Question SSD at 23% health

[DathusBR]

My SSD KINGSTON (120GB) is at 23% of health, I don't remember exactly when I brought the SSD but the Power On Hours Counter is at almost 2 years, so i guess I brought 3-4 years ago or so because i turn up my pc at 10Am and turn down at 10PM-12AM.
I saw on some forums people saying that using virtual page file for memory tear down the SSD, but others said that is a myth.
I used the SSD for virtual page file so i guess that tear down my SSD, Is that the reason for the 23% health?
I use this SSD only for Windows 10+Apps.
I should move the virtual page file to a HDD? I only have one and is at caution on Health with no percentage.
For how long this SSD will live? There is anything i could do to make things better?

 

[lvt]

My Kingston SSD info, in case you are interested.

The 120GB Kingston model generally can last 150TBW, so your SSD should have about 60% health left.

The 23% health indicator seems weird to me.

 

[lvt]

Just to add, my SSD is also a Win XP then Win 10 boot drive with pagefile on it.

 

[DathusBR]

WoW 95%, almost double my hours and with pagefile, very odd indeed.

 

[USAFRet]

You should be prepared for a drive, any drive, to die in the next 0.25 sec.
No matter what the software says.

Your particular drive seems to be closer to that than most.
Seriously consider a replacement, soon.

1. That drive is small
2. It is not one of the better ones

If you were to move the pagefile to the HDD, you end up slowing the whole system down.

 

[Maxxify]

The A400 has gone through several hardware revisions, primarily with flash as it originally had 2D/planar TLC. In any case, 67TB of writes for a 120GB drive is a large amount. The 23% value may not be, and likely is not, accurate to real endurance, as most drives will survive writes far beyond that rating. 1000 P/E of NAND writes would be 128TB for example. Your firmware dates from late 2017 so not sure if it's after the transition to 3D TLC which has higher endurance. Given the erase count (which suggests a small block size), probably you have 2D/planar flash, and given the NAND writes you are at around 523 PEC (~67TB).

Comparison to the other Kingston drive, a HyperX Fury, is irrelevant, not least because those were made with MLC and a controller that had compression (SF-2281) which reduces write amplification. You have to compare NAND writes, not host writes.

 

[fzabkar]

The health score is probably reflected in one of the other attributes that is not shown in the screenshot.

However, if we take the Erase Count, it appears to consist of two hexadecimal values, 0x02F9 and 0x0326. I suspect that these may be the minimum and maximum counts for each block. If these correspond to a wear value of 77% (= 100% - 23%), then ...

0x0326 / 77 x 100 = 1047​

0x02F9 / 77 x 100 = 988​

So this could be suggesting that the SSD is rated for 1000 P/E cycles, or 128 TBW.

 

[lvt]

WoW 95%, almost double my hours and with pagefile, very odd indeed.

Did you try Kingston Toolbox software?

The tool should give you a lifetime report like this (photo linked from Internet as my Kingston SSD is actually disconnected)

 

[lvt]

You have to compare NAND writes, not host writes.

That's a good point, maybe because my SSD has the old firmware so the NAND writes is not available.

I think NAND writes are usually much higher than the total HOST writes because when the SSD has no dedicated DRAM cache, all caching operation is done directly on the SSD's memory.

So total Nand writes = total Host writes + Cache writes???

 

[DathusBR]

Did you try Kingston Toolbox software?

The tool should give you a lifetime report like this (photo linked from Internet as my Kingston SSD is actually disconnected)

I tried the 1.0 and v2.0 both don't find the drives. :-(

 

[DathusBR]

I installed the Kingston SSD Manger, and it found the drive, but also reports 23% of SSD Wear Indicator.
And didn't found any firmware updates.

 

[USAFRet]

I installed the Kingston SSD Manger, and it found the drive, but also reports 23% of SSD Wear Indicator.
And didn't found any firmware updates.

And with that, it could last another decade. Or die in the next 5 minutes.
Really.

 

[lvt]

I installed the Kingston SSD Manger, and it found the drive, but also reports 23% of SSD Wear Indicator.
And didn't found any firmware updates.

So be it.

My opinion is that you continue to use your SSD for a while, there is no reason why it won't reach its 50,000 hours.

However, in order to avoid any bad surprise, you should have your boot drive fully backed up, in case the SSD fails you can have your computer working again in a matter of minutes.

 

[Maxxify]

That's a good point, maybe because my SSD has the old firmware so the NAND writes is not available.

I think NAND writes are usually much higher than the total HOST writes because when the SSD has no dedicated DRAM cache, all caching operation is done directly on the SSD's memory.

So total Nand writes = total Host writes + Cache writes???

The amount written to NAND is usually higher because of write amplification, which can be impacted by a number of factors. Actually, the older SF-2281 based drives could have a WAF of less than 1.0 due to compression. However, consumer usage tends to be in the 1.5-3.0 range. Technically the WAF is 1 + (NAND/host) or (NAND + host)/(host), but in any case...

 

[fzabkar]

The amount written to NAND is usually higher because of write amplification, which can be impacted by a number of factors. Actually, the older SF-2281 based drives could have a WAF of less than 1.0 due to compression. However, consumer usage tends to be in the 1.5-3.0 range. Technically the WAF is 1 + (NAND/host) or (NAND + host)/(host), but in any case...

