SSD? Still with downsides? Lower usage life?

S

spincut

[H]ard|Gawd
Joined
Jan 19, 2005
Messages
1,191
So, I have not really been following SSD's since they came out, and I need more storage, indicating I probably should still just get a solid platter drive again anyway (WD Caviar Black or somesuch), but I thought to at least inquire how far SSD has come in the past few years.

I know back 5 years ago or whenever when they first came out, there was talk about them only being able to be rewritten so many times, and that they were only garunteed to last so long (not that platter drives dont eventually break down as well due to the nature of moving parts, but by and large they would generally last longer barring any flukes).

So have SSD's firmly and fully outpaced platter drives now? Both in speed AND reliability? I guess if someone could better give me the skinny on this that would be great.
 
The whole deal about SSDs having a certain write limit is moot. By the time you reach that limit, other things about the SSD would fail. Typical HDDs would suffer mechanical failure before your SSD craps out due to write limitations.
 
Dizon248 said:
The whole deal about SSDs having a certain write limit is moot. By the time you reach that limit, other things about the SSD would fail. Typical HDDs would suffer mechanical failure before your SSD craps out due to write limitations.
Click to expand...

Mind being a little more specific of when that would be though? Yes some hard drives conk out after a while, but I know people who have been using theirs for quite a while, and let's be honest technology in that regard doesnt advance as fast as it used to, so many people are still rocking the 7200rpm HD they got, perhaps back when SATA first came out, or before that, long using the same IDE drive sa long as the space suited them.

So if the amount of time, while shorter, is truly moot, how long exactly is that still if you dont mind my asking? Has the length of time increased since they originally came out? Or have they not bothered to improve on that?
 
I consider it to be moot. Intel G2 SSDs can write 100 GB a day for 5 years straight.
Considering the average person does 2-3 GB a day and power users hit ~20 GB a day that's plenty of headroom.
 
Let do a little bit of math to determine lifespan.

MLC should have a range of 10,000+ cycles per cell.
SLC should have a range of 100,000+ cycles per cell.

Lets use the MLC type SSD in this example.

60GB SSD * 10,000 = 600,000 for total lifespan.
Lets say you generate 116GB of new data each and every day. (i guessed this number, you could be higher or lower)
600,000 / 116 / 365 = 14.17 years of life span before the cells wear out. If wear leveling works optimally.

The less new data you generate the longer it will last.

Other factors, like the reserved area, will add to the life span as well.

In most consumer scenarios, physical environment should destroy the drive before it wears out from writes.
 
Last edited:
The wear leveling algorithms are robust enough that they haven't needed to do much work on longevity. I think the numbers Intel was throwing around were 100GB a day for 5 years, or something like that. So really not a factor.
 
I'd worry more about the logic ICs crapping out before the write cells do. And when they do crap out, they tend to do so without warning. At least HDDs start making funny noises when they're about to die. Regardless, as long as you have a good backup plan, the speed gains from SSD are quite worth it.
 
I prefer to estimate the SSD lifetime slightly differently, and I will use more realistic numbers for the latest generation of flash.

First, assume non-enterprise usage. In non-enterprise usage, it should be a safe assumption that very few people will daily write an amount to the SSD greater than half the capacity of the SSD. While there may be some rare usage cases where that assumption will be false, it should be exceedingly rare. You'd basically have to fill and empty the SSD every day to exceed that.

Further, assume that the average write amplification is 2. While it is possible that some usage patterns will exceed WA of 2, it is only likely with a lot of small, random writes, and in that case, it is even more unlikely that the user is coming close to writing half of the SSD capacity daily. So, while WA > 2 may not be exceedingly rare, WA > 2 together with writing a large fraction of the SSD capacity each day, should be exceedingly rare.

With WA = 2 and writing half of the SSD capacity daily, the net effect is that you will be using up an average of 1 erase/program cycle for all of the flash each day. Assuming 25nm MLC flash with 3000 program/erase cycles, that comes to a 3000 day lifetime. That is about 8 years.
 
xnoodle said:
I consider it to be moot. Intel G2 SSDs can write 100 GB a day for 5 years straight.
Considering the average person does 2-3 GB a day and power users hit ~20 GB a day that's plenty of headroom.
Click to expand...

