SSD Drive Life

[Daemas]

Been running my OCZ vertex 32gb for over a year now (1yr, 13 days, 17 hours)

2.2TB of reads, 1.6TB of writes

Smart Data

I first started using SSDLife Pro on nov 5th and it said my drive was due to explode in may of 2011. Checked it about 2 weeks ago and it said April 11. Now it says March 29. So my SSD death has been accelerating faster and faster with no discernible differences in how I use it. (or SSDLife is a sack of crap, or my SMART field for remaining life is crap)

So, any opinions on how much longer my drive is going to last? I'm trying to get it to last until win7 SP1 and Intel G3 and hopefully socket 2011.

 

[(sic)Klown12]

SSD Life isn't a tool that you can depend on 100%. Depending on how you use the drive, it will either speed up or slow down. But since the SMART data is saying 25% left, I wouldn't do a whole bunch of benchmarking or unnecessary writing. You only have a 32GB drive, so you have very limited amount of write cycles. Your drive will last long enough for the new Intel, which should be out in a couple of months. But I seriously doubt you'll make long enough for S2011 since it's slated for late 2011.

 

[Rifter]

So SSD's last less than 2 years? really? this is making me second guess my purchase now. I thought they had the longevity issues figured out?

 

[(sic)Klown12]

So SSD's last less than 2 years? really? this is making me second guess my purchase now. I thought they had the longevity issues figured out?

Only because of how small his SSD is. If you have a 64GB, then you can expect double of what he has with the same usage. Go to 120GB, and you've almost doubled the life again. The biggest killer of write cycles is over zealous benchmarking, especially with Crystal Disk Mark which write 15GB of data on the default run. And people benchmark after every little change, so it adds up. My old 64GB Torqx has 2.3TB of host writes, more than OP, and the life is at 65%

 

[God Mode]

They need to get costs down and provision 100% for all ssd.

 

[Rifter]

Guess i will be alright then, i got a 120GB Intel on order.

 

[(sic)Klown12]

Then you have nothing to worry about. The Intel G2 is rated for writing 20GB a day for 5 years. You'll end up upgrading well before you run out of write cycles.

 

[GundamF91]

20GB per day x 365 days/year x 5 years = 36,500GB = 36.5TB.

But not sure what number is based on...probably 64GB drive or more. Is the SSD durability based on writes? It seems the SSD should last much more than just 1.6TB writes, or at least double or triple that.

 

[VeryCharBroiled]

my intel x-25 g2 80 gig shows 0.95 TB written over 6 months and 99% life left. says it will die oct 2020. SSD life has been tracking it for a few months now.

size, what is considered "spare area" and write amplification are the main factors. mine has about 50 gigs of data on it so 30 gigs are considered "spare" at the moment.

running win7 with intel AHCI drivers.

 

[frostedflakes]

I wouldn't get too hung up on the SMART data, the life time parameter is just an estimate. Also, it's not like the SSD will stop working all of the sudden, what I believe happens is you would start to lose space increasingly rapidly as cells die and the controller has to reallocate them to working ones. I think the purpose of that SMART parameter is to warn you a good amount of time before this starts to happen so you'll be able to back up any important data on the disk.

 

[(sic)Klown12]

20GB per day x 365 days/year x 5 years = 36,500GB = 36.5TB.

But not sure what number is based on...probably 64GB drive or more. Is the SSD durability based on writes? It seems the SSD should last much more than just 1.6TB writes, or at least double or triple that.

Intel's rating is based on the 120GB with 32nm NAND(5,000 cycles) on sequential writes only. That's when they achieve the lowest write amplification. Random writes will raise the write amplification, so their rating has a caveat. And yes, writes are what determines the drives' life. That's how they use up their cycles. Reads don't have much, if any effect on durability.

 

[frostedflakes]

There's no write amplification with sequential writes (or a write amplification of exactly one is another way to look at it), what you send to the disk is exactly what it writes. Write amplification is only an issue with random writes, and Intel claims it is very low with their controller, about 1.1 (meaning if you send 100MB of random writes to the controller, it will actually write 110MB).

