bit-tech.net

Flash Destroyer tests flash write cycles

Flash Destroyer tests flash write cycles

The Flash Destroyer tests EEPROM chips to destruction to see just how many write cycles they can take.

If you want to find out how much abuse your SSD can really take, there's only one thing for it: test it to destruction with the Flash Destroyer.

The Flash Destroyer is a kit from the accurately named Dangerous Prototypes site - via SlashDot - and exists for one purpose only: to write to a connected EEPROM and verify that the data is correctly stored, over and over again.

The front of the device features a counter which indicates how many write cycles the connected flash chip has suffered through - and when the drive starts to fail the verify cycles, the numbers are frozen to give a true indicator of exactly how many write cycles you can expect to get from that particular model of flash memory.

It's a neat idea - and in keeping with Dangerous Prototypes' open-source philosophy, full source code and hardware diagrams are available on Google Code if you fancy building one yourself. Sadly, the name is a bit of a misnomer - the device is designed to interface with EEPROM chips rather than the type of flash memory chips found in your average SSD, and there's no support as yet for simply hooking up a storage device via SATA.

If you do build your own - or if you buy the pre-prepared kit - it's probably worth pointing out the obvious at this point: this will destroy any connected chip in short order. While it's useful for testing, it is a destructive process - so it's probably better to just watch the device's creators destroy chips on your behalf.

Do you think that the Flash Destroyer could prove a useful tool in discovering just how the limited write cycles can affect SSDs, or does its destructive nature mean it's more of a novelty than anything else? Share your thoughts over in the forums.

23 Comments

Discuss in the forums Reply
mi1ez 28th May 2010, 14:55 Quote
good for websites (such as yourselves)
bad for the over-interested enthusiast (like us)
V3ctor 28th May 2010, 15:40 Quote
yes... good... *Mr. Burns voice*
phantombudgie 28th May 2010, 15:45 Quote
I am perfectly capable of damaging computer equipment on my own, without getting help and pointers from others!
rickysio 28th May 2010, 15:56 Quote
It seems that the only reason this would exist is for people to mind numbingly stare at the everincreasing number displayed on the screen.
Zurechial 28th May 2010, 16:04 Quote
Something like this allows us to get real-world numbers for a statistic like an EEPROM chip's lifetime instead of just taking for-granted what the manufacturers tell us.

We've all seen the disparity between the marketing claims of performance, airflow, noise etc for fans, and their real-world performance in testing.
How many other devices do we buy, taking for granted what the manufacturers tell us about them without ever having the tools or know-how to test their claims?

It's a risky line of thinking, leading to tinfoil hats and wild theories, but it's still nice to see something like this out there. :)
Flibblebot 28th May 2010, 16:17 Quote
Surely you could do the same thing for SSDs in software? You wouldn't need any additional circuitry - just plug the SSD into a PC and let the software do its thing?

Still, either way, it's not something your average user would need - I couldn't afford to throw away a couple of hundred quid just to see how long my SSD might last...;) It's more suited to test and review sites like bit - although I'm not sure how a manufacturer would like you testing review samples to destruction?
HourBeforeDawn 28th May 2010, 16:29 Quote
this could be nice for real reviewers who have the readers truly in mind and go the distance in showing their readers what you really get.
RedFlames 28th May 2010, 16:34 Quote
Quote:
Originally Posted by HourBeforeDawn
this could be nice for real reviewers who have the readers truly in mind and go the distance in showing their readers what you really get.

except it's be hard to produce accurate numbers as each drive would be different... one drive may last twice the manufacturer's quoted number of read/writes, another may keel over after only half the quoted number... it's like overclocking your mileage may vary, just because one chip will hit 4GHz doesn't mean they all will...
whiskers 28th May 2010, 16:44 Quote
Quote:
Originally Posted by Zurechial
Something like this allows us to get real-world numbers for a statistic like an EEPROM chip's lifetime instead of just taking for-granted what the manufacturers tell us.

We've all seen the disparity between the marketing claims of performance, airflow, noise etc for fans, and their real-world performance in testing.
How many other devices do we buy, taking for granted what the manufacturers tell us about them without ever having the tools or know-how to test their claims?

It's a risky line of thinking, leading to tinfoil hats and wild theories, but it's still nice to see something like this out there. :)

This. It's just a neat gadget that lets nerds like us waste our spare time on disproving manufacturer's "perfect conditions" specs. Yeah, that 30-mile radio will only work for 30 miles between two tall mountains on a sunny day. And your 54-mbps 802.11g connection can never get 54mbps.
Fizzban 28th May 2010, 17:07 Quote
I guess it's nice to know just how long you can expect a SSD to last. But the only real use for it is enthusiast websites and magazines. Not many of us can afford to burn out hardware just to see how long its life cycle would be.

Reminds me of the Tom's Hardware videos years ago, where they would take the heatsinks off of CPU's to see if they shut down or fried. Great for lab tests to inform but not for the general consumer to use.
Elledan 28th May 2010, 17:10 Quote
Quote:
Originally Posted by Flibblebot
Surely you could do the same thing for SSDs in software? You wouldn't need any additional circuitry - just plug the SSD into a PC and let the software do its thing?

Yes, you can do this :) The device in question is meant to test the individual Flash ICs, without a controller chip and such getting in the way.

