{
  "schema_version": 1,
  "name": "Quarkpad multi-GPU row record",
  "updated_at": "2026-08-17",
  "status": "OWNER ROW RECORD - NOT SHOP PROOF",
  "scope": "Manual row-level observations for two to four passive GPUs in one home AI system.",
  "card_limits": {
    "minimum": 2,
    "maximum": 4
  },
  "thresholds": {
    "minimum_duration_minutes": 20,
    "steady_window_minutes": 5,
    "minimum_steady_utilization_percent": 80
  },
  "outcomes": [
    {
      "key": "incomplete",
      "meaning": "Required row identity, setup, or measured values are missing or impossible."
    },
    {
      "key": "recheck",
      "meaning": "The manual row record is usable for support, but duration or one GPU load is too low for comparison."
    },
    {
      "key": "stop",
      "meaning": "A thermal slowdown, fan or airflow issue, or rising temperature on any GPU requires inspection."
    },
    {
      "key": "owner_recorded",
      "meaning": "The manual owner record contains the minimum row fields. It is not verified shop evidence."
    }
  ],
  "calculations": [
    "steady temperature rise above ambient for each GPU",
    "hottest steady GPU position",
    "steady temperature spread across the row",
    "total steady board power",
    "lowest steady GPU utilization"
  ],
  "stop_flags": [
    "thermal_slowdown_seen",
    "fan_cable_or_airflow_issue",
    "temperature_still_climbing"
  ],
  "limitations": [
    "Manual owner entry is not automated telemetry or a safety controller.",
    "A temperature spread is descriptive and does not prove why one position is warmer.",
    "Mixed boards, fans, shrouds, power limits, or workloads cannot be ranked as a cooling comparison.",
    "Only matched shop-controlled runs may advance the Quarkpad thermal proof status."
  ],
  "sources": [
    "https://docs.nvidia.com/deploy/nvidia-smi/index.html",
    "https://www.reddit.com/r/LocalLLaMA/comments/1v72s1f/built_a_system_with_four_p100_gpus/"
  ]
}