A reason to stick with SATA SSDs, rather than PCI-E?

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Glaring_Mistake

Senior member
Mar 2, 2015
310
117
116
Wait, it's bad for SSDs to be left unused for a long time? Like for data integrity or the drive itself? First time I hear of this.
Yes, though it varies from to drive to drive how noticeable voltage drift is.
It is primarily bad for the data on the drive but there have been occasions where the drive has stopped working entirely, though this has been with drives that have seen a great deal of writes.

The thing is, and I believe that this is mentioned in the article, JEDEC specs for P/E cycles, basically state that the SSD should retain its information at room temp for a minimum of a YEAR, ONCE rated P/E cycles were EXHAUSTED.

Yes, TLC sucks, and I guess I didn't realize how badly, but even still, this was not (IIRC) a heavily-used SSD, and shouldn't have been at the limit of its P/E cycles, and thus should have retained its information for over a year.
Actually I believe that if you go by JEDEC specs then if a drive has seen little wear then its retention should be about ten times better compared to one that has used up all of its rated P/E cycles.
So leaving it unpowered for eight months to a year should not be much cause for concern except for in extreme conditions.

Interesting, according to Johnny Lucky's website and this one that SSD has Toshiba TLC. (This is the first time I have heard of that NAND doing this)
Yes, in my experience Toshiba's TLC NAND does not leak as much as some other TLC NAND.
So I tried to find information from other sources about what NAND it uses - looks like it may actually be SK Hynix 16nm TLC NAND, see: https://www.overclockers.ru/lab/76878/obzor-i-testirovanie-ssd-nakopitelya-amd-radeon-r3-120-gbajt-r3sl120g-lish-ten-bylogo.html and https://community.amd.com/message/2724332

That makes a bit more sense to me since according to my tests SK Hynix 16nm TLC NAND leaks more than Toshiba 15nm TLC NAND.
Here you can see how much a drive using SK Hynix 16nm TLC NAND has slowed down compared to one using Toshiba 15nm TLC NAND (both using the same controller):





I think we can both agree on that the latter drive (the one using Toshiba 15nm TLC NAND) has not slowed down quite as much as the first which uses SK Hynix 16nm TLC NAND.
 
  • Like
Reactions: cbn

Ranulf

Golden Member
Jul 18, 2001
1,758
250
126
This reminds me that I have an 840 evo sitting around with a backup image stored on it. Its been at least a year since I started that experiment. I originally meant to test it after about 6 months. Oh well.
 

Glaring_Mistake

Senior member
Mar 2, 2015
310
117
116
This reminds me that I have an 840 evo sitting around with a backup image stored on it. Its been at least a year since I started that experiment. I originally meant to test it after about 6 months. Oh well.
I wouldn't mind hearing how that turns out.
I think there's still a good chance that it will manage without corruption but with pretty low read speeds of course - especially if it is using the latest firmware.
 

mxnerd

Diamond Member
Jul 6, 2007
6,392
974
126
This reminds me that I have an 840 evo sitting around with a backup image stored on it. Its been at least a year since I started that experiment. I originally meant to test it after about 6 months. Oh well.
So can you still retrieve that backup image on that pulled 840 evo SSD after one year ?
 

FelixDeCat

Lifer
Aug 4, 2000
27,370
873
126
I am hoping this turns out to be untrue. I have a old MLC Samsung 830 from 2012 used for backups. :eek:
 

ashetos

Senior member
Jul 23, 2013
254
14
76
If you boot into linux you can issue discard/trim/unmap commands directly to the block device if it's an NVM-E device without having to deal with partitioning/formatting like this:
# blkdiscard /dev/nvme0n1

This will trim all of the address space of the NVM-E SSD which in theory will make the controller ignore all checksum errors and new PROGRAM/ERASE cycles would succeed without any speed degradation.
 

Glaring_Mistake

Senior member
Mar 2, 2015
310
117
116
I am hoping this turns out to be untrue. I have a old MLC Samsung 830 from 2012 used for backups. :eek:
What exactly is it you're hoping is untrue?
Because that NAND suffers from voltage drift is pretty well established.
Now the Samsung 830 should be more resistant to voltage drift than the drive OP had issues with and even both drives that I posted results from in a previous post.

According to one study however SSDs become more likely to develop Uncorrectable Errors as they age which could be a concern given the age of that 830.
But it was not clarified why they suffered from data corruption and of course there may be differences between the custom corporate PCI-E drives in the study and your standard consumer drive.

I think it could work well as backup however if you create some hash files at the same time and check files for corruption every now and then.
 

Ranulf

Golden Member
Jul 18, 2001
1,758
250
126
I wouldn't mind hearing how that turns out.
I think there's still a good chance that it will manage without corruption but with pretty low read speeds of course - especially if it is using the latest firmware.
I think I left it un updated firmware wise at the time.

So can you still retrieve that backup image on that pulled 840 evo SSD after one year ?
I'll try to get to it later this week.
 

ASK THE COMMUNITY