I have done the math on testing Flash-based SSDs in a configuration like you suggested, and it came down to that for old-school 10k cycles MLC drives with 4k writes it'd take less than a few days to kill a drive. With current ~3k cycles MLC drives it's got to be even worse :)
HourBeforeDawn 28th May 2010, 17:57 Quote
Quote:
Originally Posted by RedFlames
Quote:
Originally Posted by HourBeforeDawn
this could be nice for real reviewers who have the readers truly in mind and go the distance in showing their readers what you really get.

except it's be hard to produce accurate numbers as each drive would be different... one drive may last twice the manufacturer's quoted number of read/writes, another may keel over after only half the quoted number... it's like overclocking your mileage may vary, just because one chip will hit 4GHz doesn't mean they all will...

true you probably would need at least 3 drives to run such a test.
Flibblebot 28th May 2010, 18:05 Quote
Quote:
Originally Posted by Elledan
Yes, you can do this :) The device in question is meant to test the individual Flash ICs, without a controller chip and such getting in the way.
Yes, but what's the point of testing individual ICs in isolation? :p
Elledan 28th May 2010, 18:32 Quote
Quote:
Originally Posted by Flibblebot
Yes, but what's the point of testing individual ICs in isolation? :p

To verify the claims on the manufacturer's datasheet for the IC?

See for example: http://nvsl.ucsd.edu/ftest.html

In the published article they test Flash ICs from 5 different manufacturers. The error rate with MLC is a lot higher than with SLC, data corruption between adjoining cells is significant with MLC, less so with SLC. MLC also has random latency spikes while reading whereas SLC doesn't have this.

It's all stuff you often won't find in those datasheets.
docodine 29th May 2010, 00:05 Quote
Did something similar (unintentionally) on a TI-83 calculator.. I wrote a 'benchmarking' program which just filled in every variable with random huge numbers and then erased them, and the point was to see how fast it could do it like 10,000 times or something. I messed up the program and it just ran endlessly, and the calculator died heroically in about ten minutes.

Called TI and the technician was the first person to tell me about the lifespan of memory chips :-\

Sorry about the offtopicness, just reminded me of one of the stupid things I did in middle school...
Elledan 29th May 2010, 08:14 Quote
Quote:
Originally Posted by docodine
Did something similar (unintentionally) on a TI-83 calculator.. I wrote a 'benchmarking' program which just filled in every variable with random huge numbers and then erased them, and the point was to see how fast it could do it like 10,000 times or something. I messed up the program and it just ran endlessly, and the calculator died heroically in about ten minutes.

Called TI and the technician was the first person to tell me about the lifespan of memory chips :-\

Sorry about the offtopicness, just reminded me of one of the stupid things I did in middle school...

It can't have been SRAM/DRAM, then. Sounds more like you wrote to the storage (EEPROM or such), which does have a very limited lifespan in comparison to the RAM.
enciem 29th May 2010, 08:29 Quote
Quote:
Originally Posted by HourBeforeDawn
Quote:
Originally Posted by RedFlames
Quote:
Originally Posted by HourBeforeDawn
this could be nice for real reviewers who have the readers truly in mind and go the distance in showing their readers what you really get.

except it's be hard to produce accurate numbers as each drive would be different... one drive may last twice the manufacturer's quoted number of read/writes, another may keel over after only half the quoted number... it's like overclocking your mileage may vary, just because one chip will hit 4GHz doesn't mean they all will...

true you probably would need at least 3 drives to run such a test.

You'd need at least 75 drives for a statistically relevant sample. But enough of that, how about ordering some drives from scan and then returning them afterwards. I love destroying stuff.
thewelshbrummie 30th May 2010, 01:35 Quote
This makes sense for my current employer - I work for a TV repair company we're seeing more problems with EEPROM chips in TVs losing data or corrupting - this would really help us out as we're resetting/replacing a fair number of them
rickysio 30th May 2010, 04:50 Quote
Quote:
Originally Posted by thewelshbrummie
This makes sense for my current employer - I work for a TV repair company we're seeing more problems with EEPROM chips in TVs losing data or corrupting - this would really help us out as we're resetting/replacing a fair number of them

Destroying EEPROM? >.>

Of course, you could test EEPROM's from manufacturers to see who has the highest average survival time, but that'd take quite a while...
deadlyavenger 30th May 2010, 20:53 Quote
I'm tempted to make one as a general gadget/art piece. Have it sitting on your desk and when people ask what it is...you can explain it to them. Especially since you can change the time of the write cycles.

Chances are...they'll already know what a massive nerd you are and let you get on with it.
rickysio 1st June 2010, 12:14 Quote
Quote:
Originally Posted by deadlyavenger
I'm tempted to make one as a general gadget/art piece. Have it sitting on your desk and when people ask what it is...you can explain it to them. Especially since you can change the time of the write cycles.

Chances are...they'll already know what a massive nerd you are and let you get on with it.

Program it to display the time instead, with a quartz oscillator writing to it every second. When the time stops instead of the "Oh, battery's flat.", the "Oh, EEPROM's dead." should be interesting.
oceanwaves7 7th July 2010, 01:32 Quote
Ok, I've come to a conclusion as to why no one is testing a real SSD (or it's chips) with this method. Because with a modern (2008-2010) SSD, you wouldn't be able to kill the drive before you were very old. Simply writing at 80 MB/sec 24 hours a day, to a 64GB SSD would last 51 years assuming typical life expectancy roughly 2 million cycles, average of 2 to 5 million write cycles. Also the bigger the SSD, the longer it would last at the same data throughput. So writing to one at 160MB/sec that was 128MB would last the same amount of time. But at 80, again it would double it's life to 102 years.

While you could test individual chips, you will never kill an SSD to any reasonable level even after years and years of trying. And even if that rate were lower, the drive still would take many years to destroy. But then again, who even writes continually to their drive anyway? In that case you can pretty much assume you won't be wearing it out. An interesting read on the myths of SSD life expectancy:

http://www.storagesearch.com/ssdmyths-endurance.html
Elledan 9th July 2010, 08:27 Quote
It's very much possible to kill an SSD by writing something like two 4kB writes/second to it. Shouldn't take more than a month or so.
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums