Hot - Samsung 850 Pro Endurance Test Results

Johnny Lucky

Member
Apr 14, 2012
92
14
71
www.johnnylucky.org
Hilbert Hagedoorn, "The Guru of 3D" just published a news report about a 3 year long ssd endurance test performed by German Magazine C'T. A Samsung 850 Pro bested other brands with grand total of 9,100 TB of data written. That's 9.1 Petabytes or about 9,100,000 Gigabytes written. Samsung's official endurance estimate was only 150 TBW!!!

I should mention that there were six different brands tested over a 3 year period and every ssd exceeded their manufacturer's conservative endurance estimates.

Here is a link to the news article:

http://www.guru3d.com/news-story/en...o-comes-to-an-end-after-9100tb-of-writes.html

The results were actually published two months ago. Unfortunately C'T charges 1,49 Euros to download the German Language test report. Here is the link to their article:

https://www.heise.de/ct/ausgabe/2017-10-Billig-SSDs-im-Langzeittest-3694404.html

GEE! I wonder how long my Samsung 950 Pro's will last?
 

Glaring_Mistake

Senior member
Mar 2, 2015
310
117
116
Hilbert Hagedoorn, "The Guru of 3D" just published a news report about a 3 year long ssd endurance test performed by German Magazine C'T. A Samsung 850 Pro bested other brands with grand total of 9,100 TB of data written. That's 9.1 Petabytes or about 9,100,000 Gigabytes written. Samsung's official endurance estimate was only 150 TBW!!!

I should mention that there were six different brands tested over a 3 year period and every ssd exceeded their manufacturer's conservative endurance estimates.

Here is a link to the news article:

http://www.guru3d.com/news-story/en...o-comes-to-an-end-after-9100tb-of-writes.html

The results were actually published two months ago. Unfortunately C'T charges 1,49 Euros to download the German Language test report. Here is the link to their article:

https://www.heise.de/ct/ausgabe/2017-10-Billig-SSDs-im-Langzeittest-3694404.html

GEE! I wonder how long my Samsung 950 Pro's will last?


This test is actually a bit interesting - not that it surprises me to see the 850 Pro as the winner or that some drives can write more than they are warranted for before expiring.
What I find interesting is that the BX200 did not exactly impress seeing as it only wrote 187TB (or 280TB, the article is unclear on that) before it stopped working.
Now both of those are well above the TBW of 72TB but the BX200 is rated at 1000P/E (which all drives using the SM2256 as a controller seem to be) and both 187TB and 280TB may be pretty close to that (unless WA was a bit high).

The other drive to go first I could see being the 750 EVO or SanDisk Ultra II given that both are rated at about 500P/E (if that hasn't changed).

A pity that some of the drives died from a power surge/peak though, would have liked to see how the Trion 150 would have done.

Samsung's official endurance estimate was only 150 TBW!!!

Not sure if that's really an endurance estimate, TBW often tends to be rather what the manufacturer wants to cover under their warranty.
For example the Kingston UV400 is rated at around 400-500P/E and their 250GB version has a TBW of 100TB while the 850 Pro is rated at around 6000P/E and their 256GB version (like the one that seems to have been used in that test) has a TBW of 150TB.
So the UV400 has just a twelfth of the endurance the 850 Pro has when comparing P/E (given 500 P/E for the UV400 and 6000P/E for the 850 Pro) but when comparing TBW you would think it had two thirds of the 850 Pro's endurance (when comparing these capacities).

Samsung have also already admitted that the TBW for the 850 Pro is not really due to (comparatively) low endurance but instead was set lower due to market segmentation, quoting Anandtechs review of the 850 Pro: "Moreover, Samsung told me that the endurance figure is mainly meant to separate the 850 Pro from the enterprise drives to guide enterprise clients to the more appropriate (and expensive) drives as the 850 Pro does not have power loss protection or end-to-end data protection for example. However, I was told that the warranty is not automatically denied if 150TB is reached under a client workload."

Also something in that review that you might find of interest is that Samsung claimed to have an 850 Pro at 128GB still running after 8PB, an even more impressive figure than 9PB (since it seems that was with a 256GB drive in that test).


GEE! I wonder how long my Samsung 950 Pro's will last?

Not that I think it likely to be much of a factor for you even should there be a difference in P/E between the 850 Pro and the 950 Pro but the same NAND in different units may be rated at different P/E.
The 3D TLC NAND in the 850 EVO is rated at 2000 P/E (or at least it used to be when they first tested the endurance) but the 3D TLC NAND in the 960 EVO is rated at about half that at 1000P/E.
 

