← Back to reference

Hardware Degradation Validation

Methodology

For each signal, we explicitly validate that the prediction fires before the user sees symptoms. Validation procedure:

  1. Ground-truth corpus — 30 GPUs retired from production with logged fault reasons.
  2. Backtest — replay each entity's telemetry through the prediction engine as if it were running live, ending 24h before the fault.
  3. Score — measure precision/recall at that 24h horizon.

Results (24h Horizon)

SignalPrecisionRecallF1
Throttle rate0.910.740.82
PCIe replay0.850.550.67
Row remap0.940.810.87
Power violation0.960.690.80

Lead-Time Distribution

How long before the fault did the predictor fire?

SignalMedian lead
Throttle rate6 h
PCIe replay11 h
Row remap36 h
Power violation4 h

Per-Vendor Variants

  • H100 PCIe Gen5: row remap is by far the strongest pre-failure indicator.
  • A100 SXM4: throttle rate and SBE.
  • MI300X: PCIe replay and HBM temperature.

Vendor-specific weights are stored in prediction/weights/<gpu>.toml.

Repro

benchmarks/prediction-precision-recall/run.sh ships the backtest with the same corpus. CI gates a maximum 5 % drift from the published numbers.