SSD Endurance - the Big Lie

I see. Well I doubt a company will wait years to extract the data. I meant more as a permanent storage when both SSD/Mechanical hard drive are working. If few writes are done, both will last a long time however the SSD will last far longer in those situations.

Platter exchanges are quite difficult as platters are incredibly sensitive. That's the type of service that reels in the few thousand dollars due to the clean room requirement. SSD would reel in the few hundred dollars if that. Anyways most companies are smart and they use massive raid arrays in secure servers using enterprise hard drives which are worlds apart from standard desktop hard drives.

Apple vs PC is irrelevant to this topic, I agree. Please keep it out of this thread :|

 

OP, it's kind of hard not to do the whole HDD vs SSD. They are competing technologies that one way or another fill the same role.

As for this so called maximum number of write cycles I gotta tell you I think it's a mute point. I've never read of anyone who has had this happen and I have had two SSD's for almost 5 years now. I got my Samsung 32GB units back when they were $500. a piece and the only thing I ever do is turn off defragmentantion in Windows.

I had them in Raid 0 for the first year with XP. Then I began to use them as boot drives in two different laptops. They have seen many windows installs from XP to Vista to Linux to W7. Most of the time I didn't do anything to the drive but have also reformatted prior to an install.

My boot times have not increased over time nor has anything degraded.

IMO, all of this SSD end of life stuff may make for an interesting technical discussion, I've yet to see it actually happen.

I have only seen and taken advantage of the great performance gains that SSD's provide.

I can only think of what a shame it would be for someone considering purchasing an SSD, to decide against it based on a theoretical end of life scenerio which has yet to happen.

 

It's actually an interesting point that you raise there. A lot of the older SSDs are actually more durable than the new ones. Your Samsung SSDs are almost certainly built with a larger process, which means that they inherently have much more cycles. The 50nm NAND usually has about 10k write/erase cycles, the 34nm now has about 5k, and it is rumored that the upcoming 20nm NAND may only have 1.5-2k.

 

Could the shrinking of the NAND be due in part because we haven't seen SSD's reaching their write cycle limits and therefore the greater amounts of NAND are in fact not necessary?

 

At work we sell data recovery; albeit very expensive.

For logical failure of a "hard drive" is 700 dollars. That is it if still spills up properly but not being recognized, usually zapped partition tables, or bad MBR or a myriad of other things. If you have RAID 0 usually you are screwed.

For physically damaged "hard drives" is it 1850. This a clean room recovery, they completely disassemble the drive and attempt to recovery data.

 

Just because something doesn't have moving parts doesn't make it more reliable than something that does. Yes there is higher probably of damage, which would fall under durability but not reliability. Something can fail for more than just mechanical reasons.

Either way this is all a bunch of theoretical hoopla. But this goes back to a point that I've made time and time again that the OEM's for these drives need to divulge this information in a way that's easy for the average consumer to digest and understand so we don't have to make all these assumptions. I agree with the OP that there's more to it than just simple total size of writes, and I'm guilty brash assumptions and spouting of information that over simplifies the issue. But it's still a lot of unknowns.

After reading this article posted by Fishon earlier, it makes me wonder if the designers of these drives even have a clue.

principles of bad block management in flash SSDs - article StorageSearch.com

Although it's kind of ironic that smaller fab process results in less life. Would be kinda funny if they had to retool to old larger fab processes, although seems like that would make them cheap. Or not. I dunno.

 

Not bad for prices, but here's one question (more of a statement from what I have seen).
Is your price for physically damaged mechanical hard drives a flat-rate fee of $1850. Or can the price increase?
Also what if it's an enterprise hard drive ?.. I can assure you it's no measily $1,850.

 

Are you talking about actual like SAS 600 GB drives or like WD Caviar Black enterprise edition?

And the service is not guaranteed. If unsuccessful your money will be refunded. Also that is a contract with DataTech Labs offered my Micro Center, I don't know if you'll get the same rate if you visit their website.

 

SAS drives, from my understanding they're a bit more difficult to work with because they may include hardware security, and sometimes hard to find thus-expensive parts, correct me if I am wrong my researched stopped at regular desktop hard drives..

I see. But is the $1850 a flat rate price regardless of the repair?
I know most places do similar fees, however it can vary drive to drive repair to repair.
Yes I am well aware no service is guaranteed in recovery of a non working hard drive, I have mentioned that already. Although most places refund most of the money, and keep a little bit I believe.

