NVME SSD Serious Problems and Validity of using SATA III/IV

[Lorcann]

Hello,
I would like to bring up the fact that I have bought 4 NVME SSD drives from Seagate and Corsair over the past 2 years and they have all broken within 6 months or so. The data transfer rate is far too high for all of them, the heatsinks run too hot and cost too much in electric bills. If they were larger, of a proportional size to HDD drives and SATA III drives, both of them are similar in dimension and shape that is why they work. I have still got 2 Samsung SATA III SSDs, a 1TB drive and a 500GB drive from about 10 years ago and they still work fine amazingly. I have a newish machine from about 3 years ago which the NVME drives crashed and broke in but managed to fix the machine by removing all NVME SSDs and then with a new Samsung 500GB SATA III SSD and it is running at the right speed, fast enough for gaming and general use including the internet. There are several companies now making SATA III SSDs and I intend to buy many of them to put in my new machines, even if they are bought with NVME SSD drives included. You can install SATA III SSDs in most new machines and can even hang them down by the SATA cables in the best place in the case so you dont need to put them in any drive bays if not necessary. SATA III SSDs are the best, they run at the right temps and I will use them for years until SATA IV is made available or NVME larger drives with new ports in newer motherboards like the Z890 series. Thanks for your time.

[Lukas]

The life cycle of semiconductor memories is limited and is associated with the wear of memory cells.

Wear is usually caused by a large number of data erasure and write cycles and lower quality of manufactured memories as well as the charging rate (QLC, 3D NAND).

If you think that SATA SSDs (especially in 2.5" plastic cases) are better than NVMe drives, well, you're wrong.
A percentage of all versions end up on data recovery tables in the same way, and the most popular models end up in the largest numbers.
SATA drives also come in M2 and mSATA versions and due to the smaller and thinner laminate, they dissipate heat just as badly, which causes problems similar to the NVMe version.

The difference between budget and "pro" versions is that the budget ones are usually made on several PCB versions on very similar components (controller / memory), degradation in them occurs less dynamically, and if it occurs, it can be corrected to a greater extent.
More expensive drives, such as Samsung EVO, WD Black, Crucial, Intel and so on, are equipped with better memory, but if degradation occurs in them, it is usually much more serious.
Due to internal encryption, it is often impossible to rebuild them in software (or replace controllers if necessary) they will get damaged - tested on NVMe WD Black SED and Samsung EVO 970).

To sum up - semiconductor media get damaged sooner or later and basing your assumption on the interface alone is very inadequate to reality.

[fzabkar]

Silicon Motion SM2508 SSD review: Finally, a true PCIe 5.0 contender:

https://www.tomshardware.com/pc-components/ssds/silicon-motion-sm2508-ssd-review

[okton]

If you think that SATA SSDs (especially in 2.5" plastic cases) are better than NVMe drives, well, you're wrong.
[...]SATA drives also come in M2 and mSATA versions and due to the smaller and thinner laminate, they dissipate heat just as badly, which causes problems similar to the NVMe version.

Isn't that a kind contradictive? If the temperature is a problem, those with larger PCB should behave better because of better heat spreading. Thickness of PCB impact only on thermal "capacity", so the SSD heats slower, but in long run should not affect temperature.

BTW, do you remember there were a series of SSD made by Intel in 2.5" metal case like 5-10 years ago, do you know how those were reliable?

[WebClaw]

Depends on the model.

The Intel drives back in the day used a SandForce controller... I can tell you first hand it wasn't a good model and SandForce controllers were widely used. OCZ anyone?

That said, I agree with you. I see more and more NVMe's come into our shop of decaying NAND and once that happens it's usually a situation were data cannot be recovered.

[okton]

I see more and more NVMe's come into our shop of decaying NAND and once that happens it's usually a situation were data cannot be recovered.

I understand by decaying NAND you mean the cells are loosing charge and the content of data becames rubbish. Do anybody has any observation what is range of those damages in data? Does it happen to specific blocks or planes or is it like a "spot" in silicon crystal which affect everything around?

