Is an SLC SSD worth it for the longevity?

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

PandaBear

Golden Member
Aug 23, 2000
1,375
1
81
At the moment SLC last about 100k cycle, MLC (2 bit) last about 5k cycle, and 3LC (3 bit) about 100-500 cycles.

All before wear leveling.

So if you intend to fill a drive to near full and perform lots of writing to it, then I'd go with a small SLC drive. If you are planning to just store lots of data to it and read most of the time, then MLC with a good controller is sufficient.
 

Idontcare

Elite Member
Oct 10, 1999
21,118
59
91
Originally posted by: PandaBear
At the moment SLC last about 100k cycle, MLC (2 bit) last about 5k cycle, and 3LC (3 bit) about 100-500 cycles.

All before wear leveling.

So if you intend to fill a drive to near full and perform lots of writing to it, then I'd go with a small SLC drive. If you are planning to just store lots of data to it and read most of the time, then MLC with a good controller is sufficient.

Where are you getting your lifetime data from? It does not match the published/stated numbers.
 

jimhsu

Senior member
Mar 22, 2009
705
0
76
Originally posted by: fleshconsumed
Originally posted by: AyashiKaibutsu
A bit off topic, but what happens when most of the drive is full and what is free is constantly be rewritten for the page file/system restore info/etc? Couldn't that cause an issue with some sectors wearing out much faster than expected?

AFAIK in this situation yes, those cells will wear out faster. This may or may not kill the drive faster depending if it can keep track of those cells and somehow rearrange the data to avoid writing to worn down cells in the future. However, this will entirely depend on the drive's wear leveling algorithm. Second SSD article on Anandtech mentioned this problem.

Most good wear leveling algos find the cell with the LEAST number of writes whether it is full or empty, appends the content to be written to the cell, and writes the result to the cell with the MOST writes. Hence, whether the cell if empty or full is irrelevant. Having more "free space" simply gives the SSD more breathing room and makes the process faster (hence why SSDs slow down if a lot of free space is used).
 

PandaBear

Golden Member
Aug 23, 2000
1,375
1
81
Originally posted by: Idontcare
Originally posted by: PandaBear
At the moment SLC last about 100k cycle, MLC (2 bit) last about 5k cycle, and 3LC (3 bit) about 100-500 cycles.

All before wear leveling.

So if you intend to fill a drive to near full and perform lots of writing to it, then I'd go with a small SLC drive. If you are planning to just store lots of data to it and read most of the time, then MLC with a good controller is sufficient.

Where are you getting your lifetime data from? It does not match the published/stated numbers.

I work for SanDisk.
 

PandaBear

Golden Member
Aug 23, 2000
1,375
1
81
Originally posted by: jimhsu
Originally posted by: fleshconsumed
Originally posted by: AyashiKaibutsu
A bit off topic, but what happens when most of the drive is full and what is free is constantly be rewritten for the page file/system restore info/etc? Couldn't that cause an issue with some sectors wearing out much faster than expected?

AFAIK in this situation yes, those cells will wear out faster. This may or may not kill the drive faster depending if it can keep track of those cells and somehow rearrange the data to avoid writing to worn down cells in the future. However, this will entirely depend on the drive's wear leveling algorithm. Second SSD article on Anandtech mentioned this problem.

Most good wear leveling algos find the cell with the LEAST number of writes whether it is full or empty, appends the content to be written to the cell, and writes the result to the cell with the MOST writes. Hence, whether the cell if empty or full is irrelevant. Having more "free space" simply gives the SSD more breathing room and makes the process faster (hence why SSDs slow down if a lot of free space is used).

Very true. A file system knows whether a space is empty or not, but any storage device like SSD works at a sector/lba level and it does not know whether it is empty or not. However, the assumption of the usage model is that a typical workstation/user read and write the same amount of data regardless of the size of the drive, so there are more space to cycle over on a larger drive, therefore it last longer. If you increase the workload by the same ratio as the storage size increase, then you are right, it will last the same amount.
 

taltamir

Lifer
Mar 21, 2004
13,576
6
76
Originally posted by: PandaBear
Originally posted by: Idontcare
Originally posted by: PandaBear
At the moment SLC last about 100k cycle, MLC (2 bit) last about 5k cycle, and 3LC (3 bit) about 100-500 cycles.

