it like that forever, but, oh the reality, i must think again so thanks
Not only does QLC seem like a horrible deal but they are always trying to make it denser too. I know everyone wants more but they don't realize what they are asking for.
Yes, exactly. Many QLC SSD's are no faster than, or in many cases even slower than even a laptop hard drive. The only real benefit are fast seek times.
I get why years ago there was a push for QLC since at the time flash was still expensive, but boy has that changed now! In todays newegg email there was a teamgroup 4TB SATA TLC SSD for $137! At those prices it's not worth using QLC drives now, even though their performance is almost the same as TLC on SATA drives. But during those years of QLC development and improvement TLC wasn't standing still. They went from 32 layer to 64 to 96 to 128, and they say more layers will keep being added. What I would love to see happen is that now that flash has gotten so damn cheap, why not make 2 & 4TB drives that are made of 100% SLC NAND? So instead of how it is now where a small portion of a drive can be used as a SLC cache, and after it's full write speeds fall off a cliff, 100% of the drive would be able to write at full speed 100% of the time. Yes, it would cost more than the TLC drives we are using now, but these would allow the SSD makers to once again have their "halo" products like the Samsung Pro drives used to be for Samsung. People would pay more for 100% SLC drives like those that do 4K video editing since as you noted SLC has the highest write cycle endurance.
I have a 240GB Samsung 840 EVO, which was the SSD that sparked the whole "Why are SSDs losing data when unpowered" frenzy many years ago. Samsung issued a software fix that they wouldn't say what it does to combat the problem. But that drive is still in this system and works fine, so whatever the fix was hasn't appeared to harm the drive. The rumor that went around was that it would "refresh" the nand by rewriting the drive over time. If true it must not do it excessively because the last time I looked at the drive health(which has been a year or so) it was at 99%.
2.5" SSD are mostly empty space, so why not pack more chips inside which are less dense? Problem solved.
@@radry100because it costs money. Chips cost per square millimetre to make. The packaging is a small part of the cost. The more you can store in a square millimetre the cheaper you can make an SSD of a given capacity.
Oh yes they do. They really don't care. "But it's still going to last longer than you will". No, not every ssd lasts people 100 years, dudes.
The 1 year retention spec is a minimum based on end of life with all write cycles exhausted. SSDs with only a few cycles on them will retain data for far longer.
What impresses me is how much he blew the write cycles out of the water, and it's still perfectly fine. Definitely interested for the two year data hold next year.
I started using SSD at work in around 2006. I'd backup configs and drivers to encrypted zips across two drives.
Out of 24 drives. 2 are totally dead (I haven't tried logic analyser on actual chips yet), rest are fine . Mixture of IBM, Samsung, Fujitsu, Seagate, Kingston drives. Some I'd already re-purposed for RPI. All the SSDs from 2012 are still fine in the Last PC I built.
It's really weird how the TH-cam algorithm works. I normally don't watch videos with only 40 views, but this one piqued my interest.
Great video! I'm looking forward to future updates. See you again in the coming years. :D
Hey, glad you liked it. Thanks for the positive comments.
This is great to test. There are some considerations, however.
Planar TLC drives are especially bad for cold(offline) storage, as 3D nand flash seems to have much better data retention characteristics across a worn drive.
JEDEC specifications for TBW are intended to be performed with the drive half full for 75% of the writes, I believe? This is down from previous testing standards that required 80% or 90% full TBW ratings, which some of these drives are rated for.
Colder operating temperatures for nand flash is more damaging for drive writes, as the electrons don't flow as easily into and out of the cell. Warm storage temperatures will also more quickly lose data, as the electrons more easily flow into and out of the cell.
There are actually whitepapers on testing NAND flash of various kinds, though they're awfully hard to find these days. I've read parts of one such paper that suggest early(32 layer) TLC Nand flash has a data retention time of around 18~24 months depending on wear leveling, whereas planar TLC had something more like 3~18 months as I recall? But, my memory is hazy on the numbers. Planar data retention decreased linearly with wear, though, whereas stacked NAND for some reason beyond my understanding retains data much more consistently until much closer to cell failure.
MLC has been used by Samsung to misleadingly describe 3bit and 4bit cells. Scummy samsung.
In theory, the number of bits per cell should predictably decrease data retention time. Taking SLC as a value of 1, 2bit should be 0.5, 3bit 0.25, and 4bit 0.125.
