Meta’s July 2025 discussion describes about one SDC fault per thousand devices without a stated exposure period
~1 fault per 1,000 devices; exposure unstatedobserved
| Value kind | observed — Reported measurements, counts, and specifications keep the precision and scope stated by their source; an exact specification is not treated as a range. |
|---|---|
| Scope | Historical statement in Meta’s AI-hardware reliability discussion; no time denominator is supplied. Not a measured annual incidence, per-job cadence or proof that 0.1% of GPUs currently harbor a defect. |
| As of | 2025-07-22 |
| Source | Meta Engineering, How Meta keeps its AI hardware reliable, 22 July 2025. · Silent data corruption: contrast with soft-error-related bit flips. |
| Review | checking…review by 2027-03-05 · standard cadence |
| Recorded changes | last 2026-09-16 · 2 revisions tracked |
| Claim id | machines-affected-by-silent-data-corruption-sdc |
Where the guide uses it
- 10.6 Observability, Telemetry & GPU Health
- 13.8 GPU Node Burn-In, Diagnostics & Stress Validation
- 14.1 Operational KPIs, Goodput & the Reliability Economics of AI Factories
- 14.3 Component Failure Modes, Failure Rates & Fleet Reliability Data
← Full numbers register — every date-stamped figure in the guide, with revision history.