All before wear leveling.

So if you intend to fill a drive to near full and perform lots of writing to it, then I'd go with a small SLC drive. If you are planning to just store lots of data to it and read most of the time, then MLC with a good controller is sufficient.

Where are you getting your lifetime data from? It does not match the published/stated numbers.

I work for SanDisk.

interesting, so you say it is 5K and not 10K for MLC?
 

jimhsu

Senior member
Mar 22, 2009
705
0
76
It may depend on grade, application, confidence intervals, etc.

Say that the mean lifetime of a cell is 20000 cycles and the standard deviation is 5000 cycles. A 95% confidence interval with sample size 1 (one drive) is 10200-29800 (+- 1.96 stdev), close to the 10K estimate by intel. But an enterprise application might demand a 99% confidence interval, which is 7120-32880 (+- 2.58 stdev), closer to sandisk's estimate. It could be in estimates of enterprise life, sandisk chooses to use a more stringent estimate, while intel clearly targeting the M drives to consumer uses a more relaxed confidence interval.
 

Idontcare

Elite Member
Oct 10, 1999
21,118
59
91
Originally posted by: PandaBear
Originally posted by: Idontcare
Originally posted by: PandaBear
At the moment SLC last about 100k cycle, MLC (2 bit) last about 5k cycle, and 3LC (3 bit) about 100-500 cycles.

All before wear leveling.

So if you intend to fill a drive to near full and perform lots of writing to it, then I'd go with a small SLC drive. If you are planning to just store lots of data to it and read most of the time, then MLC with a good controller is sufficient.

Where are you getting your lifetime data from? It does not match the published/stated numbers.

I work for SanDisk.

Are your numbers specific/limited to Sandisk products or are you communicating to us that other sources which are implying/communicating higher-cycle lifetime numbers are wrong?

http://www.anandtech.com/stora...howdoc.aspx?i=3631&p=6

http://pc.watch.impress.co.jp/...s/211/964/kaigai11.jpg
 

toslat

Senior member
Jul 26, 2007
216
0
76
Q: When a cell loses integrity from excessive writes, what do u lose: a bit, a page, or a block?
 

VirtualLarry

No Lifer
Aug 25, 2001
56,479
10,137
126
Originally posted by: toslat
Q: When a cell loses integrity from excessive writes, what do u lose: a bit, a page, or a block?

Also, when are errors detected? During a write cycle, or a read cycle? If it's a read cycle, then do SSDs do background scanning to pre-emptively detect failures?
 

Idontcare

Elite Member
Oct 10, 1999
21,118
59
91
Originally posted by: toslat
Q: When a cell loses integrity from excessive writes, what do u lose: a bit, a page, or a block?

When a bit dies you lose that bit, but because the system wouldn't know if the bit is dead (i.e. wrong, its a zero when it should be a one, or vice versa) you don't lose anything, just the resultant file when all the bits are added up will contain corrupted data as a bit will be flipped at some point in the file.

At any rate supposedly you don't lose the bit from excessive writing, you can continue to read the bit but you can't write and change the cell (reliably anyways) anymore.

Originally posted by: VirtualLarry
Also, when are errors detected? During a write cycle, or a read cycle? If it's a read cycle, then do SSDs do background scanning to pre-emptively detect failures?

It's detected in the read, but the read is done immediately after the write to verify the write was successful (this happens internally, you don't see it going on).

If the read shows that the write attempt was unsuccessful then a whole other calibration loop is invoked to mark the cell as potentially bad while the data are written again to the next set of known good cells, with subsequent verification by way of read as well.

Once it is written and verified to be a good write there is no more "refresh" verifications that go on, unlike dram and sram, unless you specifically invoke a read operation of the cell contents. Then there are more diagnostic loops invoked in parallel to the read that determines whether the cell's charge level is good enough to leave it be or whether it needs to be refreshed or potentially marked as bad. Very similar in concept to how read/write verifications happen on spindle-based media as well.
 