 

Okay, even if that's true, the argument still, is they are going to reach a write cycle limit even though it hasn't happened yet?

The point is, everything has an end of life cycle. How old is the average users cell phone? For that matter, how old is the average users laptop?

"SSD Endurance-the Big Lie"........If it could be demonstrated with evidence of short lived use due to maxing out the write cycles I would give that statement credence. Until/if we ever see this happening, I can't get on board with this notion because my own experience and the lack of examples of failures due to limited write cycles demonstrates otherwise.

 

endurance is not an issue for me. i remember the old hdd's failing after 2 years or so. so really 4-5 years isnt bad. but just like a mentioned before, its the performance degradation that some people are reporting, that bothers me most

 

I haven't experienced any of the degradation I've been reading about. I wonder if it has anything to do with most of the drives being MLC rather than SLC like mine.

 

OP... This thread is going to get shot down by uneducated people with anecdotal references. Mostly because some spent hundreds of dollars on their SSD, thus they have a conflict of interest to hear that SSD aren't that durable.
Much like my air cooling explanation at overclock.net that hot air does not rise under room environments as well as desktop cases with or without fans. Got shot down pretty badly, even moderators wouldn't do anything even though I receive more then a few comments/private messages saying I was correct.

 

You can force Firefox to use system RAM as follows:
Moving Your Cache to RAM for Speed Boost | Firefox Facts

I have also set the disk-cache integer values to 0.

 

Ah, there you have it SLC are easily 10x more durable, maybe even moreso. I think how they're circumventing the lower life cycles of the cells is having larger drives. With a 32GB, it will see tons of writes across each cell. With a 256GB it will take a long time before each cell has reached its P/E limit unless you do tons of writes.

I think tiller has shown that doing lots of daily writes destroys performance rather quickly. But the average user won't see that degradation for a couple years. Then a secure erase will at least bring back some of the performance for a while longer. I personally never keep my laptops more than two years, but in my desktop, where I put an SSD in each one as an OS drive, I've had for over three years. Swapping out a $50 hard drive isn't too hard to swallow, but swapping out a $200 SSD is if it has to be done every couple of years.

 

I once used a computer nearly unchanged for 8 years. It was my main computer that was running every day. The end of that period was 4 years ago and the computer including its now 11 and 12 years old HDDs still works.

I can hardly see irony in that. What basically happens when a flash cell of an SSD degrades is that the usually insulating layer around the cell that turns the transient cell into durable memory corrodes due to the voltage application that is used to write data to the cell. On a certain point the corrosion is just to bad to change the semiconductors behavior again and to write new data.
If one now decreases the structures of the process, the insulating layer has to shrink too, which means that there is less tolerance for corrosion.

One thing that has not been mentioned here so far is, that SSDs usually have a certain amount of backup cells that you can't see. If one cell fails, another one from the backup pool will take over. This means that you won't notice any degrading until the backup pool is used up. It's the same idea like never filling an SSD completely.
The pro of that extra pool is that you'll have a fully functional SSD for a much longer time, the con is that the apparent degrading will go quite quickly once you start to notice it.

Yes it has, in two ways:
1. SLCs only have two states: 0 and 1, charged and discharged. MLCs contain multiple states like bit A charged (0) or discharged (1) and bit B charged (2) or discharged (3). If you read a charge and have to decide which data it stands for it's much easier if you only have to tell the difference between 0 and 1 instead of 0,1,2 or 3. The same goes of course for writing. So SLCs have a bigger tolerance.
2. Since a single MLC contains more data than a single SLC it will more often be accessed which also means it will be written more often.

 

So it is sounding like they may have to stop using MLC altogether and run with SLC which will hike the prices way back up again for the consumer end, lol. Back to 32GB drives for $300! Or give us huge capacities to offset the low life cycle life of the cells, again back up to $300+ range.

 

Yeah, Rob, IIRC today's SLC cell has about 100k cycles, which is like 20 times more than today's MLC. So an old SLC drive might have like...

 

I always wondered why they did not decide on SLC. Seemed like much more robust tech. This thread has a very scary ring of truth to it. I am not enough of a scientist to know for sure which side is correct. It is a very interesting thread.

I have my SSD simply for performance. But also reliability. But not necessarily longevity. Never had a hard drive more than 2 years before I wanted a newer faster one.