Older microSD or thumb drives can retain data offline for a very long time, because multi-level cell nand flash wasn't as available. Some of these may have not even used NAND flash at all. Some flash storage technologies can retain data for multiple decades reliably, making flash media technically the longest lasting electronic cold storage media, for very low-density storage.
They do tend to slow down with old stale data before they lose the data. ECC just kicks in more and more. I've seen that symptom countless times on high end name brand SSD's all the way to generic. The better drives of course will re-write a borderline sector if ECC worked a bit too hard. Moving the sector as a last resort. Point being is some drives might never lose any data if you read them each year. Too hard to call, too many controllers on the market. Good luck.
My dad recently found a micro SD card he lost in our gravel driveway. He plugged it in and it still had all the video footage from 2014 completely intact. Who knows how many times cars have driven over it and its been rained and snowed on during those 9 years!
This valuable research. Thanks for the time you spent so far on this. Keep on going! We are watching!
The world is heating up right now and we don't know if we'll be around next year, but if a the worst happens and i have a usb stick with 512 gb in a metal cage bured in the ground and i can recieve it if i survive, and it takes like 20-30 years before electronics industry starts happening again, is there any possibility that I could ever keep the data for that long? what type of digital backup is most long lasting, and also least sensitive to EMP damage?
A USB stick/SD card is NOT recommended for long term cold storage. From what I know, data tapes are the best long term solution of data storage.
I have experienced it myself 3 times, 2 SD memory cards and 1 USB stick with problems after years of just sitting offline with files on it. The USB did show some files, but ~60% was corrupt. Same thing with 1 SD card. The other card was not even recognized, and needed reformat again to be recognized = 100% loss.
If you really really want to do something like this, the obvious solution is to use it all: HDD, SSD, USB stick(s), memory card(s), CD, DVD, BR and data tape(s). That would really increase the likelihood that something survives, without breaking the bank.
I fear the day the NWO will deploy an EMP that mainly destroy crypto wallets, so they can fuck up everyone, and MAKE us use their global CBDC crypto BS to control the world. Are you naughty and jay walk when there's no traffic? Too bad, the CCTV with AI will catch it and draw money from your bank account right away. I started to convert from crypto to physical gold and silver years ago, and the gold price have really done well and made me a lot of money. Next in line is silver, I think, unless total destruction of the world. The only bad thing with silver is that it's heavy to carry 10x1KG bars. ;) Hopefully silver go up up up the coming next 10-25 years!
I like this video. It got my curiosity up, so I dug out a box of old thumb/usb drives as well as ten year old (give or take a year or two or more) sata ssd drives that were used in car washes, pos's, etc. While it took about five days to go through them, none of them shown any loss of data. It blew my mind that they were more reliable than the spinning hard drives that I had stored, and went through because I was curious. out of the 87 various thumb drives and 32 ssd drives none were bad. Out of the 148 ide drives four were bad and out of the 59 sata drives 3 were bad, they were all stored in working condition. Thanks for the video, and now I got a "gonna do someday" project behind me.
Still amazes me how much information we can get on such a small scale
I was concerned about Total Writes, but THIS is very bad - in fact I have a PC that hasn't been on for a couple years that has a PNY SSD in it (however I think I only put the swap file on it - so that isn't too dangerous.)
recently found an old 512mb usb stick, data is easily 13 years old and still fully survived.
as you note, TLC has an endurance of about 3k cycles, so your 128G SSDs should have a rating of about 385TB, and you took them to 280 or less. I don't know where the endurance rating of 60TB would come from - that sounds like QLC, actually.
thanks for doing this, even if it turns out not to be quite the worst-case you were aiming for...
Good point. I made an error. Newest JS600 SSD's seem to be QLC with 60TB TBW for the 128GB version: www.leven.com.tw/en/ssd/view/JS600-128GB
But original were labeled as TLC, and Leven confirmed 1000-3000 P/E cycles. Images for clarification: imgur.com/a/IkBmwf1
@@htwingnutback in November 2012 when I bought my LG Optimus l5 running android 4.0.3, I bought a silicon power (made in china) 32 gb micro SD card which was a class 4 speed card, pretty much common back in the days. I used it a lot and even though, it's now slow, it still works perfectly after eleven years.