Golgatha

Lifer
Jul 18, 2003
12,691
1,617
126
Originally posted by: Idontcare
Originally posted by: Russwinters
I would NEVER use the same hard drive for anything over two year because after that the chance of failure increases significantly.

Using a hard-drive, or any hardware component, to failure is not advisable.

Preventative maintenance includes planned obsolescence and planned replacement. SOP across many industries and fields.

Obviously the basis of a decision for pursuing SLC-based SSD's ought to include some manner of rationalizing and estimating the duty-cycle (GB of writes per day versus GB capacity of the drive) at which you intend to operate the drive.

That is assuming the performance advantage of SLC is not a factor.

But if all things were equal for my personal usage pattern then I would not opt to buy the SLC just to have an SSD with extended lifetime. I'd buy an MLC and invest the difference for a few years time and then buy whatever latest and cheapest and fastest MLC-based SSD the market has to offer then.


I agree about rotating hard drives. I'd be very wary of keeping a spindle drive for more than 3 years. My storage needs have grown substantially over the years and much of my data is precious enough to me to justify the small extra costs. So far my need for more storage has outpaced my personal 3 year deadline to replace drives, but I'd say those days are becoming short, as my current drives are around 1.5-2 years old and still going strong.
 

taltamir

Lifer
Mar 21, 2004
13,576
6
76
Originally posted by: toslat
Q: When a cell loses integrity from excessive writes, what do u lose: a bit, a page, or a block?

generally you don't lose bits, the "longevity" figure (10k writes) is the point at which the drive will simply refuse to write to it anymore, considering it read only isntead. Which is supposed to be BEFORE any bits actually burned out. However if a bit did burn out, you will just have lost a single bit.
 

PandaBear

Golden Member
Aug 23, 2000
1,375
1
81
Originally posted by: taltamir

interesting, so you say it is 5K and not 10K for MLC?

It is batch/generation/revision/stepping specific.

Just like saying Intel's 45nm CPU can run at 4GHz, but forgot to mention if it is when it first come out, or after they worked on it for 6 months.
 

PandaBear

Golden Member
Aug 23, 2000
1,375
1
81
Originally posted by: Idontcare
Originally posted by: toslat
Q: When a cell loses integrity from excessive writes, what do u lose: a bit, a page, or a block?

When a bit dies you lose that bit, but because the system wouldn't know if the bit is dead (i.e. wrong, its a zero when it should be a one, or vice versa) you don't lose anything, just the resultant file when all the bits are added up will contain corrupted data as a bit will be flipped at some point in the file.

At any rate supposedly you don't lose the bit from excessive writing, you can continue to read the bit but you can't write and change the cell (reliably anyways) anymore.

You are assuming that there is no ECC on a flash device. Almost all NAND card/SSD out there have ECC nowadays rating from 6 bit per sector to 122 bit per page (16 sectors max as of now). The failure model of worn out NAND is usually very Gaussian, so before the ECC reach the maximum correctable level, the firmware should reassign/relocate the block and mark it bad. When the reserves are used up, the card is dead (no name cheap stuff) or becomes read only (SanDisk).


Originally posted by: Idontcare
Originally posted by: VirtualLarry
Also, when are errors detected? During a write cycle, or a read cycle? If it's a read cycle, then do SSDs do background scanning to pre-emptively detect failures?

It's detected in the read, but the read is done immediately after the write to verify the write was successful (this happens internally, you don't see it going on).

If the read shows that the write attempt was unsuccessful then a whole other calibration loop is invoked to mark the cell as potentially bad while the data are written again to the next set of known good cells, with subsequent verification by way of read as well.

Once it is written and verified to be a good write there is no more "refresh" verifications that go on, unlike dram and sram, unless you specifically invoke a read operation of the cell contents. Then there are more diagnostic loops invoked in parallel to the read that determines whether the cell's charge level is good enough to leave it be or whether it needs to be refreshed or potentially marked as bad. Very similar in concept to how read/write verifications happen on spindle-based media as well.