And I never really did believe the Read Only when it dies thing. Much as I would like to, sounds too good to be true.

But I bet it is more true with SLC than MLC.

Where oh where did Mtron go who had a nice line of SLC? Did they get a visit from Micron/Samsung/Intel telling them to stop making drives???

In the mean time, I am enjoying the heck out of my Corsair NOVA and looking forward to the next round of SSD's. SLC where are you???

 

Because people didn't want to pay SLC prices. I remember 1 year ago 60 GB OCZ SLC drive was 600 bucks. MLC drives keep selling because they are cheaper. Ain't much of a market when your SSD for the same price can be sold for 20% of the price.

 

Most people do not want to pay for a Ferrari either. And yet, there are still Ferrari's

 

The industry seems to be moving away from SLC. What is replacing it seems to be eMLC which is a tweaked MLC with more P/E cycle but much shorter retention life(intended to be used in 24/7 server).

As for reduced P/E cycle in 25nm process, it is compensated by much larger size as when we look at SSD, it is the total P/E cycle that is important.

 

Yes, but when people buy a 256GB SSD they expect to use the 256GB SSD, not a pseudo-128GB SSD. I'd be very unhappy if I was told that my 500GB HDD lasts half as long as my 250GB HDD unless I only use half of it.

So if the 25nm process produces cells with half the P/E cycles of the 3xnm SSDs, then they better doubly overprovision, something I doubt they will do.

 

Yeah, SLC vs. MLC is part of the answer and the other part is the much larger (and much higher quality) process node it was manufactured at that explains why a 5 year old drive is still going so strong.

I too would pay for Ferrari prices if I actually got a Ferrari.

Although MLC and eMLC tech along with ECC and smaller nand will give us an 'affordable' and 'capacious' SSD fairly soon, all the marketing spin in the world will never make me believe that this very fragile tech (at its core) will ever be more reliable than mechanical HDD's.

A HDD platter has a known, physical layout of the data it has recorded - it might take extreme measures to get to that data (electron microscope, clean room, etc.) but in almost all cases, if the data is important enough, it can be done.

An SSD on the other hand is hardly 'readable' like this - especially if the ECC which will soon be on the nand itself (the 'ClearNAND' article I posted earlier) is what fails.

The claims that 'SSD's will still be readable 10 years after the cell life cycles have been exhausted' is highly dubious and smells more like marketing than anything we can trust our data to. It may have applied to the original SSD's (circa 2008) that it was written about - but notice how quiet SSD manufacturers about making any recent/current claims?

They put 3 year, 5 year and 10 year (Patriot) warranties on the drives for little to no cost, and most people assume that the warranty shows how much faith the manufacturer has in the product. Nothing is further from the truth.

Like mentioned earlier, warranties are used by only a small part of the people who buy things - most will simply ditch them and get something newer/better and the only 'winner' is the manufacturer who got many, many more sales because of advertising the warranty, instead of simply offering a truly superior product.

Most consumers have very selective memory: they'll hear the part about the 10 year readability, they'll see the increased warranty periods, the increasing capacity of the drives and the cost of the drives plummeting - and they'll put it all together subconciously and think that today's drives are better, cheaper and more reliable. When they are obviously not.

Warranties do not make a product better: it is simply one more tool the marketers have to put pressure on consumers to buy now.

 

unfortunately, that is fact of life of SSD so anyone seeing a $/GB, one has to add the OP premium as well. I put it at 30%.

It is this OP requirement that is why I don't want small(just fit) SSD.

Though similar thing happens in HDD. My Scorpio Black 500GB is used as the main drive for only the first 150GB or so as I want the best performance. The rest are used as 'secondary' storage.

 

Me.


What's the big deal here? There is no SSD conspiracy...

No one held a gun to my head when I bought one either. If someone buys a bad product due to hype or dubious marketing information... that's their own fault. Not the industry's. If they're flat out lying that's one thing, but there's no proof of that here. Hyundai says their cars are just as luxurious as a benz... and they offer a 10yr warranty to boot! You gonna buy one just because the nice man at the store told you that?!

Seems like a crusade without cause here.

If someone is truly suspect of MLC SSD's... buy an SLC one.

If you bought an SSD for data reliability in terms of long term shelf storage... you're using the wrong medium.

