SDC · Silent Data Corruption
Errors that corrupt computation without any alert; at fleet scale they silently spoil training and must be hunted.
Current numbers
~1 fault per 1,000 devices; exposure unstatedMeta reported about one SDC fault per 1,000 devices; exposure period unstated (22 July 2025)
~2.5 billionSDC test seeds per month run by Ripple across Meta's CPU fleet (2022; Fleetscanner counted separately)