How is it possible that TLC gets more endurance than MLC

kensiko

Limp Gawd
Joined
Oct 5, 2011
Messages
136
2 weeks ago I discovered that my trusty 970 Pro with MLC NAND had lower endurance rating then a MP510 with the same capacity. How is this possible ? Was there any test done to prove that ? Because on the MP510 once the NAND is filled it will start to reorganize the data and this causes wear.

https://www.tomshardware.com/reviews/corsair-force-mp510-ssd,5848.html

ProductTBWDWPDWarranty
Force MP510 960GB1,7000.935
MyDigitalSSD BPX Pro 960GB1,6650.915
Samsung 970 PRO 1TB1,2000.665
Adata XPG GAMMIX S116400.355
Samsung 860/970 EVO 1TB6000.335
WD Black 1TB6000.335
Intel SSD 760p 1TB5760.325
WD Blue 3D 1TB4000.373
Crucial MX500 1TB3600.25
Intel SSD 660p 1TB2000.115
 
TLC drives usually have a large SLC and/or DRAM cache and newer SSD controllers have become very good at utilizing that cache and evening out wear on the TLC.

With that said, these endurance figures are borderline arbitrary, and likely more defined by warranty concerns than the technical limitations of the drive. A drive can easily exceed it's endurance rating.

If you fill your drive, that will result in negative side effects with almost any SSD, so don't fill the drive...
 
i use them 128gb ssds from samsung 840-850 pros for caching in my systems with primocache. they have helped immensely when paring with even 8tb -12 tb 14 tb 5400 rpm hamr drives.
 
2nd'ing rice's statement, frankly its about how much the manufacturer wants to garuntee/warranty not so much its true endurance.
Drives can be worn much past it's rated endurance or fail to write prematurely, at the end of the day its similar to the MTBF for HDD.
Its not a hard stop, but they have an increased likely hood of failing/unable to write further to the drive once it's reached that amount of writing.
 
2nd'ing rice's statement, frankly its about how much the manufacturer wants to garuntee/warranty not so much its true endurance.
Drives can be worn much past it's rated endurance or fail to write prematurely, at the end of the day its similar to the MTBF for HDD.
Its not a hard stop, but they have an increased likely hood of failing/unable to write further to the drive once it's reached that amount of writing.
I agree. People freak out over SSD endurance too damn much. Lots believe once you get close the the rated writes the drive is dead.
 
I have been using SSD's since 2008 and have not had a single endurance issue. I have only had two SSD's go bad all this time. Both of those were because of the controller (Sandforce) going bad and not the memory chips. The endurance ratings kind of remind me of short expiration dates on canned goods. They have them to cover themselves.
 
I have been using SSD's since 2008 and have not had a single endurance issue. I have only had two SSD's go bad all this time. Both of those were because of the controller (Sandforce) going bad and not the memory chips. The endurance ratings kind of remind me of short expiration dates on canned goods. They have them to cover themselves.
Same here, really! I remember paying 350 CAD$ for a 64GB SSD OCZ core! Yes the ones slowing down with random writes
 
Damn times have changed ATM Amazon has Samsung 850 Pro 128gb for $304 crazy 3rd party sellers.
 
99% of that is marketing/warranty. See the old 840 Pro tests where they wrote a full petabyte to the drive before it started failing...

I did some quick looking, as I have a background in NAND storage, but I can't easily pull up the wear life specs on the raw samsung vNAND chips to figure out their actual wear life. It used to be a lot easier to get at that data =P
 
I have been using SSD's since 2008 and have not had a single endurance issue. I have only had two SSD's go bad all this time. Both of those were because of the controller (Sandforce) going bad and not the memory chips. The endurance ratings kind of remind me of short expiration dates on canned goods. They have them to cover themselves.

I have bought many many SSDs and curiously the only ones that ever show any wear are the Samsungs.
 
There is different binnings within each Nand type e.g. there is good TLC nand and there is junk TLC nand.
Also there is the size of the manufacturing node (smaller means less endurance and performance).
Also 3d nand vs planar nand.
Then there is dram vs dramless, which affects endurance of mapping tables.
Finally you have the controller which by itself has an impact as it effects how well things like wear levelling work.

Initially been loyal to samsung I am actively now trying to avoid them when buying new ssd's, 970 evo's seem to have a high rate of problem reports, and I feel they are been over agressive in topping performance charts. Drives with larger SLC caches e.g. I suspect should also have higher endurance. Not to mention the 840 fiasco.

The endurance tests run by a couple of reviewers were good, but bear in mind they not mimicking endurance during power off, as well as general lifetime decay. Also they were constantly refreshing data, which didnt emulate what happens if you write rarely and read often. (which is the failcase for the 840).
 
Well I guess we'll never have a clear answer, but I like those specs and I've purchased a M510 4TB. Would have loved a M600 not qlc but no 4TB version available
 
I am waiting for the 4TB NVMe TLC drives to hit around the $250 to $300 mark before replacing the 3.2TB io scale drive I have.
 
Endurance ratings are not hard limits when the flash chips stop working. Usually when you reach specified TBW it means that the data retention time (time during which the SSD keeps your data while unpowered, while still meeting specified uncorrectable error rates) dropped to the rated amount.

If the manufacturer only guarantees a data retention time of three months like is common with TLC/QLC SSDs, then they can of course specify much higher TBW than if they specify a year which was often the case with MLC.
 
2 weeks ago I discovered that my trusty 970 Pro with MLC NAND had lower endurance rating then a MP510 with the same capacity. How is this possible ? Was there any test done to prove that ? Because on the MP510 once the NAND is filled it will start to reorganize the data and this causes wear.
Usually SLC > MLC > TLC > QLC on endurance, but there are other variables, and one of them is the internal overprovision of the drive.
Micron 5300 MAX 2TB (TLC) has a endurance of 17,520 TBW
Samsung 860 PRO 4TB (MLC) has an endurance of 4,800 TBW

So a 2TB TLC drive has 4x the endurance of a 4TB MLC, but its a matter of how they design the drives not only the type of NAND.
 
Still have one of the OCZ V3 128Gbs pulling daily duty with no issues. Decade or more old now? It's been pretty full most of the time too lol, some SSDs are much more redundant and not plagued by the 'platter dust of death destroying your platter surface' issue that HDDs often had when failing.
 
I thought Samsung gave up on Multi-Level Cell being 2 like many people want and they have 128L 3D TLC (three, tri, etc.)? Sites like Anandtech and Servethehome say it is 3-bit.

But are QLC drives failing? They have been out for years.
 
Last edited:
Back
Top