[Arch Stanton]

I understand by decaying NAND you mean the cells are loosing charge and the content of data becames rubbish

It's not this one dimensional, you're describing charge-loss or gain (due to retention errors, disturb errors etc.) while cells themselves degrade over time with each write and erase operation. And it is this degradation itself that will accelerate decay, it will degrade the cell's ability to retain data.

[okton]

while cells themselves degrade over time with each write and erase operation. And it is this degradation itself that will accelerate decay, it will degrade the cell's ability to retain data.

Now I am more confuser, FLASH degradation with write/erase is of well know problem with technology. FLASH manufacturers are documenting this effect by number of writes, there is existing wear leveling algorithm, and there are magic algorithms calculating/checking ECC, and if needed moving data to spare sectors. In theory everything looks correct, so what it a problem? FLASH manufacturers are lying in specifications or die quality is getting worse or FLASH controllers logic is not properly alligned to FLASH wear?

[HaQue]

specifications in die is getting finer, which logically means less can go wrong with them before they fail. They counteract this with fancy error correcting algos. The sooner people stop worrying about the ultimate search for the unicorn reliable drive, the easier they will sleep. The solution is and always has been a decent backup solution.

a well maintained Holden 186 will last for ever, put in whatever oil you want as long as it has some, and coolant or water, as long as it is cooled. whereas a 2024 GM engine... well..

Also, a single experience with drives is not enough to make any assumption on quality. The most reliable stuff can break, and the most unreliable stuff may last for some people.

You dont say what you are doing with these drives, or if they have had firmware updates that solved anything and you did/didn't apply them, or if your power supplies are good, etc etc etc. Too many variable conflated here to be anything other than a statement of experience

[Arch Stanton]

while cells themselves degrade over time with each write and erase operation. And it is this degradation itself that will accelerate decay, it will degrade the cell's ability to retain data.

Now I am more confuser, FLASH degradation with write/erase is of well know problem with technology. FLASH manufacturers are documenting this effect by number of writes, there is existing wear leveling algorithm, and there are magic algorithms calculating/checking ECC, and if needed moving data to spare sectors. In theory everything looks correct, so what it a problem? FLASH manufacturers are lying in specifications or die quality is getting worse or FLASH controllers logic is not properly alligned to FLASH wear?

'You' (a drive manufacturer) can try mitigate and adapt, you can not stop physical wear over time. 'You' can not address charge leakage over time when a drive is not connected to power. 'You' can not change the physical reality.

I do not understand your comment at all, you being more confused, then providing answer yourself and claim manufacturers are lying. What's, where's the confusion, what's, where's the lie?

[igen]

Hello everyone,

The problem with rapid degradation of NAND memory cells is the resultant exponential function of two key factors:

1). Smaller and smaller lithography. The distances between transistor junctions (base, emitter, collector) are already single semiconductor atoms (the native diameter of a silicon atom is about 2nm). And manufacturers are already trying to make conductor paths in which the average atom size is about 0.25 nm.

2). Faster and faster clock speeds. Current memories are forced to work at a clock speed of several GHz. Memories from old NANDs (where the clock speed was 50 or 100 MHz) were not as failure-prone and withstood many more write or read cycles.

Very narrow lithography plus increasingly higher frequencies have the effect that fewer and fewer molecules (now even single ones) are forced to vibrate faster and faster (there is also the factor of increasingly higher temperature in the micropoint).

Hence, the increasingly lower stability of such solutions.
The failure rate of such a system is increasing.
And the durability (as a resultant function of time and cycles) is decreasing.

To illustrate this, I will use an example.

Let's assume that we are a wizard and can slow down time a billion times. And at the same time, our vision magnifies such a NAND structure a billion times (like an electron microscope). We will see a pot of boiling liquid, which, as it boils, pours out of the container. Such an analogy comes to my mind.