 

tiller - manufacturer warranties almost never cover "normal wear and tear". It's specifically mentinoed in the warranty information in most cases. I have OCZ about this before, and was told that wearing out the NAND is just your own problem. The warranty covers things like the controller unexpectedly dying, or manufacturing defects.

 

3.94TB worth of writes in under a year and its running as strong as day one here.

 

No, but they also don't divulge information that is important to its use and longevity. Holding back information is almost worse than misinformation. To use your analogy, the nice man at the car dealership tells you his Hyundai is as nice as a Benz, but fails to tell you that there's only a three gallon gas tank in the car. There's features and performance expected to be there from a customer standpoint.

 

are you sure ? If their warranty is termed as '3 years or 7TB written', that may be the case. That is how car manufacturers say. Otherwise, I doubt it can withstand the legal challenge. 3 years are 3 years.

 

Yes, I am sure. A manufacturer offering a "7TB written" warranty can be potentially out a huge amount of money if a customer can destroy it at will whenever he wants a new SSD just by running some script that writes a single '1' to each block and then erases it.
The 7TB written thing is just a marketing thing, carries no significance.

Here's OCZ's warranty info:

Limitation of Warranty:

• Connection to a faulty power source
• Alteration, Modification, Disassembly or unauthorized repair
• Improper use of product
• Normal wear and tear
• User inflicted intentional or accidental damage
• Any other cause not resulting from defects in materials or workmanship.

 

I don't quite agree with you.

Nowadays SSD manufacturers adopt the notion of Write Amplification to take this special P/E characteristics of the NAND flash into consideration. In short, WA is the ratio between actual flash writes and actual data recorded.So the correct formula of life expectancy will be:

(Spare)SSD Size * P/E cycle = WA factor * Wear leveling inefficiency * Amount of Data can be written

With the current technology, P/E cycle is about 5K~10K and Wear leveling inefficiency is about 1.04. Intel X25-M has a WA factor of 1.1 and the latest SandForce products has a WA factor of 0.5. Early SSDs could have a WA factor of as bad as 5.

SSD size is a tricky part. Typically manufacturers use available blocks in user accessible spare space to write new data onto on a daily basis. A SSD with a spare space of 10% could have 400% more erasing than a SSD with a spare space of 50%.

But even blocks with infrequently updated data must be erased and reprogrammed over the time due to a property called "read disturb". Besides, reserved blocks are put into use gradually. Of course, the wear leveling algorithm will take care of everything.

Let's assume a user who have 20% of spare space on a 120GB SSD, with 5K P/E cycle and WA factor of 1.04. This means 108T of data can be written before the SSD expired.

On average, a typical Windows 7 notebook user writes 5GB of data daily. But considering nowadays SSDs are mostly accepted by hard core users, we double this number. Even so, this translates to 30 years before the P/E cycle is depleted.

Of course, this is with the lasted generation of controllers. With early implementations of a 5 WA factor, this life expectancy could be reduced to only 5-6 years. The situation is even worse on old OS which don't support the TRIM command.

Anyway, P/E cycle depletion comes far behind other possible malfunctions or technology replacements.

 

Poor analogy. All of the faults you mentioned about the car (brakes, gas tank) affect the performance of the car. SSD memory lifetime affects the longevity, not the day-to-day performance.

Again, this is theorycraft.

1) Does anybody have an actual confirmed case where an SSD has reached the end of its write lifecycle, and reverts to read-only mode?

2) How long does that take?

3) How many people actually keep their drives for that long? How many people replace their drives for performance or storage reasons within that time frame?

4) What evidence is there that a mechanical hard drive would be more reliable within that time period?

Just because SSD's aren't perfect, doesn't mean that they are useless. They are still, by far, THE technology to use for people looking for speed and resistance to physical shock.

 

I wonder how they can defend themselve under the 'normal wear and tear'. If I have a HDD that I can write say 100T and a SSD failed, what would a judge think ?

 

I would also like to address to the question "What happened at the end of life of a SSD?".

Will the SSD become read-only? Of course not. Read-only blocks are marked as BAD blocks and replaced by blocks in the reserved space. When the reserved space is depleted, this is almost the end of life of the SSD and the user should seriously consider replace the SSD.

But things are not that simple. Data stored on a SSD are not meant to be last forever. Electron charges could shift over the time and the pair of bits in a single cell become erroneous. But thanks to the built in 4-bits ECC, this is still correctable. But there is a point when even the ECC cannot save the data.

The more P/E operations is performed on a page, the shorter the page is likely to hold the data safely.