These are all done inside the NAND instead of by the controller/ASIC of the card. There are a lot of partial replacement of capacity than just the entire block nowadays, but I can't comment on the details due to work.

 

PandaBear

Golden Member
Aug 23, 2000
1,375
1
81
Originally posted by: VirtualLarry
Originally posted by: toslat
Q: When a cell loses integrity from excessive writes, what do u lose: a bit, a page, or a block?

Also, when are errors detected? During a write cycle, or a read cycle? If it's a read cycle, then do SSDs do background scanning to pre-emptively detect failures?

During write the NAND is responsible for adjusting the cell's data to correct level before timeout or send back error/completion status to the controller.

During read, the NAND wouldn't know if the data is already bad, but the ASIC has ECC to detect failure scenarios and correct it.

Background scan do not happens because too much read will actually weaken the signals.
 

PandaBear

Golden Member
Aug 23, 2000
1,375
1
81
Originally posted by: Idontcare


Are your numbers specific/limited to Sandisk products or are you communicating to us that other sources which are implying/communicating higher-cycle lifetime numbers are wrong?

http://www.anandtech.com/stora...howdoc.aspx?i=3631&p=6

http://pc.watch.impress.co.jp/...s/211/964/kaigai11.jpg



Look, I'm not saying anything is right or wrong or there is a supreme being in the FAB that say these NAND will fail at a particular cycle immediately, all of a sudden.

The published numbers are just a guarantee by the manufactures that these NAND will still be functional after so many cycles if you use it correctly (i.e. sufficient level of ECC).

It is more like a warranty length than a accurate prediction of one particular NAND's life, and can be increased if you improve the process, or reduced if you shrink your node and ship products that can tolerate reduced life. This is where good controller that cost more money can actually save money, by using cheaper but reduced life NAND before your competitors can.
 

Idontcare

Elite Member
Oct 10, 1999
21,118
59
91
Originally posted by: PandaBear
Originally posted by: Idontcare


Are your numbers specific/limited to Sandisk products or are you communicating to us that other sources which are implying/communicating higher-cycle lifetime numbers are wrong?

http://www.anandtech.com/stora...howdoc.aspx?i=3631&p=6

http://pc.watch.impress.co.jp/...s/211/964/kaigai11.jpg



Look, I'm not saying anything is right or wrong or there is a supreme being in the FAB that say these NAND will fail at a particular cycle immediately, all of a sudden.

The published numbers are just a guarantee by the manufactures that these NAND will still be functional after so many cycles if you use it correctly (i.e. sufficient level of ECC).

It is more like a warranty length than a accurate prediction of one particular NAND's life, and can be increased if you improve the process, or reduced if you shrink your node and ship products that can tolerate reduced life. This is where good controller that cost more money can actually save money, by using cheaper but reduced life NAND before your competitors can.

You lost me there, I'm merely trying to close the gap between your stated lifetime numbers (as a sandisk rep) and those of other sources such as Intel and Samsung.

If you are defining your lifetime metrics differently than how Intel and Samsung define them then you aren't adding much value here because we are all accustomed to thinking about lifetime metrics within the context as they define them.

(you say 5k writes, they say 10k writes, if you mean actual lifetime versus Intel meaning warranty lifetime then you are needlessly creating an apples to oranges comparison here that people may not realize as they attempt to rationalize the disparity in lifetime numbers)

As it stands now I can't tell if your numbers were meant to be restricted in application to just sandisk products or if you meant to suggest the numbers from other manufacturers are overstated.

If you meant neither then please just explain what you meant. We aren't trying to trip you up or call shens or anything here, we are just trying to rationalize why everyone out there but PandaBear and Sandisk has set our expectations to assume MLC chips are designed to meet a minimum lifetime threshold of 10k writes with the understanding that some chips will have higher lifetimes depending on process-induced manufacturing variation but that presumably Q/A has been engineered such that very few chips will fall short of reaching that 10k minimum spec.
 

