Elevated temps significantly impact the retention and are included in some datasheets by memory vendors, but often they are omitted and you need to request them.

Some earlier HN discussion here [0]

[0] https://news.ycombinator.com/item?id=35382252

This has inspired me to conduct my own drive longevity experiment. Thank for sharing.

Powered vs unpowered SSD, Hard drive, cheap vs expensive, NVME vs SD Card flash quality.

All good stuff to know for backups, and long term planning on building a computer, using a raspberry pi for projects, etc.

Does someone have a TL;DW ? I can't stand youtube videos.

(Universal Summarizer by Kagi)

Title: How Long Can SSD Store Data Unpowered? Year 2 Update

- The video provides a two-year update on SSD data retention tests conducted by HT Wingnut, focusing on SSDs left unpowered.

- Four inexpensive 128 GB TLC SSDs were tested: two were heavily used (over 280 TB written), while two were kept fresh.

- A folder with over 100 GB of random data was created for testing, with hash values generated for validation.

- The testing timeline was extended, with the initial data written on September 2, 2022, and the follow-up conducted on November 23, 2024.

- The fresh SSD maintained its integrity, with 100% data retention and no performance degradation after two years.

- The heavily used SSD showed significant corruption, with four files unaccounted for and a performance drop during read tests.

- The audit time for the well-worn SSD increased from about 10 minutes to over 42 minutes, indicating severe performance degradation.

- The fresh SSD's read performance remained stable at approximately 470 MB/s, while the worn SSD's performance was inconsistent and slower.

- Both SSDs showed some error recovery, but the heavily used SSD had 12 uncorrectable sectors after testing.

- The fresh SSD will be re-evaluated in one year, while the worn SSD's corruption raises concerns about long-term data integrity.

Might this be due to the fact (or at least something I've heard) that TLC SSDs maintain some initial portion as SLC for speed reasons?

Also I'm not sure simply keeping the drive powered on is sufficient to prevent degradation you'd have to actually read from the blocks in order for ECC and relocation to kick in. But maybe the SSD firmware does that in the background.