synchronous-job overprovisioning cut when GPU availability rises 99.5% → 99.9% (4x)
20% → 5%modeled
| Value kind | modeled — Interpret this value according to its displayed kind, scope, source, and as-of date. |
|---|---|
| Scope | A projection in the Delta paper ('we project the impact of GPU node availability on larger-scales'), not a measured stocked-spare or inventory experiment. The 20%-to-5% step is a modeled availability sensitivity on a 1,056-GPU A100/H100 system; it does not establish that deeper spares caused the availability improvement. |
| As of | 2025 |
| Source | Characterizing GPU Resilience (arXiv 2503.11901) · Abstract; 5–20% overprovisioning and fourfold reduction at 99.9% GPU availability |
| Review | checking…review by 2026-11-09 · standard cadence |
| Recorded changes | last 2026-06-29 |
| Claim id | synchronous-job-overprovisioning-cut-when-gpu |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.