16-year-old SATA II SSD survives 1 petabyte of writes - 25x more than the drive's endurance rating

Experiment Overview

As part of an experiment, an enthusiast has written one petabyte of data on a legacy Sandisk P4 SATA II SSD that was released 16 years ago.

A fascinating experiment conducted by the YouTube channel WolfyTech shows that SSDs are more durable than we think, even if they were released 16 years ago. Over the course of the experiment, the channel wrote one petabyte of data to the drive, and the SSD, despite having over 60,000 hours of power-on time, continues to function and shows no signs of catastrophic failure.

Understanding SSD Endurance

The NAND in SSDs gradually degrades over time as you write or erase data, similar to the wear and tear your car or virtually any other electronic device in your house experiences. Just as cars come with a manufacturer's warranty defined by either years of use or a certain number of miles, whichever comes first, SSDs also come with a warranty defined by either years of use or a metric known as TBW (Terabytes Written).

However, there is a common misconception that when an SSD exceeds its TBW rating, it will immediately stop working or become unusable. In reality, the TBW value is simply a guideline that manufacturers establish for warranty coverage. The statistical-based rating is not a definitive indicator of when the drive will fail. Contrary to popular belief, chipmakers don't program NAND to self-destruct when it surpasses the TBW threshold.

To revisit the car analogy, just as a car can run fine beyond 100,000 miles, an SSD can continue to function after exceeding its TBW rating. However, just as older cars may require more frequent maintenance and become less predictable over time, SSDs that surpass their TBW threshold may gradually become less reliable. This is due to the physical wear that accumulates in the drive's NAND flash memory cells through repeated Program/Erase (P/E) cycles.

The Sandisk P4 Specifications

Manufacturers engineer SSDs to run beyond the rated TBW by a significant margin, and the Sandisk P4 is a testament to that. The P4, which launched in 2010 in various formats, including mSATA, primarily targeted OEMs of netbooks, tablets, or ultra-thin notebooks.

In its defense, the P4, available in capacities from 4GB to 128GB, used 32nm MLC NAND. Although it's on the archaic side, 32nm 2D MLC NAND is physically larger and can withstand more write cycles than today's 3D TLC or QLC NAND.

Information on the P4 is slim, given that it was released more than a decade and a half ago. Nonetheless, we dug up an old specification sheet showing that the P4 64GB, the model used in WolfyTech's experiment, has a 40 TBW endurance. Therefore, 1 PB or 1,000 TB of written data exceeds the TBW by 25X. The drive also logged over 60,000 power-on hours and over 1,100 power-ups.

Caveats and Community Discussion

It would seem that the user created a workload that kept making cached writes to the SSD. This isn't the first time we've seen an SSD outlive the manufacturer's specified TBW limit. If you look online, you'll find many SSD endurance tests that challenge the conservative TBW numbers that vendors slap on their products. However, this particular case was pretty interesting, given that it was on an older drive with MLC NAND that has since disappeared from the market.

Even though your SSD will likely have a longer lifespan than its TBW rating, it doesn't mean you should carelessly push your SSD to its breaking point. On the contrary, given the current market situation, you should be taking extra good care of your SSD.

• pjmelect replied: "However, there is a common misconception that when SSD exceeds its TBW rating, it will immediately stop working or become unusable. In reality, the TBW value is simply a guideline that manufacturers establish for warranty coverage. The statistical-based rating is not a definitive indicator of when the drive will fail. Contrary to popular belief, If I remember correctly did not Intel SSDs stop working when the TBW limit was reached."

• edzieba replied: "It would seem that the user created a workload that kept making cached writes to the SSD. So an entire article on NAND wear, and a mere hint of an implication that the '1 petabyte of writes' never actually touched the NAND: the user running the test created a tool that would send a write to the DRAM cache on the drive that would be flushed before the NAND was written. Physical wear from NAND erase cycles is irrelevant here, as no physical wear occurred in the first place."

• Jabberwocky79 replied: "I will need my NVMe drives to last this long before I can afford to replace them LOL"

• King_V replied: "Yeah, I was wondering about that - the still-shot on the video reads like it's saying 'no writes ever touched the NAND flash.'"

• IntelUser2000 replied: "I dislike these hype articles that really is irrelevant in reality for 99.9% of people. SSDs fail mostly because the electronics and the firmware fail, not due to overwrite on the NAND. My X25-M, Lenovo Thinkpad's M.2 SATA SSD, my brother's Adata SSD, all failed well, and far before it reached even 1/2 of rated endurance cycles. It probably wasn't even 10% for any of them. Electronics fail all the time. People shouldn't fool themselves into thinking getting an SSD with high write cycles will make it last 50+ years. And mostly likely they'll fail because they cheap out on the components and also push the spec for marketing teams and run them too hot. Most failures will happen in the 5-7 year range. Even devices that are properly designed and don't run hot fail sometimes too."