SSD Endurance - the Big Lie

Let we use this thread to compare real data, in this aspect it would be helpfull for others to decide which SSD to buy. Everybody should post theirs:
- SSD model (firmware)
- time used in months
- host writes
- wearout indicator (if is)
- type of use (laptop, desktop, video editing, browsing net etc.)

---------------------
- Intel G2 80GB X-25M (2CV102HD)
- 11 months
- 2,10 TB host writes
- 98%
- laptop, most browsing but also installing lot of software. 8h a day

 

ah crap ... this thread made me thinking again before spending 400+ USD for that Samsung 256GB ... when I found they claim only 3,000 write cycles ?!?!?!

Samsung's 470 Series solid-state drive - The Tech Report - Page 1

still, everyone that has this drive is pretty happy :|

great info on here though, I just went through the whole thread again. 512GB SLC ... where are you ...

 

Well, even "only" with 3000 cycles, a 256GB drive has a nominal limit of 768TB writes to it.

People like Tiller who do lots of writes probably won't be happy with that, but it's more than plenty for the majority of us.

 

It would take years for you to notice a performance drop...and then SSDs will be dirt cheap.

 

How did you get it to display estimated lifetime?

 

If you install SSDlife on a new SSD it can't measure the life until you've used it a bit. Give it a couple days/weeks usage then run it.

 

Crucial M225 256GB (Indilinx Barefoot) - Shipped 14th Sept 2010 (about 5 months usage):
HDD Life OnLine Drive is actually on firmware 2030.
After I got the C300, I moved all my data from the M225 to the C300. Near the end of it's use on my main laptop, the M225 was reporting about 3 months of 'life' left.

Crucial C300 256GB - Shipped 10th Feb 2011 (maybe a month of use):
HDD Life OnLine Drive is actually on firmware 0006.
Please carefully note that while the 'life' reading is accurate for this SSD with the SSDlife application, a number of the other SMART statistics are mislabeled.
Currently reports about 5 years of life left.

Both are 'rated' for 5000 write cycles.

 

no, that part is true (+amplification, as stated)

that doesn't make the above non-true. the amount of ssd-written data doesn't always correlate to the amount of os-written data, there's the actual point you wanted to make.

problem there: the os does not write your 1bit as one bit instantly, anyways. file systems have caches, and are more complicated to update, even if the file you save just has one bit (the filename doesn't, and the updated file structure means even more writes).

in short: the file system collects small writes to a certain amount, and writes them at once. the ssd, AGAIN, does that, too. so the one bit write will not happen for a while (most likely by then, you've written other stuff, too), and the ssd will wait to write it onto flash for a while, too. again, collecting more stuff.

oh, and even if it can't wait for long enough, it'll just write to a free spare cell. then, once enough such spare cells have "1bit data" on it, it collects it to one. so while even in the worst case, yes, more clears have to happen, they're not that much (not the same as if you'd re-clear the same cells all the time).

so in short: yes, it isn't a 1:1 thing, no it's not way off in real world use cases (for home/work systems).

BUT this is the reason why one should use slc flash ssds for databases. there, caching is not allowed for data-reliability reasons. and there, tons of small writes happen (like up to millions of times more "1bit write" scenarios than will ever happen on your desktop/laptop). there, it actually matters.

so your point matters where ever no small write caches are allowed. that is no-where in homeland. it is everywhere in database land.