Bug in Ryzen 3000 NVMe controller causes data loss

reddit.com

15 points·LeoNatan25··6 comments

>There's also a report of it happening on even a Haswell-E system and the common factor being NVMe on Windows 10 version 1903.

Where did you see this? I have not heard of this issue before the launch of the current Ryzen CPUs and x570 boards.

An AMD rep has responded claiming no data loss

https://www.reddit.com/r/Amd/comments/cd7pqb/warning_samsung...

> Due to the way NVMe writes, and the lack of power loss protection, it is absolutely possible to bork a few files in day-to-day operation. That's why enterprise-grade SSDs are built differently from consumer disks.

He’s actually trying to explain away the problem saying data loss is to be expected. Poor show, this device should be fixed in software or recalled.

Yeah. Data loss from "day to day operation" is complete and utter bull.

It's literally not "day to day" if you're losing data. It's "replace the device, then restore from backup".

The contract between a block device and a host is well known.

When a write is issued by the OS and acknowledged by the device, it must persist.

If write barriers are used, all writes before a barrier must persist before any writes after it.

You don't need battery backup to achieve this, but battery backup helps you get more performance while still meeting the requirements.

I should be able to cut power to the device a million times, and these basic requirements should still hold. If they do, great. If not, the device should be blocked by Microsoft (or at least a very scary warning on every bootup)

Reply on news.ycombinator.com