Some people just love to put fake information on the internet like how SSD couldn't be trusted and how HDDs are great when it's the other way around
Nice. I am sure it will last very long. Otherwise we would be hearing more widespread issue from people about this. I do not think being unpowered changes anything really compared to being powered and not being written / read in particular pages / bits. Maybe it is related to temperature. Also according to JEDEC, there are derating tables, that depend on active use temperature and offline temperature. With higher active temp, and lower offline temp, helping prolong retention. A lot, sometimes by a factor of 10 times.
On the other hand I did not see many experimental publications on it either, so it would be interesting to see more results.
I have a server rack with few servers at home. Two of these servers I unpowered about 4-5 years ago, and they are offline. They are mostly storage servers with spinning rust, but boot drives are 2x SAMSUNG PRO SATA SSD drives in one, and 2x Crucial MX500 in the other. Or something like that. Before unpowering they were running for about 6-7 years. Not much load on SSDs, or too important data. Just usual Linux updates, some logs in /var, etc. So maybe 30GB written during their entire life. I should power them up, and check out.
I do have also an old laptop with SSD, that I maybe power once a year, with pretty old drive, and it is still working perfectly.
Since there are so many comments on finding old working SSDs or thunmb drives, i have one example for the opposite. I had an old 128GB SSD from a thinkpad, built around 2015. That was powered off for about 2 or 3 years. Afterwards it was no longer bootable.
My question is: when left unpowered, no charges can be refreshed (obviously); but what happens on that initial power up, after the storage period? Does the controller set to actively refreshing all the stored charges? It might be interesting to see the power data (of course SSDs are quite power efficient, one would need precise equipment to measure that) of a drive that is in regular use at boot or wake from standby, compared with a drive that has been stored for a year or more.
Measuring power draw is a clever way to figure out whether it's doing refreshing at power. Store the drive at a high temperature (e.g. 50C) to accelerate data degradation and speed up testing.
Some SATA drives record NAND writes in addition to host writes (e.g. Samsung enterprise drives), so that would be an easier way to see refreshing it's doing.
While not the same technology, the ROMs on videogame cartridges keep functioning even after 40 years.
Its tape obviously
Given that SSDs use non-volatile memory, pretty much indefinitely...
I never really gave this a thought. But I just dug out an old laptop that has an SSD that has not been turned on for at least 3 years and it booted without issue.
Good to hear. It seems there's a lot of unfounded fear online about SSD long term storage. However, there is no substitute for a proper backup to keep your data safe.
@@htwingnut Absolutely. Data is the one thing that cannot be easily replaced. I image all of my machines regularly (at least the ones I use).
just recalled that I have a 512G ssd just sitting somewhere with some cold data. I would love to move them to HDD's but because of other reasons cant really afford more now haha.
I need to check if it's still having my data. It probably is not powered for just like 8 months now but eh
@@hariranormal5584 just try to full error check so the data is rewritten (recharged), not to often tho.
I had on SSD (samsung 970 evo) which was left unused for 5 years in my old laptop. I then put them in my new laptop after 5 years and to my surprise the data on the ssd was still intact. So its 1 year at the very least.
Does the act of powering on to read and verify the data integrity reset the 'timer' on its longevity, or does it only reset when it's overitten?
From my understanding, it will only "reset" when it's overwritten. Of course SSD's are "black box" anomalies with highly secretive algorithms, so we never truly know what goes on inside unless we work for one of those companies, but then it'd be under NDA.
But as I see it, a read and write operation are separate entities. There's no reason to affect the state of a cell with a read operation that would cause it to reset it to the appropriate voltage. That would take more time and wear on the cells. SSD's can enter a "read only" state eventually because they can no longer write to it, otherwise it risks data corruption from the breakdown of the insulating layer.
SSD don't have refresh, so powered or not is exactly the same. This video is made with ignorance.
I recently dig up some microsd card that were unpowered for like 5 years. It's most likely MLC, since it's a 4GB model. For now i had less luck restoring data from HDD, but controller failure are more common on cheap drives, not happened to me yet.
That’s just unproven data. I bet that using a good modern ssd you’re safe for minimum 15-20 years. For best models ranging even 50 years of expected data loss in case of outage.
