[TR] SSD endurance experiment ends at 2.4 PB

[johny12]

it sucks the intel did that. I under stand why it would do it, but as a consumer I would want the ability to keep going (ie: a pause in the boot once reached it's limit on every power up). But going dead without warning is not acceptable to me for a drive sold to the general market. If being sold as a server grade device, fair enough.

Re sandforce, that is just dirty. Care more about making it to the warranty expiry more than given the consumer what they paid for / expect. I have not been a fan of sandforce controllers (most of my usage patterns are not friendly to compression), but that warranty idea just makes me think even less of them. Might was well add them to the group of companies that put page counters in the ink cartridges of printers to "kill" them after a set number of pages are printed, even if there is still ink left, or even by selling printers with "starter" cartridges which make the whole new printer cheaper than a single replacement cartridge.


/end rant

That warranty period comment got me worried, but after reading around on some other sandforce controller SSDs it seems that setting is not universal and most of the SSDs with that controller don't do that.

 

[JimmiG]

Read the article on how the intel 335 died! "Intel's 335 Series failed much earlier, though to be fair, it pulled the trigger itself. The drive's media wear indicator ran out shortly after 700TB, signaling that the NAND's write tolerance had been exceeded. Intel doesn't have confidence in the drive at that point, so the 335 Series is designed to shift into read-only mode and then to brick itself when the power is cycled. "

And some sandforce drives had lifetime throttling which reduced throughput according to the warranty (from previous article on SSD endurance test) So you would write to the SSD for 3 months straight and it would then punish you for the next 21 months at 1K/sec write speeds to protect its wear!

I can understand switching into read-only mode, but why self-destruct when the power is cycled? Didn't it occur to Intel that people might still want to access the data on the drive?

As for HDD's, I've only had one sudden HDD crash with a 74GB Raptor way back. It just started making metallic noises and you could hear the spindle motor working harder. Ouch.

Some have developed bad sectors and struggled a bit when trying to read the data, but it has never become completely unrecoverable. My oldest drive right now is a 640GB WD from 2008, still no SMART-warning anything.

If you back up regularly, a HDD/SSD crash is just a minor inconvenience anyway. If you have the bandwidth, cloud backups work great. You could also use NAS or simply plug in an external USB drive. Personally I back up to one internal drive, one always connected USB drive (it goes into standby, so not spinning 24/7), and occasionally to another external USB drive.

 

[CiPHER]

All these 'endurance tests' are misleading because it implies the SSD would actually be usable when it is still working. That is, however, not really the case because as the flash wears, the retention drops. And the retention is the whole reason SSDs have limited cycles.

When the MWI (maximum cycles on paper) reaches 0%, it is assumed under JEDEC spec that the flash still has 12 months of retention left. When the MWI is 100% thus a brand new SSD, the retention is about 10 years to 100 years. Once you go below the 0% MWI, the retention can become a big problem in reality.

However, this problem is obscured through these sort of tests because the flash is overwritten very rapidly and doesn't get a time to settle where retention becomes an issue. So these endurance tests are not all that helpful, and in fact can be very misleading.

 

[Cerb]

All these 'endurance tests' are misleading because it implies the SSD would actually be usable when it is still working. That is, however, not really the case because as the flash wears, the retention drops. And the retention is the whole reason SSDs have limited cycles.

When the MWI (maximum cycles on paper) reaches 0%, it is assumed under JEDEC spec that the flash still has 12 months of retention left. When the MWI is 100% thus a brand new SSD, the retention is about 10 years to 100 years. Once you go below the 0% MWI, the retention can become a big problem in reality.

However, this problem is obscured through these sort of tests because the flash is overwritten very rapidly and doesn't get a time to settle where retention becomes an issue. So these endurance tests are not all that helpful, and in fact can be very misleading.

For actually determining longevity, yes. For placating the fears of those that would go and put their temp folders and page files on HDDs, it's just the sort of testing and results that are needed.

 

[CiPHER]

For actually determining longevity, yes. For placating the fears of those that would go and put their temp folders and page files on HDDs, it's just the sort of testing and results that are needed.

Explain please!

Because, if their fear is that such usage would actually get them close or even below 0% MWI, these tests do not mean much at all. They do not give any measuring of how common usage would lead to wear. For example, pagefiles mostly just sit there doing nothing. The actual act of swapping is quite rare and as such does not wear an SSD much at all.

All these tests do, is hammer the SSDs until they break. They give misleading results because people think 'oh see it takes much more than 0% MWI for them to break'. But what the tests do not reveal is that beyond 0% MWI the SSD becomes unreliable to the point of being unusable. This is something that is not measured, nor mentioned, and as such is misleading in my opinion.

 

[Cerb]

Explain please!

Because, if their fear is that such usage would actually get them close or even below 0% MWI, these tests do not mean much at all.

And, how are you going to get there? Those fears are unfounded, because it will take many PCs' lifetimes to get even close.

 

[CiPHER]

True, but if you want to make a substantive argument for that case, you should think of a test that actually addresses that point. So use the SSD in various scenarios that match several consumer profiles; like one who games a lot, one who only emails and browses the web and one who transfers a shitload of data with video editing or whatever.

Then you can say - without testing for a long time - that after x months of such usage, the MWI is still at 100%. So then you can calculate the actual time it would take for those usage profiles to bring the SSD down to 0% MWI. That is the boundary where JEDEC says 12 months retention is the minimum for consumers. And i think that is about right. Because those 12 months is an average; individual pages will vary and thus weaker pages will turn into unreadable pages. This hurts the reliability of the drive. This, however, does not get tested and in fact totally obscured by the endurance tests that everyone is so fond of, because NAND is overwritten over and over and does not get any time to settle so that retention is nullified. That is totally unrealistic and no usage profile would tally with that kind of testing.

 

[Cerb]

True, but if you want to make a substantive argument for that case, you should think of a test that actually addresses that point. So use the SSD in various scenarios that match several consumer profiles; like one who games a lot, one who only emails and browses the web and one who transfers a shitload of data with video editing or whatever.

Then you can say - without testing for a long time - that after x months of such usage, the MWI is still at 100%. So then you can calculate the actual time it would take for those usage profiles to bring the SSD down to 0% MWI. That is the boundary where JEDEC says 12 months retention is the minimum for consumers. And i think that is about right. Because those 12 months is an average; individual pages will vary and thus weaker pages will turn into unreadable pages. This hurts the reliability of the drive. This, however, does not get tested and in fact totally obscured by the endurance tests that everyone is so fond of, because NAND is overwritten over and over and does not get any time to settle so that retention is nullified. That is totally unrealistic and no usage profile would tally with that kind of testing.

But, then you have to explain what all those things mean.

 

[CiPHER]

Well, informing the public about these things seems like a more noble goal to me then misleading them with tests which are unrealistic and obscure the things that are truly important to ordinary consumers.