I've been using my Intel G2 for over a year with OS, applications (except games), pagefile and hiberfil still in place, no tweaks and the wear/lifetime rating is still at 99/100.
 
satterth said:
Let do a little bit of math to determine lifespan.

MLC should have a range of 10,000+ cycles per cell.
SLC should have a range of 100,000+ cycles per cell.

Lets use the MLC type SSD in this example.

60GB SSD * 10,000 = 600,000 for total lifespan.
Lets say you generate 116GB of new data each and every day. (i guessed this number, you could be higher or lower)
600,000 / 116 / 365 = 14.17 years of life span before the cells wear out. If wear leveling works optimally.

The less new data you generate the longer it will last.

Other factors, like the reserved area, will add to the life span as well.

In most consumer scenarios, physical environment should destroy the drive before it wears out from writes.
Click to expand...

Forceman said:
The wear leveling algorithms are robust enough that they haven't needed to do much work on longevity. I think the numbers Intel was throwing around were 100GB a day for 5 years, or something like that. So really not a factor.
Click to expand...

alamone said:
I'd worry more about the logic ICs crapping out before the write cells do. And when they do crap out, they tend to do so without warning. At least HDDs start making funny noises when they're about to die. Regardless, as long as you have a good backup plan, the speed gains from SSD are quite worth it.
Click to expand...

Hmmm, so ok, this i what I am driving at though, both hard drives, platter or SSD, are prone to failure, but it also seems like "how" they fail is also signifigant. Since it is moving parts, platters can fail before their time, or go long beyond it depending on things like luck and how they're cared for. And if/when they do fail it seems like there is a bit of warning.

So when these SSD's fail, or when/if they do reach the end of its lifespan (because let's face it, people do use their HD's for that long sometimes or keep old ones as backups), do you just try and switch em on day and they dont switch on? Is there any sort of counter measures to at least let you know the drive is on its last legs and you probably should save your info? One would think that, of all things, should be a major benefit of a device like an SSD, so it can better keep track of its health and there would be less unexpected and unresecuable situations over platter drives.

Also, I am rather unfamiliar with SSD's in general, so differences between MLC and SLC, as well as write amplification, are terms I am not really sure about....might need a little extra clarification.
 
I believe that when they fail for write lifetime, the do so gracefully and you can still read out the data that is currently on the chips. However, the ungraceful failure where a logic chip fails, or something similar, is more problematic because data recovery on failed SSDs is more problematic (you can't just swap out a bad logic board like you can for a hard drive for instance).

I'd say in general, from what I've seen so far, when they fail they don't often give a lot of warning - you'll just turn on the computer and the drive won't be recognized, but that's all based on anecdotal evidence. It's not like a hard drive where you might start getting a lot of errors, or it'll start making noises to tip you off. Still early to draw any long-term conclusions from that though.
 
So when these SSD's fail, or when/if they do reach the end of its lifespan (because let's face it, people do use their HD's for that long sometimes or keep old ones as backups), do you just try and switch em on day and they dont switch on?
Click to expand...

Are you talking about failure do to the NAND write endurance or other failures?

For the first one I have yet to see any forum post anywhere where someone has worn out their SSD due to writing too much data. Now the second we have seen quite a few of those with varying failure behaviors. Two of these behaviors are the drive shows up as a few MB or does not show up at all.
 
Dizon248 said:
The whole deal about SSDs having a certain write limit is moot. By the time you reach that limit, other things about the SSD would fail. Typical HDDs would suffer mechanical failure before your SSD craps out due to write limitations.
Click to expand...

I agree. A typical consumer hard drive is expected to last 5 to 7 years. On top of that it is expected that hard drives will have around a 1% to 7% annual failure rate overall for drives in use in this expected lifetime.

SSDs expected lifetime is at least 2 times this. I have not seen enough data to give an expected annual failure rate on SSDs. I believe Intel estimated theirs to be less than 1%.
 
