ECC is not mandatory for Ceph, but I wouldn’t risk it.

No, it’s not required. Ceph has checksums, replication, and scrub, so non-ECC RAM can technically work. But you’re playing with fire.

Ceph’s checksums catch corruption in transit and on disk, but memory bit flips can corrupt data before it ever gets checksummed — in the RAM buffer where the data sits during reads/writes. ECC catches those flips; non-ECC silently passes them along. Replication won’t fix that because all copies might see the same corrupted data if the bit flip happens at the source node.

For a small homelab or test cluster, non-ECC is probably fine. But if you care about the data — even a little — spend the extra $20-50 on the motherboard and used server RAM. Used ECC DDR4 is cheap. Your uptime and integrity matter more than that.

Your data’s not worth the $50 savings.