AFR · Annualized Failure Rate
The fraction of a component population expected to fail per year; the fleet-planning number behind spares pools and maintenance budgets.
Current numbers
148 GPU + 72 HBM interruptionsobserved GPU- and HBM-attributed job interruptions in one 16,384-H100, 54-day Llama 3 run; not component AFR or per-node λ
148 GPU + 72 HBM interruptionsobserved interruption attribution in one 16,384-H100, 54-day Llama 3 run: 148 GPU + 72 HBM events; not equipment AFR