Need Help? Contact the Espiya Helpdesk. CLICK HERE


Author Topic: SSD stress test  (Read 1476 times)

kenji_kulet

  • 2008 Guardians
  • Active - Top Level
  • *
  • Posts: 5529
  • Karma 461
  • Gender: Male
  • Special Skills: kalabit penge gang , taga jumper ng kuryente , sipsip gas at mang-buriki ng bigas
SSD stress test
« on: August 25, 2013, 10:33:09 pm »
the guys from Hardware.info did a stress test to Samsung 840 SSDs to see how long will it take for your SSD to break down

here is the final conclusion of the test , there are other test on the website i'll post the link below

but here is the final update

http://us.hardware.info/reviews/4178/10/hardwareinfo-tests-lifespan-of-samsung-ssd-840-250gb-tlc-ssd-updated-with-final-conclusion-final-update-20-6-2013


Final update (20-6-2013)

Two and a half months ago we started the endurance test of two Samsung 840 SSDs to find out whether an SSD with TLC memory chips has a shorter lifespan, and after that period of time did one of them fail. The second one still works, but it's running on fumes. The is no doubt about it, there is nothing wrong with SSDs with TLC memory chips.

Our previous calculations of the lifespan of TLC-based SSDs were clearly off the mark. We made the erroneous assumption that static data remains in the same location on an SSD. We had put 160 GB of static data on the SSDs. Considering how much data was written to the SSDs in two and a half months, the static data was moved by the controller to other memory cells at regular intervals in order to keep an even load on all cells within the SSDs. That means that a nearly full SSD performs more operations in the background in order to move that static data around. However, that does not have a negative impact on the lifespan of the SSD. There's no difference between an almost full SSD and an almost empty SSD in that regard.

The first SSD, the one you could follow live on the stream, had its first re-allocated sector on May 22, after 2,945 program/erase cycles and 707 TiB of written data according to the S.M.A.R.T. data. That's quite remarkable, as Samsung guarantees a lifespan of 1,000 cycles, while in reality their SSD last three times longer.

A re-allocated sector means a sector that no longer works as it should, and that the SSD avoids it and instead uses reserve memory. As long as no uncorrectable errors have occurred no data has been lost, but it should serve as a wake-up call to make sure that you are making enough back-ups. The first uncorrectable error came four days later on May 26, after 3,187 cycles and 764 TiB of written data. This is the first time data was irretrievably lost, and we consider it to be the end of the line for that SSD, even if it was just 512 kB. It wasn't until June 7, after 3,706 cycles and 888 TiB of written data, that the first SSD died completely.

During the test we received many requests from readers to also test the data retention. It's a valid point, as it could very well be the case that after many write cycles, the memory cells could lose their data when they don't receive power. To find out, we took a second system and turned it off half a day each week after about 3,000 cycles. We even turned it off an entire weekend. It didn't run into any problems, so the retention part seems to be in order.

The second SSD had its first re-allocated sector on May 28 after 3,152 cycles and 745 TiB of written data. The first uncorrectable error occurred on May 30 after 3,247 cycles and 768 TiB of written data. It has completely died yet, and it's passed 4,000 cycles now, but it's also accumulated more than 58,000 uncorrectable errors. We'd call it clinically dead.

So let's calculate the lifespan. We consider the first uncorrectable error to be the end of the SSD. For the first one that was after 764 TiB of written data, and for the second one that was after 768 TiB of written data. That the two failed so close together could indicate that the lifespan of the Samsung 840 SSDs is fairly constant. However, two SSDs aren't representative of the thousands that are out there.

If we take the 764 TiB and an average of 10 GiB of writes per day, we arrive at a lifespan of 214 years. Keep in mind that we sequentially write and fill the SSD which gives us write amplification factor of only 1.04 or 1.05. That's the difference between the write commands sent to the SSD and the writes executed by the SSD internally. The general assumption is a WAF of around 3.0 for normal consumer use with SSDs that don't employ compression tricks. That translates to a lifespan of 75 years. Even when you push an SSD to the max by downloading lots of movies everyday up to an average of 30 GiB per day, the SSD will still last you 24 years. Nomatter how you use it, it will last longer than the period you plan on using it (most people like to add more capacity after a certain time). A functional lifespan of 10 years is already unusually high for a storage medium.

With an average lifespan of 75 years for the TLC memory chips, consumers have absolutely nothing to worry about. It doesn't mean the SSD will actually last 75 years, but the number of available write cycles will not be the bottleneck. That means we will amend our conclusion from a couple months ago. A Samsung 840 SSD with TLC memory is just as reliable as SSDs with MLC memory, and the type of memory should not be a reason to choose one SSD over another.