Also, I think Intel's claim was 100GB/day for five years with the original 80GB X25-M based on 50nm flash. Not sure if this was rated for sequential or random, but because write amplification on these drives is so low, it doesn't really matter, they are effectively the same thing.

At least this is my understanding of the issue.

 

[Mark R]

This seems to be a write amplification thing. The indilinx controller appears to have terrible write amplification - especially when the drive is quite full (more of a problem o. Small capacity drives)

Looking at the OP's figures, his drive has a write amplification factor of over 50 (that's not a typo, I do mean fifty).

By contrast, I have a 1 yo intel 160 GB, with nearly 8 TB of writes on it - it's currently on 97% health - suggesting that the internal write amplification is much better - probably only 2-3.

That said, the stats on the OP's drive are interesting - some flash cells on his drive have had over 50,000 cycles - and none have gone bad! not bad considering the flash is only rated for 5,000 cycles. There is scientific evidence now suggesting that traditional methods of testing flash reliability severly underestimate cycle life in a real life situation. In reality, we don't know for sure how reliable SSDs are, just that the indilinx controllers seem to chew up the flash memory 10-20x faster than Intel or Sandforce controllers. (I've seen quite a few similar threads from people with indilinx drives on this and other forums, which have shown similar figures for write amplification.)

 

[zerogear]

How does SSDLife compare to CrystalDiskInfo? Seems like both just simply rates the drive based on the SMART values, why would I pay for SSDLife?

 

[Voo]

Looking at the OP's figures, his drive has a write amplification factor of over 50 (that's not a typo, I do mean fifty).

How do you compute that value?
Even if we assume that the Total Count of Write Sectors includes sectors written by the controller (I assume that's possible), we only end up at 3387871621*4*1024/1024^4 = 12.6TB data, which puts at an impressive WA=7.6, but far from your 50.
Intel puts its own WA for usual desktop data at 1.1 btw.


Anyhow: Considering that SSDLife Pro surely just uses something similar to the usual approximation [<total data written>/(<size of device> * <nr you can write to 1 cell>)] and I'd think they use pretty conventional values (like Intel advertising) there's a pretty large margin of error here.

 

[frostedflakes]

Yeah those write amplification values seemed a bit off. Indilinx Barefoot controller isn't great, but write amplification should be much lower than 50. The 7.6 you came to seems more in line with what I would have thought it should be. I think even the older, crappy JMicron controllers only had a write amplification of like 20-25, in that area. Good controllers like Intel will have write amplification very close to and slightly above 1.0, and SandForce is actually able to achieve write amplification <1.0 due to the ingenious data compression they use. Kind of OT, but it's just so incredible to me all the creative ways that controller designers have figured out to deal with the inherent limitations of NAND flash technology. Amazing how much the controllers have advanced just over the last 3-4 years.

 

[Mark R]

How do you compute that value?
Even if we assume that the Total Count of Write Sectors includes sectors written by the controller (I assume that's possible), we only end up at 3387871621*4*1024/1024^4 = 12.6TB data, which puts at an impressive WA=7.6, but far from your 50.
Intel puts its own WA for usual desktop data at 1.1 btw.

We use the SMART data to calculate 2 values:

First: total amount of host writes
3387871621 (C7 total count of write sectors) * 512 (bytes per sector) = 1.6 TiB

Second: Amount of actual flash cycles
3753 (D0 average erase count) * 32 GB (total size of flash drive) = 109 TiB

Write amplification = 109 / 1.6 = 68

And for my Intel

First: total host writes (7.6 TB) - from Intel SSD toolbox

Second: Total flash erases
(100 - 97 [media life remaining]) / 100 * 5000 (presumed cycle life of flash) * 160 GB = 21 TiB

Write amplification = 21 / 7.6 = 2.7

--

The OP's is not the only indilinx drive I've seen with similar WA figures.
I stand by my original assertion that the indilinx controller has shockingly bad write amplification, which wastes 90-95&#37; of the life of the flash.

