This is a bit of an out-of-cycle update as the bargain-hunting season starts to ramp up and flash memory prices seem to have bottomed out, people may be on the lookout for SSDs. I am certainly no stranger to a good deal, but it seems that my optimism about SSDs is all but lost due to some recent experiences.
TL;DR – If you have an SSD built using YMTC flash memory, you should probably check its data integrity immediately and keep an eye on it. Many low-end models appear to be released with a mixture of different parts – the only way to know if you have YMTC is either to examine the chips or know the chipset and retrieve the Chip IDs using the appropriate diagnostic tool.
Background
I’ve been an SSD user since the early 2010s, starting off with an OCZ Vertex 3 – a Sandforce-based SATA SSD reputed for its unreliability and firmware bugs. However, that SSD is still with me today and still boots my Phenom II x6 1090T test rig, clocking over 28,000 hours. A second drive I purchased for a friend is also still alive too.
Since then, almost 100 units of SSD have passed my hands, and sporadic monitoring of them suggests that I have accrued a total of around six bad blocks across the fleet up until last year with zero data loss. In the same span of time, I’ve had to decommission about ten hard disks from a smaller pool due to signs of failure, but overall, with very minimal data loss. I definitely enjoyed the reliability of SSDs, regardless of manufacturer.
The tables turned last year, when Samsung put egg on my face with their 870 EVO, for which I had to apologise for recommending as four drives failed within the same week and data loss was incurred. It seems others suffered issues which were rumoured to be either firmware related or due to poor quality flash. Even a leader in the market was not above failures – perhaps due to the pressure of COVID-19 related chip shortages or the rapidly falling price of NAND putting pressure on yields. The unimpeachable SSD was beginning to falter.
The race to the bottom with regards to pricing has resulted in some options reaching close to AU$50/TB, a price that one could only dream of in years past. My first SSD, by comparison, was at around AU$2300/TB. But it seems something had to give …
Patriot Burst Elite with Amnesia …
I reviewed the Patriot Burst Elite 1.92TB SATA SSD back in March this year. After the review, I left the H2testW test patterns on the drive and promptly forgot about it, left in the case of the test computer until recently when I needed a “2TB-class” SSD for a project.
I expected to be able to shove the drive into a USB enclosure, wipe the drive and get going … but it had other plans.
It instead was busy, all the time, only providing short bursts of data on access. The data was only written a little over six months ago!
The Illness
Attempting a read doesn’t get far – 7.1MB/s initially and soon erroring out. Was this just an issue with a few sectors? Maybe an incompatibility with the enclosure?
According to the SMART data, it had no reported uncorrectable errors and it was not reallocating sectors either. It seemed healthy and happy.
I promptly restored the drive into the test computer, which had a dead CMOS battery and somehow “lost” the ability to run SATAIII on the chipset, reverting to SATAII rates on all ports. Nevertheless …
… no dice. The drive had issues. It would start off returning data until it didn’t, dropping off the bus entirely.
A little hot-plugging and it was back online … so let’s just try to overwrite the data on the drive. Usually a drive that can’t read might accept writes happily and that would “restore” it to usability … but no. It was slow initially, just 4.4MB/s until it fell off the bus again.
Now, it had a tendency of not appearing properly at all, being “not ready”.
After a few attempts, even its ID stopped appearing correctly. The drive is very sick and yet, its SMART data seemed to claim otherwise, matching the above but with now two uncorrectable read errors rather than zero.
Salvaging the Pieces
Unfortunately, the drive was disassembled in the process of the review, voiding the warranty, as I had naively assumed that having passed the barrage of commissioning tests, it would behave like all my other SSDs which have been extremely reliable. Instead, this will become a failure that I will have to “eat”.
The next logical step for me was to try and recover it by secure erase. Booting into partedmagic, getting the drive to detect was not easy and required a few hot-plug attempts.
Once detected, I tried committing the secure erase command …
… but with all the warnings of bad things that could happen …
… it somehow failed. Such issues can sometimes end up bricking a drive completely. Trying again was no good, so I reluctantly power cycled the drive and tried again. This time, I succeeded!
Making it an Experiment
Now that the drive is secure erased, will it write and read as normal?
The answer, surprisingly, is yes. The SATAII rates is because of a motherboard issue, but the drive seems workable again.
The SMART statistics didn’t change much either except for 0xF3/0xF5 – noting that there’s been quite a few unexpected power losses due to my desperate hot plugging/unplugging. The spare block count remains the same, so in spite of the whole ordeal, the flash was still deemed “fine” by the controller.
H2testW shows data integrity is just fine as well when using the USB 3.0 enclosure that I initially was using.
Also via the enclosure, the read speeds were pretty much as expected, with the unsightly dip, as the drive was pretty much packed to the brim.
Perhaps I got lucky that the data loss had not been severe enough to render the drive’s internal metadata and firmware corrupted, otherwise I wouldn’t even be able to get this far. That being said, this is not a device to be trusted … so it’s going to be a bit of an experiment in seeing how quickly data is lost instead.
Giving it a Week
From what I know about flash memory, JEDEC standards stipulate an end-of-life data retention time of two years at room temperature. Many of my SSDs in older machines don’t get a boot-up for a year here and there, yet I have yet to lose any data. So surely, one week is not catastrophic, right?
I came back in a week and ran the throughput test on the drive –
Initial signs were not good as access was slow and seemed to occur in bursts with periods of long “waits”.