This is confusing. Your WAF formula can never have a value that is less than 1, yet your claim that write amplification (WA) can be less than 1 in cases of hardware compression is clearly true (and verifiable).

Therefore it would appear that WA and WAF are not the same thing. In fact Micron's tech note clarifies this:

TN-FD-23: Calculating Write Amplification Factor:
https://www.micron.com/-/media/clie...age/tnfd23_m500_smart_attributes_calc_waf.pdf

The WAF is calculated using the following equation:

WAF = 1+ (Attrib 248 / Attrib 247)

Attrib 247 = Host program page count
Attrib 248 = Background program page count

Attribute 247 is the total number of NAND page program operations initiated by the host computer.

Attribute 248 is the number of NAND page program operations initiated by the SSD's Flash Translation Layer (FTL) and are in addition to the operations programmed by the host.

Elsewhere I see that WA is defined as the number of host writes / NAND writes. That's a different quantity.

 

[Maxxify]

1 + (NAND/host) or (NAND + host)/(host) is the same thing mathematically: the second can be separated into (NAND/host) + (host/host). These two equations are shown in Micron documentation, yes.

 

[fzabkar]

1 + (NAND/host) or (NAND + host)/(host) is the same thing mathematically: the second can be separated into (NAND/host) + (host/host). These two equations are shown in Micron documentation, yes.

Yes, I understand that. However, WAF will always be greater than 1, even for an SSD that incorporates compression. Clearly the formula needs to be modified to reflect the latter.

Something like this ...

WA = (NAND data writes + NAND overhead writes) / (host data writes)​

If WAF is rigidly defined according to Micron's formula (or yours), then WA and WAF must be two different metrics. That's why I'm confused.

 

[Maxxify]

Depends on the controller's accounting, e.g. for SLC cycles if applicable. If the drive has no compression or SLC caching it will generally be > 1.0 (as on the M500). If it has SLC caching, static and dynamic have different effects including additive writes with the latter, etc. The SF-2281 can be < 1.0 (obviously). Micron's accounting is not universal. I do not own a Crucial/Micron drive but I've gone through all the WA calculations for my SSDs on my discord server to show these differences, including with several SF-2281 drives. For example, the WAF on my WD Blue 3D as defined by TLC/host is < 1.0, but I have more SLC writes, and of course with static-only SLC the wear zone is separate. Something similar also on my 545s. My SL308 is also < 1.0 by that metric...

Traditionally WA = NAND/host, yes. I do not know why Micron accounts for it differently, it may even be incorrect at least for some of their drives, and this was important for the "MX500 write amplification bug" issue.

 

[Maxxify]

Elsewhere I see that WA is defined as the number of host writes / NAND writes. That's a different quantity.

WIkipedia, sourcing from IBM, OCZ, and Intel, says NAND/host. But using an actually decent source:

https://imgur.com/zBnED6t

View: https://i.imgur.com/zBnED6t.png

 

[fzabkar]

WIkipedia, sourcing from IBM, OCZ, and Intel, says NAND/host. But using an actually decent source:

https://imgur.com/zBnED6t

View: https://i.imgur.com/zBnED6t.png

Yeah, that's what I meant to write.

 

[USAFRet]

While the technical aspects of WA are a good conversation, it has little to do with the OPs question.
"For how long this SSD will live? There is anything i could do to make things better? "

The answer to that is literally - ¯\(ツ)/¯

It may last several years, it may die tomorrow.
Prepare for that.

 

[Maxxify]

While the technical aspects of WA are a good conversation, it has little to do with the OPs question.
"For how long this SSD will live? There is anything i could do to make things better? "

The answer to that is literally - ¯\(ツ)/¯

It may last several years, it may die tomorrow.
Prepare for that.

Agreed.

I was trying to estimate the flash being utilized to get a general idea of PEC plus his amount of actual wear, to which I state he's probably about halfway through the rated endurance but you would want to replace earlier than that (pretty close to when the health hits 0, in this case). But of course, everybody should have a solid 3-2-1 backup plan.

 

[fzabkar]

While the technical aspects of WA are a good conversation, it has little to do with the OPs question.
"For how long this SSD will live? There is anything i could do to make things better? "

The OP needs to find out what is consuming the drive's P/E cycles at an excessive rate. That would answer the second question.

 

[USAFRet]

The OP needs to find out what is consuming the drive's P/E cycles at an excessive rate. That would answer the second question.

For this drive, the damage, if any, is already done.

Little or no investigation has been done into what software may be running to cause these P/E cycles.

 

[Maxxify]

It's an old drive, but his use of it as a "virtual page file" is likely the culprit or a primary contributing factor. That many writes on a relatively early A400 means it's about halfway through the flash's rated lifespan (67416/128 = ~527 PEC, 15nm Toshiba TLC is 1000 PEC). However you start throwing errors and rotating in spare blocks earlier than that, as early as 2/3 rated, which would be approximately when the health would hit 0%. So while he can still use it for a while he should start considering retirement/replacement before it hits 0%. Obviously, it should not be used for anything mission critical, and the standard backup/redundancy rules apply.