Plus you can minimize this risk by simply connecting power to your drive once every 15 years to refresh the state of the transistors and “charge” them for another 15 years
@@w140 Unlikely. Older SSD's were based on SLC or MLC, which are a lot more robust, because they have a lot more room tolerance for voltage leakage. Most modern SSD's are TLC or QLC, and soon PLC (5 bits per cell), which have to be very precise with the voltage level they read, otherwise it could result in incorrect data, meaning your data would be corrupt.
Connecting power to your SSD doesn't "charge" the transistors. They have to maintain a specific voltage and refreshing the cells would mean it would have to read every cell, assuming every cell was accurate, and then rewriting every cell to ensure the voltage levels were "refreshed". This would take an inordinate amount of time and extra wear on the transistors.
@@htwingnut well, I never ever saw a hard drive to fail due to “sitting long”. And I’m into this game since 386’s. I personally have 14 year old ssd that still works to this day. And 14 year old info is there and accessible. I didn’t “rewrite it”. SSD’s wear out because of cycles and if you’d like to use them in archive - running a time costly refresh every 15 years is forgivable, you don’t use much cycles in archive mode anyway + you’d better buy new hdds or use an alternate method for data storage like cloud + copy on external drive + raid1 if you’re extra paranoid. But I don’t see any issue in ssds longevity so far except cheaper models tend to stop working altogether at all.
@@w140 Thanks for the feedback. I've also been in the PC and IT field since the 1980's, and understand what you're saying.
I'm not concerned about longevity personally, as I follow proper 3-2-1 backup for my important data, as well as the same for my clients. But there is always a question out there about the reliability of data when left untouched.
Like any data that is left untouched, it's Schrodinger's Data. It both is and isn't good until you verify it's valid.
This is really just a fun test to have a somewhat controlled experiment and see what happens.
@@htwingnut I get what you’ve said and I just wanted to say, that it might take way more time than expected:) maybe create a branch out of your experiment of test with some extreme variable, like temperature? I’m personally very interested, how hot ssd can get and still remain the data. I just believe that they’re pretty robust and might exceed your expectations on that behalf. And I truly believe it would be a fun experiment to watch. Thanks for your videos and replies also!
Hmmm, I'm not sure this test would work using this method since connecting the SSDs every now and then runs a built-in standard refresh operation in the background, so every year you would be practically just refreshing the decaying cells for a potentially infinite period of time until the cells actually get worn out and die.
Interesting. just popped an old 16gb usb stick that hasn't been used since 2012 in my computer. looks like all the data is there.
Great research, brother. I'm sure it will make many computer-nerds like meself sleep better at night.
Probably no need to continue the test year after year. The data storage will out live us leaving perfect data intact. That's the nature of molecular charged states. Metallic hard disks also will last for many years but already fading away just like the Model T Ford did.
I understand that the temperature at the time of data being written also has an impact on powered off data retention time, general idea being that low write temperature = bad, high write temperature = good, while storage temperature is opposite - low storage temperature = good, high storage temperature = bad.
Did you come across this in your research?
I came across this: An end of life device (i.e maxed out on writes) when stored at 10 C was good for about 75 years if written at 55 C. The JEDEC standard says that a "client" device written at 55 C and stored at 25 C should be good for 7.75 years, but if written at 25 C it is only guaranteed good for about one year. Who knew? Well Intel actually. They provided the data to JEDEC.
why did CrystalDiskInfo show good on those 50% drives?
Does the health status depend solely on errors within the drive?
This gave me an idea for a video..how long can my 4 week worn socks be stored in a drawer and retain their smell.
I have 3 SSD's I did not use for over 3 years and 2 of them when connected to a PC would not be recognized by the operating system any more..... one survived but I no longer trusted it. These were a mixture of older OCZ Vertex and Crucial BX/ M drives.
OH NO !!! I have a MX500 now I am scared :C
@@Scr3amer42 I found them very reliable over many years. It was my fault to leave them unplugged for too long.
@@micb3rd I am not sure plugging them is enough. The author of the video mentionned in a comment plugging an SSD doesn't refresh the data. Data is refreshed when rewritten.
I need to investigate that claim.
I was listening and not watching and thought I was listening to a young Norm MacDonald
An SSD that is powered on after being off for a year has its memory cells refreshed. Any subsequent retention tests for the SSD essentially start from scratch.
That is not how I understand it. SSD's can read data by the cell, but have to write data by the page (containing thousands of cells). This takes time to do across the entire disk and added wear to the cells. Plus an SSD does not know how long it has been unpowered for. Usually a file deletion or TRIM operation will trigger a garbage collection and/or wear leveling routine. But if left untouched, there's no reason for the data to be "refreshed".