The head of the drive seems to fare poorly, but the block between 768GB and 1728GB seems to fare much better. None of the drive reaches the speeds that it did when freshly written.
This suggests to me that maybe either the flash itself is leaking charge and corrupting the data badly, with error correction consuming time and retries to correct the data (which has not been entirely lost, yet). Or perhaps there’s an issue with the controller and its firmware settings for this flash memory that causes poor performance as the data “ages”. Nevertheless, to see such a visible effect after one-week of unpowered storage is not a good sign.
While Patriot has never been my “go-to” brand for much, their products usually aren’t terrible. But this one probably takes the cake.
Fanxiang S501 Falls Off The Bus Forever!
So, what might be the cause? If it’s bad flash, well, I have another YMTC product right here in the form of the Fanxiang S501. That one is only four and a half months old and is in-service in my desktop which gets powered up at least on a twice-weekly basis. Surely, that’s working just fine … so I decide to check.
Uhh. What the …
No. The Fanxiang is also sick. It was reporting errors all over the place and running slow. Before running the test, the SMART data looked just fine.
As the test was running, I saw the Available Spare attribute count down from 100 … to 89. This is a sign that blocks are being reallocated.
After a little more, it reached 70. The threshold is 10, so this was not supposed to be a “critical” situation just yet … but one second later …
… we’ve lost the patient. It fell off the bus, never to be seen again. I couldn’t get it detected regardless of using a USB enclosure or various PCIe slots – it just won’t come up. The only obvious jumper pads put the unit into a download mode where the red LED blinks but no action on the PCIe bus.
This was frustrating, because this drive was one of my scratch drives where work-in-progress is stored. The majority of it, save for a couple of gigabytes, was restored from backup to a Kingston NV2 1TB for the time being. But the other data is well and truly lost for now, making this a remarkably bad investment when considering time and effort.
This also brings to light another issue regarding data recovery from SSDs – the act of reading can be destructive in a sense … so when issues are spotted, perhaps it’s too late to salvage your data. But perhaps more often than not, the failure is “sudden death”, so this point may be moot.
Thankfully, after a week of repeatedly contacting the seller, they agreed to provide me a full refund which I’ve put towards another Chinese-made SSD (from Lexar) that hopefully will have a bit more longevity. If not, then I’ll chalk this up to yet another experiment …
Conclusion
It was my sincere hope that with YMTC’s entry into the flash market and their qualification in SSD devices in the China domestic market and beyond, that they would offer the market an even lower cost alternative that would serve to drive competition in the flash space. Apple were considering them as suppliers at one stage, only nixed due to political issues.
As my track record with YMTC flash is now two failures from two drives, both from different vendors and utilising different controllers, I am now inclined to distrust YMTC-based devices. While I know full well that two samples is not a large number in the grand scheme of things and I could just be doubly unlucky, hitting the jackpot twice. The probability of this is also arguably slim. Or perhaps the manufacturers were taking liberties with the grade of flash memory they chose to equip their SSDs with.
As a result, erring on the side of caution, I do not recommend YMTC-based SSDs anymore regardless of the low price and would encourage anyone with YMTC-based devices to keep an eye on the data integrity especially if left unpowered for some time. Depending on the controller, having it powered regularly is no guarantee either, as controllers may not “scrub” the data automatically and re-write data that is faulty or “weak” depending on their firmware.
While the data loss from having 3TB of SSD storage going offline only amounts to a couple of gigabytes, the time and effort this cost me was not worth the savings. However, as someone interested in technology, I still feel that it was a bit of an interesting journey. Nevertheless, it seems the Patriot Burst Elite will now become a bit of a retention time experiment – I’ll check in on it on a weekly basis to see just how much it struggles …
But for everyone else – check your YMTC drives.
You might be in for a nasty surprise.
Epilogue
That being said, this wake-up call caused me to do an audit of all my SSDs within arms reach to check for data integrity – so far, I have found zero unreadable sectors, although some minor slowdowns on some models, amongst:
- OCZ Vertex 120GB
- 2 x Fujitsu 120GB (Memorite rebadge)
- Intel 730-series 240GB
- 2 x Kingston SSDNow V300 120GB
- 2 x Kingston SSDNow V300 240GB
- Kingston A400 960GB
- Mushkin Source 2 1TB
- Kioxia Exceria 960GB
- 2 x Kingston NV2 1TB
- Samsung 840 PRO 256GB
- Samsung 850 EVO 1TB
- Samsung 970 EVO 1TB
- SKHynix PC401 256GB
- Transcend MTS400 256GB
- Kingmax SME35 240GB
- Transcend SSD340 256GB
- Crucial M500 240GB
- Netac N530S 120GB
- WD SN730 1TB
- PNY CS3030 2TB
- 2 x VAVA Portable SSD 1TB
- 2 x Samsung T7 Portable SSD 2TB
- Lexar Professional Workflow 512GB
That’s a total of 29 drives and zero problems, which is what I was expecting. There may still be a few hiding around that I haven’t audited. I have a feeling that the older flash memory is just more reliable by being more conservative (higher endurance, larger lithography) and the race to the bottom in terms of pricing may be letting sub-par products onto the market which are “good enough” on paper, but not in practice. It reminds me of the days of optical discs where, as CD/DVD burning was popularised, low-cost media flooded the market with terrible longevity.