Typically an enterprise SLC SSD will have a data retention period of 10 years while a consumer MLC SSD will have a data retention period of 1 years at the end of life.

So even if you retire an old SSD and think the data are safe on it, they are not. You'd better either check your data occasionally or transfer all of them to other medium.

 

Is it the other way round ? Based on what I have seen in Intel's next generation drive, their enterprise line is geared towards short rention but higher P/E cycle, consumer line being the opposite.

 

You proved my point. THERE IS NO INFORMATION! That's why we all sit here and speculate.

 

Judge? Lol. More theorycraft. You're thinking about this too hard. You're assuming that the SSD manufacturer would be hard-headed about it and try to fight you on your warranty claim for "wear and tear".

There's a simple rule: If it's in warranty, it gets replaced. If it's out of warranty, it doesn't. ​

It isn't worth a company's time or money to try and play "gotcha" with their customers and try to prove that the warranty should not be honored.

The only exception to this rule is if there is blatant damage (e.g. physical damage or abuse, a water immersion strip being triggered on a smartphone, etc)

 

dynkin, I think the point of my post was to show that you can't calculate SSD lifespan like that. Like I said, my 120GB Vertex has 115 wear cycles after 160GB written, unless the drive is lying to itself.

 

I have 1026 wear cycles after 1.3 Terabytes written with a remaining drive life of 90% on a Vertex 60GB. That's with 1263 hours of use. The drive is 1 years old. Free space 30GB.

Continued use in similar fashion would yield about 10 years out of it.

On a 120GB Vertex, I have 1816 wear cycles after 3.8 Terabytes written with a remaining drive life of 82%. That's with 3233 hours of use. The drive is 1 year 7 months old. Free space 95GB.

Similar usage pattern would yield about 8.5-9 years left of write use.

Both Win 7.

I have other SSDs with similar stats.

 

It seems you're right. I checked the article "Intel's 3rd Generation X25-M SSD Specs Revealed" on Anandtech and it says:
"Standard MLC will last for 12 months after all erase/program cycles have been consumed. Enterprise grade MLC will last only 3 months after exhausting all erase/program cycles but will instead support many more cycles per cell."

As I said before, the more P/E are performed on a cell, the less stable it becomes and the shorter time it could hold the data safely. So if the cell is designed to be wore out in 10K cycles and the P/E threshold is set at 9K, the data retention period is definitely shorter than the one with the P/E threshold set at 6K. There is always a trade off.

I guess the logic behind this is that enterprises always replace their SSD arrays on a regular schedule while consumers are less likely to do so.

 

The life span is affected by a bunch of factors. But most importantly, the OS used, the spare space and the WA factor.

Unfortunately, OCZ Vertex with Indilinx controller is the old generation of SSD which has a WA much higher than 1. If WA is 5 and you have 20% of spare space, 160GB writing means 33 cycles.

160GB writing doesn't mean you can transfer 160GB of data. There are a lot happen on the file system level.

Even so I think your 115 cycle is still exaggerated. Could you be more specific about which OS are you using, how many spare space do you have and what is the 160GB number come from?

 

1) Windows 7 professional 64 bit
2) 120GB drive down-provisioned to 100GB (60 + 40GB partitions), 25% space usage, 75% free.
3) Cycle / written amount from Crystal Disk Info, IndilinxSSDStatus, SSDLife

 

I thought that down provisioning was only really beneficial with Intel drives?

 

benefit any SSD.

 

Although MLC and eMLC tech along with ECC and smaller nand will give us an 'affordable' and 'capacious' SSD fairly soon, all the marketing spin in the world will never make me believe that this very fragile tech (at its core) will ever be more reliable than mechanical HDD's.

I have often wondered about the above. Hmmmm.

 

SSD is not fragile. It has "measured" life expectancy. It is predictable when the SSD will reach its EOL.

Unfortunately, many consumer don't like the word "measured". If I tell them their HDTV will break down in 3 years, laptop will stop work in 5 years, they would translate these to poor qualities. On the other hand, they would perhaps buy a new one in merely 2 years without knowing they were actually replacing the old unit.

On the contrary, HDD life is quite unpredictable. It could die all of a sudden due to shock, motor failure, power outage, etc. This could happen within the first couple of years or never happen in 10 years.

Which one would you prefer, a measured life of 6 years or a sudden death between 3-10 years?