Regardless, this will at least test for a two year timespan, or three depending on how you want to look at it. Which any data left untouched for more than a year without validating it's still good is as good as dead data in my opinion.
@htwingnut Boruut209 is correct in what they say. You will find that any data read, if there are errors, will be refreshed. I would suggest you check the number of crc errors at test start, and after doing the read verification after a year/2 years to see if you see more over time. Unfortunately it’s too late to get the original values.
I have several flash drives that been unusable for several years and the data on them still intact
So what happens when you read the data, but the firmware silently rewrites the data if it detects any errors or weak reads?
if the bit rot isn't bad, the controller silently "fixes" the results. this is actually the norm. the firmware may decide to move the block to a less-worn place. if the rot is bad enough, the error correction will fail and it'll return an error.
@@markhahn0 right, I am just pointing out a potential issue with this periodic read the data test.
What would be nice would be the option (via a vendor-supplied app that interfaces with the drive's firmware) to irreversibly downgrade the drive to MLC or SLC, of course there would be a reduction in capacity but write endurance/useful life would be far extended
This would probably depend on if the controller chip can do that, too
at this point its safer to keep on a ssd then spenning hard drive
Specification requires 12 months retention without power. Research shows minimum of 2-5 years w/o power. And some SSD manufacturers claim 15-20 years for their products:
"Will SSD lose data without power?
Overall, if SSD is not getting power for several years, it may lose data. According to research, an SSD can retain your data for a minimum of 2-5 Years without any power supply. Some SSD manufacturers also claim that SSD can save data without a regular power supply for around 15 to 20 years. Jul 20, 2023"
"Modern SSDs are only guaranteed to retain data for 12 months without power. They usually exceed that, of course, but that's all the specification requires. All SSDs gradually lose data without power, as the tiny charges in the NAND flash floating gate transistors slowly leak out."
You number are WAY off. MLC has 3,000 write cycles, TLC has only 1,000 and QLC has a shocking 250 write cycles.
I've seen it vary, but generally speaking from my research, QLC is 1000 or less. Most TBW ratings are overly conservative rating it from about 200-400 TBW per TB, which is usually based on host writes, not total writes like from write amplification.
7:49 too many read/writes on the brain, just like me
Like the other commenters, I have also found super old thumb drives and such without any issues. My favorite is a little mini mp3 player I bought around 16 years ago. I put my entire library on it at the time and used it daily pretty much up until 2010 when I replaced it with my first smartphone. It has sat in various drawers and whatnot over the years and still works to this day.
Thank you for great testing. If only you didn't use the word "basically" all the time multiple times per sentence...
Basically, that's good feedback, basically.
With the test plan being read every two years, won't the test be a matter of "will aging drives retain data and be useable two years between power ups, and does thrashing a drive affect that?" A useful test up to its limits but not a test of "how long will a drive retain data unpowered?" For that you'd want something like say ten sets and read one set every year.
If I wanted to avoid data loss how long would I have to keep an SSD NVMe drive plugged in for in order to ensure multiple terabytes get charge refreshed? Is the process instant or does it take time to apply to each cell?
It takes time. Unfortunately firmware code is under lock and key, so we can only guess based on subjective testing. But if you manually initiate a TRIM command, let it sit idle for a couple hours, it will likely go through and refresh everything that needs refreshing.
If you want to force it, then only sure way is to copy the data off, format the disk, copy data back.
ChatGPT:
For consumer-grade SSDs, you can generally expect data retention periods of 1 to 2 years at a minimum, with many exceeding 5 years if stored under typical conditions.
For enterprise-grade SSDs and high-quality, well-maintained SSDs, you can often expect data retention periods of 5 to 10 years or longer.
Thank you. I also asked ChatGPT and got a similar answer. However I also asked if it could site source references and it said this was from general consensus. Not to mention, those spans are for SSD's within TBW ratings not 3-4x the TBW rating.
Also, the Jedec standard indicates the opposite. Enterprise disks tend to be run harder and longer and hotter as well as stored in higher temps and indicate data retention less than a year.
