unexpected job interruptions (419 of 466 total) over a 54-day snapshot of Llama 3 405B pre-training on up to 16,384 H100s (~1 every 3 hr)
419observed
| 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 | Whole-job interruptions for one named run; ~78% of the unexpected ones were confirmed hardware issues. The other 47 were planned (automated maintenance such as firmware upgrades, operator-initiated configuration or dataset updates). Not per-GPU MTBF. |
| As of | 2024-07 |
| Source | Meta, 'The Llama 3 Herd of Models' (arXiv:2407.21783) · §3.3.4 Reliability, Table 5 |
| Review | checking…review by 2027-07-23 · standard cadence |
| Recorded changes | last 2026-10-15 · 4 revisions tracked |
| Claim id | unplanned-interruptions-over-54-days-on-16-384 |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.