The most [H] use for an i can think of is put one as my HTPC TV buffer. In HDTV that is like 10GB/hour 5 hours/day of "live" tv. I have yet to buy one for this purpose but certainly a good test for those wearing algorithms.
 
I can safely assume that by the time my SSD is supposed to have crapped out, I would have already replaced it with something better.
 
SSDs can still theoretically fail in several ways.

In theory, if an SSD fails due to NAND wear it will fail on erase - not write and not read. This means that the data which is "lost" was already marked for erasure, the controller will see the failure and can mark the cell "bad" and remap around the problem. This is all in theory. If I had to pick a vendor to implement this properly my money is on Intel.

SSDs can still suffer controller failures which are not in the NAND. The risk and outcomes of this are the same as for an HDD. The drive will become completely inoperative. I suppose one key difference is whether all of the mapping data is lost in the case of the SSD controller failure. With an HDD you could, in theory, swap in a working controller board and continue using the mechanical part of the drive without a problem. I'm not sure if the same kind of operation would ever be possible with an SSD since critical mapping data could be lost forever.

Either way, I don't lose sleep at night worrying about it. SSDs have the same risk of controller failure and the data failures are more predictable. HDDs often wake up one morning going "whack whack whack" and they are totally useless. Having single cells fail on erase is far easier to deal with.
 
satterth said:
Let do a little bit of math to determine lifespan.

MLC should have a range of 10,000+ cycles per cell.
SLC should have a range of 100,000+ cycles per cell.

Lets use the MLC type SSD in this example.

60GB SSD * 10,000 = 600,000 for total lifespan.
Lets say you generate 116GB of new data each and every day. (i guessed this number, you could be higher or lower)
600,000 / 116 / 365 = 14.17 years of life span before the cells wear out. If wear leveling works optimally.

The less new data you generate the longer it will last.

Other factors, like the reserved area, will add to the life span as well.

In most consumer scenarios, physical environment should destroy the drive before it wears out from writes.
Click to expand...

Just wanted to make a correction.

50nm MLC is 10,000 Writes
34nm MLC is 5,000 Writes
25nm MLC is 3,000 Writes

What everyone has posted so far is correct you will have something else fail in the ssd or the system before writes become an issue.

My current G2 intel drive is at 1.5TB host writes after 15 months of use at this rate i'll be able to retire and still use the drive.
 
Makaveli@BETA said:
Just wanted to make a correction.

50nm MLC is 10,000 Writes
34nm MLC is 5,000 Writes
25nm MLC is 3,000 Writes

What everyone has posted so far is correct you will have something else fail in the ssd or the system before writes become an issue.

My current G2 intel drive is at 1.5TB host writes after 15 months of use at this rate i'll be able to retire and still use the drive.
Click to expand...

though the sheer storage desnity increase will help somewhat offset the lower write capacity... I think?
 
Yes a Larger SSD has more cells to spread the writes to.

So a 160GB SSD should last longer than a 80GB SSD when it comes to write endurance when given the same type of data load.
 
The real thing to learn about data is no matter what it is stored on, if it is important have a backup of this data.

No HD (even one made of pure unobtanium) is perfect, and no SSD (even if made of pure unicornium) is perfect. Either might fail at any time.

If your data is important, back it up.

But the chance of hitting the read/write cycle limit of a flash based SSD is terribly unlikely.
 
Still interesting to find out that SSD's are now supposed to last twice as long as platter based HD's, never would have figured that to be the case.
 
spincut said:
Mind being a little more specific of when that would be though? Yes some hard drives conk out after a while, but I know people who have been using theirs for quite a while, and let's be honest technology in that regard doesnt advance as fast as it used to, so many people are still rocking the 7200rpm HD they got, perhaps back when SATA first came out, or before that, long using the same IDE drive sa long as the space suited them.

So if the amount of time, while shorter, is truly moot, how long exactly is that still if you dont mind my asking? Has the length of time increased since they originally came out? Or have they not bothered to improve on that?
Click to expand...


It only really needs to last until the next upgrade....for me, anyway. Right now, I'm a 1 year and nearly 5 months. We shall see.
 
You must log in or register to reply here.
Back
Top