You need to read SMART or other internal diagnostics/metadata from the drive. There is always error correction that fixes limited errors. Then the error rate overwhelms the error correction and the drive fails quickly with a lot of unreadable blocks. You need exact error rate before the error correction is applied otherwise the test is kind of useless.
If you have the error rate before error correction you can somewhat reasonable predict when it starts to fail for real because you can see how the error rate increases
Thanks a ton. This is a really valuable underrated experiment and I'm grateful that you have the patience and the go to try this.
Time powered on is the enemy of spinners, writes/rewrites are the enemy of ssd.
Very interesting. I'm going to guess that, as long as they're kept in a cool, dry place 5 years shouldn't be a problem, 10 years should still be fine but maybe expect some errors and 15 years + is pushing it. I would also guess that mechanical drives are probably better for storing really long term, like for family photo's and things, but then again they have moving parts that could fail. The problem is that it's quite a hard variable to test without spending the actual time testing!
No problem, just live longer. I'm testing vinyl records since 20th century, soon i'll start with ssd to finish testing in 22nd century.
I have 20 year old smart media that is stil good
think about how many times youve found an old usb drive in a drawer and plugged it in. I dont think i've ever encountered corrupted files. Usually when I encounter corrupted files it happens pretty quickly for a wide range of reasons, but never due to age alone.
See you next year on this subject. 👍
Doesn't powering them on after a year to read them for validation reset that clock though?
Yes! In order to test data retention over time, the SSD must be totally unpowered for the test period.
the data quantity remained in tact ..but was there any errors/or bit flips .. eg:same quantity but imperfect data .. i wonder how it will be after 5 years
anyone else find it.. rather deceptive that no SSD maker lists how long the drives retain data unpowered?
It would seem that they don't know exactly 🤷
Though not an intended experiment, I've kept an SSD with Windows 10 installed for a couple of years now which I use to boot newly built PCs to test their function before I install an OS permanently; usually some version of Linux. I've not noticed any degradation at all nor ever thought I might until I saw this video. Certainly my experience is not definitive.
Cool experiment 😉. thanks!
it's very satisfying to see the proof in the pudding. thanks for running these tests and making the video
Does the second test of the same drives make sense? They've been plugged in for reading, so they have received power.
It will still have gone 2 years without power at that point. Reading technically should have no impact on its retention.
Those who know the secrets don't need ssd
I have seen drives losing their data after just 3-4 years.... some friends didnt listen and just stored their photos on an SSD and when it was full just shelved it. Now they regret it but there is basically nothing you can do at that point - many of the images were corrupted (like serious double-digit percentage).
Also by reading the entire drive the controllers will check the voltage-level and depending on the manufacturer/model already start refreshing the data. Some years ago Samsung had the problem that their entrylevel SSDs had some serious performance and integrity degradation and the fix was that the controller would permanently monitor the sectors and re-write them if there were too many nearby reads and after some time-interval. Pretty sure since then most SSDs have implemented something similar as that was like 10 years ago already.
Still though, that is a very good point on why to not keep a backup on just one type of storage media and also why it may be best to avoid buying the cheapest SSDs. 1000 write cycles is not really a lot at all.
Early SSD were not exactly known for reliability, controller issues of big brands like Samsung were in IT news almost every day about ~13 years ago. The situation has improved a lot.
Yeah, you always need to have backups, electronics go bad sooner or later, some can last very very long and some very short.
You could have a job for life.
Hi. Very thorough, good scientific video. May I ask, do you know if data retention is better with nvme vs sata ssd? Or, due to the fact that they are both flash, they both suffer from the same issue? Thank you.
They both use the same type of NAND flash chips. So it would be the same result. I do personally think it's a non-issue for the most part. Just don't let a disk sit whether it's SSD or hard drive for any length of time without at least powering it on once or twice a year and verifying the contents are ok.
By "verifying the contents are ok" can I assume that simply plugging it in and randomly playing a video or an mp3 from the disk should suffice? Or should I scan the whole drive with Windows' Disk check or Mac's Disk utility in order to check for errors?@@htwingnut
i have yet to ever have a problem and i have been using SSD drives for years and never a problem when off. so i would not take to mich value in this sort of thing because its just not a thing
So the better solution is Hdd for data you want to keep for years? Or Tape...
Yeah, sounds like it, although some people are saying that some of their flash storage has lasted a lot longer than that.
