NodeRAYA reads the telemetry your GPUs already report and raises a rule finding only when a defined pattern persists. Separately, each GPU receives its own tamper-evident daily receipt after the day closes, backed by deterministic analysis and telemetry from the GPU itself.
Deterministic rules, inspectable logic, byte-level evidence. Fully read-only. It sits alongside your stack (Grafana, New Relic, Datadog) and runs fully automated.
Detects GPUs drawing excess power relative to their clock-aware mathematical physics baseline for 10+ minutes.
Reads the GPU's own firmware throttle bitmask directly. Real physics from the silicon, not an inference from room temperature.
At day-end, each GPU receives a daily record with a SHA-256 integrity hash. Receipts are retained permanently for later verification.
NodeRAYA reads telemetry your GPUs already emit, matches it against deterministic rule patterns, and raises a finding only when the pattern holds through the persistence window.
Reads metrics your GPUs already export: utilisation, power, SM clock, temperature, and the firmware throttle bitmask. Zero deployment on the silicon. Fully read-only.
Each rule is a deterministic predicate over a window. RULE_01 for compute inefficiency. RULE_02 for thermal throttling. PEER for cohort deviation. The rules are inspectable and reproducible.
A finding fires only when the pattern holds through the full persistence window. No single-sample triggers.
At the end of each day, NodeRAYA finalises one receipt for each GPU with recorded telemetry. It captures the daily telemetry summary, data-quality result, hardware status and SHA-256 integrity hash. Receipts are retained permanently in the daily archive.
Each receipt is finalised after the day ends, records one GPU’s daily telemetry and hardware status, carries an integrity hash, and is retained permanently.
At the close of each day, NodeRAYA finalises a receipt for every GPU with recorded telemetry. Each receipt preserves that GPU's daily record, data-quality assessment and hardware status, together with its integrity hash. Receipts are retained permanently and remain available for later review.
The pattern that fired. The thresholds that were in force. The exact telemetry rows behind it. Nothing is reconstructed after the fact.
NODE 12 · NVIDIA H100 SXM5 · HwThermalSlowdown held 60.0 min · 16:54 → 17:54 · config hash 1e39f4725cd2
Low utilisation sustained through the persistence window while observed power exceeded the clock-aware baseline for this GPU profile.
| Time | Util % | Power W |
|---|---|---|
| 20:00 | 6.4 | 483 |
| 20:05 | 6.6 | 484 |
| 20:10 | 6.9 | 486 |
| 20:15 | 6.7 | 485 |
| 20:20 | 6.8 | 487 |
| 20:25 | 6.5 | 484 |
| 20:30 | 6.9 | 485 |
| 20:35 | 6.6 | 483 |
| 20:40 | 6.7 | 485 |
| 20:45 | 6.8 | 486 |
| 20:50 | 6.5 | 484 |
| 20:55 | 6.9 | 485 |
| 21:00 | 6.7 | 484 |