 

[Hacp]

There's no write amplification with sequential writes (or a write amplification of exactly one is another way to look at it), what you send to the disk is exactly what it writes. Write amplification is only an issue with random writes, and Intel claims it is very low with their controller, about 1.1 (meaning if you send 100MB of random writes to the controller, it will actually write 110MB).

Also, I think Intel's claim was 100GB/day for five years with the original 80GB X25-M based on 50nm flash. Not sure if this was rated for sequential or random, but because write amplification on these drives is so low, it doesn't really matter, they are effectively the same thing.

At least this is my understanding of the issue.

When you say write amplification, people mean write amplification factor. There is always going to be some write amplification above one. That's the nature of wear leveling, even if the write is sequential. Write amplification factor of one means twice the amount of data is written on the drive per gb of data.

 

[Voo]

We use the SMART data to calculate 2 values:
[..]
Second: Amount of actual flash cycles

Oh that's a nice value - didn't know that even existed, because the Intel toolbox SMART reader doesn't show it, which of the countless SMART readers out there do you use? Want to see what my Intel drive shows

 

[Mark R]

Oh that's a nice value - didn't know that even existed, because the Intel toolbox SMART reader doesn't show it, which of the countless SMART readers out there do you use? Want to see what my Intel drive shows

I use Intel's SMART reader just because it reads the total number of writes correctly on the intel SSD. Most other SMART readers don't get that value right.

Regardless, that value (D0 Average flash erase cycles) is an indilinx only SMART value. No other drives have that, or if they do, they use it to mean something different. Intel, JMicron and Sandforce don't report that value directly.

 

[flamenko]

wow...los of thoughts on this one but the reality is that nobody is really sure and we base our calculations on guestimates as SSDs are so new. I have used somewhere in the area of 40-50 ssds in the past 3plus years and have yet to have any reach end of life. In fact, I am as much in the ssd environent as you can get and I dont know of any person who has used his ssd to death.

Simple as that...they will last much longer than a hdd.

 

[hal2kilo]

These drives would all probably last longer if people would stop being obsessed with showing off their benchmarks. Where you will notice the difference mostly is during boot-up and shutdown. My Vostro is up and running (except for acquiring virus updates) in 40 seconds. Beats friggin 2 minutes anyday.

 

[Daemas]

These drives would all probably last longer if people would stop being obsessed with showing off their benchmarks. Where you will notice the difference mostly is during boot-up and shutdown. My Vostro is up and running (except for acquiring virus updates) in 40 seconds. Beats friggin 2 minutes anyday.

I've probably run some form of SSD benchmark maybe 5?? times in the last year. Just windows 7 ult 64bit and programs (no games)

 

[IntelUser2000]

20GB per day x 365 days/year x 5 years = 36,500GB = 36.5TB.

But not sure what number is based on...probably 64GB drive or more. Is the SSD durability based on writes? It seems the SSD should last much more than just 1.6TB writes, or at least double or triple that.

Intel claims on sequential writes, the 160GB 34nm X25-M G2 can write 370TB of data before it dies. IMFT's Troy Winslow is quoted as saying that the 25nm cells have the same write lifespan as the 34nm cells at 5000 cycles.

In theory=160GB x 5000 cyles = 800TB

In reality=(160GB is 149GB usable GB) x 5000 cycles = 745TB but the actual lifespan is claimed to be 370TB so the controller can only manage half before it dies out

Lifespan of 160GB 34nm X25-M G2 on 100&#37; random writes = 15TB

Lifespan of the same drive in usage for most people=70-150TB

Lifespan of the OPs drive=Who knows?

 

[Troll Trolling]

Just a question:
If we assume that the Sandforce controller do, on average, a write amplification of 1; it had 120GB, and it's NAND can do 2500 cycles, would it mean that I could write in it about 300.000GB before it failed? Or about 105GB per day in 8 years?