And supposedly optical discs would last several decades, but then on the other hand sometimes you hear about bit rot and delamination that will happen way before, so those are concerning also. And of course different disc qualities will play a role in that storage length too.
Powering a drive renders it's next test meaningless as it will refresh it's cells.
You will not have any drive to test after next update. Unless you want to wait 3 more years.
Powering a drive on does not render its next test meaningless. SSDs store data by voltage state trapped in a cell. SSD's would have to rewrite the entire disk to do so, and that does not happen.
Some data would be "refreshed" with wear leveling and garbage collection routines, but that typically doesn't happen until you delete files.
Reading the data just reads the voltage state. It doesn't apply a voltage. This is why SSDs will go into "read only" mode after they exceed a certain amount of wear.
@@htwingnut Reading the data on a NAND flash with multilevel recording will force any pages with cells showing signs of drift to be refreshed. How the drive does this varies, and tolerances will probably vary too, but they may be reallocated in the same process. This is invisible to the OS of course.
Edit: IN order to use the same drive for multiple tests over many years, I suggest writing multiple copies of the data and only reading the first copy after 1 year, then both first and second after 2 years etc. This way you can even catch the drive if the first copy always works, while later fails.
Drives go into read only mode if the writing has too many errors, not from a set amount of wear. Some drives never do this. They just keep returning crap data once cells are too worn.
I found a similar test to this on the internet, but it was a lot more comprehensive, and it included recording temperature and storage temperature. The higher the recording temperature, the longer the endurance; the lower the storage temperature, the longer the endurance, so you want to record hot and store cold. You can also use a parity archive tool, which uses a repair file to fix your corrupted files. The larger your repair file, the larger the amount of data you can fix. About once a year, I make an image of my hard drive which I put on an SSD, add a parity archive file (made with Multipar), and store the whole thing in the freezer. I could be wrong, but I reckon that shold be safe for a minimum of 5 years, perhaps even 10.
Each one of those gates is like a mini battery, holding a cluster of electrons as a charge....
If you toss a battery into the freezer then that's gonna go flat real soon.
I don't know why or how your strategy could work, but I hope for your data's sake that you're right.
I'm not confident about my 1:1 comparison of battery:NANDflash transistor either, but one of us is gonna be right in the end haha.
@@Yezpahrbatteries rely on chemical reactions. NANDs not so much.
Thank you for your contribution. Experiments like these are hard to come by and very interesting to me.
This is why I doubt HDDs and Tape drives will vanish anytime soon, they are super reliable as backup drives using tapes as tertiary backup to said HDDs.
Usually for commercial use... SSDs >offload> HDDs >tertiary backup> Tapes/"DAT".
However SSDs are suprisingly reliable and can and will hold readable data for quite some time, tested this on an old 40GB SSD. 3 years later data was still perfectly readable ive been using the same 180GB OCZ SSD as a boot drive through three different PC builds, when I first baught it back circa 2012. according the health stats its still 98% MTBF health.
I bought one in 2022 and barely used my computer for almost a year. When I booted up the bios couldn't find my WD ssd. I kept it but replaced with an NVME. Pulled it apart and everything looks fine. I'll throw it in my laptop and see if it goes.
This is seriously incredible. What a wonderful experiment. Thank you for doing this!
Well, wow! A test that takes 3 years to finish, thanks for doing this!
Very good methodology. These kinds of videos are always interesting. And kudos on being able to resist checking them before the year is up. I couldn't do it, lol.
The Intel 2.5" enterprise/datacenter-class 256K SATA3 SSD that I used as my primary/boot drive in my 2015-2019 build sat in a box from 2019 to 2024. I recently put it into a 2.5" USB enclosure & all the data is still intact & it works fine. Also to note that approx. 4 years of use in my 2015 active daily-use gaming/home system only used approx. 1.5% of it's stated endurance rating (1700 TBW) for writes. These sorts of professional-grade, higher-end component SSD's are worth the extra money IMO just for the ultra-reliability peace-of-mind. Intel sold their SSD business to Solidigm some years back & those are the types enterprise-class SSD drives that I have also used in both my 2019 & current in-progress 2024 systems (both M.2 form factor & 2.5" SATA3 versions). Love them.
An absolute wonderful, gem of a video. Loved to see the long term planned video, this what makes the Internet so incredible.
The longest unpowered time is 2y for both cases between checking steps, would've been nice to have 1y,2y,3y instead.