Independent Research
We've written 28 petabytes to one SSD. It still hasn't failed.
An independent, unsponsored endurance test run from our Belfast laboratory — Samsung's 990 Pro, 990 EVO and 870 QVO against budget Chinese NVMe SSDs, streamed live and pushed until something breaks.
The experiment
Every SSD has a rated TBW figure. We wanted to know what that number is actually worth.
Manufacturer endurance ratings are conservative by design, and most reviews only measure how fast a drive performs on day one. We're measuring what happens years — and petabytes — later.
At Data Forensics Ltd in Belfast we're running one of the most extensive endurance tests attempted on consumer SSDs: Samsung's premium NVMe and SATA lineup against budget Chinese alternatives, written to continuously, power-cycled regularly, and monitored on camera for as long as the drives survive.
Every drive was purchased directly by the lab. This is not a vendor-sponsored benchmark — no free samples, no input from Samsung or anyone else into which drives were chosen or how they were tested.
The headline figures so far: the Samsung 990 Pro has passed 28 petabytes written — nearly 50 times its official 600TBW rating — with no SMART errors or warnings. The Samsung 990 EVO has passed 20 times its own 600TBW rating. Both drives are still fully functional.
Under test
Three Samsung drives, three budget NVMe challengers.
Samsung 990 Pro (1TB NVMe)
Rated 600TBW. Now past 28 petabytes written — around 50x its rating — with sustained write speeds still above 900MB/s and no SMART warnings.
Samsung 990 EVO (1TB NVMe)
Also rated 600TBW. Past 20x that figure. Less consistent than the 990 Pro, including a "bad drive" warning at roughly 16TB written that cleared after a firmware reset.
Samsung 870 QVO (1TB SATA)
Rated 360TBW. The weakest performer of the three — sustained writes drop to around 80MB/s once the cache is exhausted, slower than many mechanical drives.
KingSpec (1TB NVMe, Gen3x4)
The cheapest of the budget challengers. Failed outright at just 200GB written.
Fikwot M.2 (1TB NVMe, Gen4x4)
Started strong but drops below 100MB/s under sustained heavy load.
Fanxiang S880E (1TB NVMe, Gen4x4)
Marketed as a performance drive; inconsistent in practice and not enterprise-ready.
How we test
Built to reflect real usage, not a marketing benchmark.
- 01
Continuous write cycles
Each drive is filled with random data, then wiped, repeatedly — simulating heavy real-world workloads like video rendering or database operations rather than a single sequential pass.
- 02
Random data, random blocks
Random data bypasses the drive's cache and forces writes directly to NAND, distributed across the entire drive surface rather than sequentially.
- 03
Power cycles every 20TB
No PC runs forever. Regular shutdowns stress SSD firmware and reflect everyday use, rather than months of uninterrupted writing that no real system experiences.
- 04
Continuous SMART monitoring
HDD Sentinel tracks write speeds, temperatures, bad blocks and health percentages throughout, with results streamed live and daily SMART logs published for download.
What we've learned so far
A wide gap between premium and budget drives — and endurance that varies even within a model.
The Samsung 990 Pro is the standout so far: even at 28 petabytes written it is still delivering sustained write speeds above 900MB/s, with no errors or SMART warnings. A separate, previously used 990 Pro brought into the test had already endured over 900 power cycles and 22,000TB written before this round began, and was still outperforming brand-new budget drives.
The Samsung 990 EVO has also comfortably exceeded its rating, though less smoothly than the Pro — it threw a "bad drive" warning at around 16TB written, which cleared after a firmware reset and has not recurred.
The Samsung 870 QVO, a budget SATA drive, is the weak link of the Samsung range under sustained writes: once its cache is exhausted, throughput drops to around 80MB/s.
Of the budget NVMe challengers, KingSpec failed outright after only 200GB written. Fikwot and Fanxiang are both still running, but neither sustains anywhere near their advertised speeds under continuous load.
The clearest lesson so far is that official TBW ratings are conservative, but also that endurance is not purely down to brand — two drives from the same batch can behave very differently, the same way two chips from the same CPU line can overclock very differently. Premium NAND and controller engineering narrow that variance considerably; budget hardware does not.
Every terabyte written here teaches us more about how Samsung and budget SSDs actually degrade — knowledge that goes directly into every SSD recovery case that reaches the lab.