Elixer

Lifer
May 7, 2002
10,376
762
126
This isn't really news though, we have known that write endurance isn't really the problem these days.
It is everything else that is the issue.
I got a few SSDs that died from no obvious reason, and they were not even close to 1 TBW, so, yeah, the very last thing that is mentioned in the "article" holds true.
...always backup your data.
 

Glaring_Mistake

Senior member
Mar 2, 2015
310
117
116
Off topic but Johnny Lucky I've seen some things you might want to update in your SSD database:
MX200 is shown to come in capacities of 120GB and 1TB in the M.2 section but they never made those capacities for M.2 (and the 120GB not at all).
Also it is missing from the SATA 3 section, though that may have been because you removed some drives from the database seeing as the MX100 and BX100 are also missing.

MX300 does not have an M.2 2TB version, it just goes up to 1TB.

OCZ RD400 has a maximum TBW of 592TB, not 296TB.

Mushkin Reactor uses the controller SM2246EN, not SM2246FN.

850 Pro has a TBW of 300TB from 512GB to 2TB.

SanDisk Ultra II has gone over to using 15nm TLC NAND.

Toshiba A100 has a TBW of 30-60TB and the same goes for OCZ TL100.

Toshiba Q300 has a maximum TBW of 240TB, not 340TB.

The Intel 600p has an updated TBW that goes from 72TB to 576TB.
Also might consider updating the name of its controller to SM2260.
 

Johnny Lucky

Member
Apr 14, 2012
92
14
71
www.johnnylucky.org
Glaring Mistake - A great big thank you for the information you provided.

I just corrected all endurance discrepancies by deleting that column from the database. There were too many variables for the information to be accurate. The first problem I ran into were that some companies did not bother to publish any information. In addition not all companies estimated endurance the same way. Some publish TBW, some publish MTBF, and some publish GB written per day. The problem is further complicated due to different capacities having different ratings. Worse still, the information in early technical reviews sometimes doesn't match the data when an ssd is finally released for sale. However, it was the three long term endurance tests that made the decision to delete the info an easy one. All three long term tests suggest that company endurance estimates are extremely conservative.

I'll check and make the rest of the corrections tomorrow when I do another update.
 

bononos

Diamond Member
Aug 21, 2011
3,894
162
106
Whats the read/write speed before it bites the dust, thats going to be the real problem before the drives are totally dead. The performance of planar tlc drives are going to crash hard well before their rated endurance and its going to be a pain using them even while they have plenty of life left.
 

Glaring_Mistake

Senior member
Mar 2, 2015
310
117
116
Glaring Mistake - A great big thank you for the information you provided.

I just corrected all endurance discrepancies by deleting that column from the database. There were too many variables for the information to be accurate. The first problem I ran into were that some companies did not bother to publish any information. In addition not all companies estimated endurance the same way. Some publish TBW, some publish MTBF, and some publish GB written per day. The problem is further complicated due to different capacities having different ratings. Worse still, the information in early technical reviews sometimes doesn't match the data when an ssd is finally released for sale. However, it was the three long term endurance tests that made the decision to delete the info an easy one. All three long term tests suggest that company endurance estimates are extremely conservative.

I'll check and make the rest of the corrections tomorrow when I do another update.

Well, the TBW can be a bit conservative (it varies a lot) and often seems to be set by exactly what they want their warranty to cover rather than representative of the actual endurance.
I think that at least a few drives have a TBW representative of that but then we're talking under specific workloads, if we're talking about sequential writes with plenty of time for it to recover a drive may not see as much wear as it would if it was constantly bombarded by random 4k writes.

However those tests are just about how long a drive can last while being asked to just write and write and not retention which is what the rated P/E cycles are.

If a drive is rated at a certain amount of P/E cycles (say 2000 like the 850 EVO has) it does not mean it will stop working after using up those 2000 P/E but instead it is just telling you: "Stop here and I will reliably store data powered down for one year (under specific conditions)."

Something that Nordichardware has experienced with both the Adata SP550 and the 850 EVO.

After using up around 4000 P/E for the 850 EVO (so twice the number of P/Es it is rated for) they left it powered down for two months, after which it developed a lot of read errors and then died.
With the Adata SP550 they also used up around 4000 P/E (that's however four times the number of P/Es it is rated for) but left it powered down for just a week.
After which it was no longer detected by the computer (okay, strictly speaking they say Windows so it's possible it showed up in BIOS).

So you can write a lot more to a drive than it is specified for but not without retention suffering severely.


I can see why that section would be a pain to maintain though.



Some more corrections:
The Intel 545s has 128GB as their smallest capacity, not 126GB.
In the same panel TweakTown has been misspelled.

The Intel 540s has SM2258G as its controller (and it comes in the M.2 form factor too).

The Kingston UV300 uses TLC NAND, not MLC NAND

And finally, the Integral UltimaPro X uses MLC NAND, not TLC NAND.