PandaBear

Golden Member
Aug 23, 2000
1,375
1
81
I understand what you mean. The endurance is a bit confusing and what I meant to say is that manufacture may decide to release a product to a customer when they meet an endurance target, or jump the gun if they have controllers that can handle it.

Each new process takes a while to mature, and what your link provide is the typical mature product endurance: 100k for SLC, 10k for MLC, 1k for 3LC. My assumption is that these were based on 40-60nm processes that are already mature, and can buy from any manufacturers and used any ASICs from Phison, 3S, TDK, etc with little to no problem.

The industry is migrating to 32-34nm at the moment and some manufacture (i.e. Micron, SanDisk/Toshiba) are already shipping in this generation. The life isn't that great yet, but good enough if you have a controller that can "wear level" and "ECC" your way out of the low life. This is how integrated memory / card manufacturers like SanDisk, Toshiba, Samsung, Lexar/Micron made their money, by being the first out of the gate with products that can handle lower quality memory with internal knowledge of the weakness of the NAND. When the processes mature, 3rd party card manufacture start buying mature memory with cheaper ASIC from companies like Phison that cost less than what integrated memory manufacturers cost to make.

So, what you say is correct, that most memory manufactures don't release a memory until they get close to these target, but many of them can still use the weaker ones in house if they know what they are doing, and take advantage of the memory that would have been scrapped.
 

PandaBear

Golden Member
Aug 23, 2000
1,375
1
81
Also, what the finished products sell is a "wear leveled" and "ECCed" card/drive that mask/obsolete the endurance's number of the NAND used. Intel may sell you an SSD that does not do partial page update internally until they need to flush its DRAM buffer, yet Micron may sell you an SD card that must flush a partial page update at the end of every command before the 250ms timeout and power down. This difference alone may change the endurance count up to 16x, despite using the same NAND.

 

Idontcare

Elite Member
Oct 10, 1999
21,118
59
91
Thanks for explaining PandaBear, makes sense to me now that you explained in that way. Particularly from the maturity vs. cost and margins standpoint.

With 3LC, do you foresee the technology iterating to the point where endurance is high enough to enable the use of them in consumer grade SSD's?

And if yes, do you expect the cost savings (over MLC) to be large enough to justify transitioning to the lower endurance 3LC chips in as much as the cost savings of MLC enables their employment versus SLC-based SSD's?

(taking into account that endurance is also a function of cost, both design and production, so naturally I am assuming that a portion of the 3LC endurance improvements will come with an increase in the cost of 3LC over the lower endurance 3LC chips that will be used in throw-away gadgets)
 

PandaBear

Golden Member
Aug 23, 2000
1,375
1
81
At the moment or foreseeable future, I don't think 3LC will be anywhere near ready for SSD usage. It is not just the endurance of 3LC being significantly lower, the NAND is also significantly slower to a point that you cannot finish a garbage collection fast enough in the worst case scenario. Not so long ago most NAND were fast enough that manufactures underclock their memory for low end cards, but now with 3LC they barely have enough speed to qualify the lowest performance requirement, and move the 2LC to medium performance products.

This is the main reason many FABs and card manufactures are not even considering 3LC any time soon.
 

VirtualLarry

No Lifer
Aug 25, 2001
56,479
10,137
126
Originally posted by: PandaBear
Background scan do not happens because too much read will actually weaken the signals.
Really? Does that mean that if I Ghost my system image onto an SSD, and then don't write to it, but read from it daily, that my data will fade quicker?

Would you recommend doing a full image backup, followed by a secure erase, and then a full image restore, for an SSD, to refresh the data? How often would one need to do such a thing?

 

PandaBear

Golden Member
Aug 23, 2000
1,375
1
81
Reading daily will be fine, unless you read it millions of time daily.

If active background scan is operated continuously, then it will wear out the data rather than keeping the data fresh. So it is doing more harm than good, isn't it?

Normal ECC during user read will be sufficient in determining the health of the data.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |