Why does used enterprise SSD pricing make no sense?

I am seeing the same flood in Singapore. Local broker on Carousell has trays of PM1733, same firmware, selling at SGD 460. Claims they came from a Vultr refresh out of their Singapore DC. I checked the serials against Samsung's warranty lookup, all show "End User: ODM Direct" which means no retail warranty transfer.

Bought two to test. One had 31,000 hours, the other 29,500. Both reformatted to 512 byte fine, both passed full read-write-verify. The third I opened had a stripped screw on the heatsink, obvious rework. Sent it back.

The consistent hours are the tell. Random e-waste would be a distribution. This is a single retirement event.

Running these in a Windows Server 2022 Storage Spaces Direct cluster. The PM1733s report temperature through NVMe-MI but Samsung's Windows driver does not surface it in WMI. Had to pull it through Intel's NVMe CLI tool instead, which feels wrong.

The -0000 firmware does not block format. The -0000 means no TCG Opal, no encryption. The -001B has self-encrypting but the key is already wiped. Either way you are fine for sector size changes.

Power loss capacitors on the v3 are replaceable if you are handy with a soldering iron. I have done two. Samsung part, 2.5V 330mF, about $4. Not saying you should, but you can.

That saves me a test cycle. Thank you.

Ordered the 24 yesterday. Seller in Phoenix, 18 in stock, had to split with another buyer for the last 6. Arriving Friday. Will run badblocks and format over the weekend. If anyone wants a follow-up with actual power numbers and fio results against the QLC alternative, I will post in the benchmarking subforum.

Still want to know the original owner. The coordinated retirement story makes sense but I would love confirmation. Vultr https://www.vultr.com, Contabo, someone else? The serial prefix KOI-2023Q2 does not match any of the known Samsung customer codes I have seen.

Before you benchmark, consider the verification gap. You are trusting the firmware to report its own health. A compromised firmware can lie about sectors, temperatures, power hours, everything. Your fio results will be valid for throughput but tell you nothing about integrity.

My recommendation stands: treat these as untrusted hardware. Separate management network, no IPMI to the scrubbing layer, verify writes with periodic hash checks against a known good. The $340 is not just a drive, it is a risk allocation.

I have no evidence these are implanted. I have no evidence they are clean. That is the point.