{
  "name": "nvbandwidth-all-gpu-power-watch",
  "description": "Full nvbandwidth across all GPUs at once (the crash never reproduces on a single socket alone), with IPMI sensors and GPU power/temp sampled every 2s to check for a power-delivery correlation.",
  "timeout_sec": 1800,
  "jobs": [
    {
      "name": "ipmi-sensors",
      "type": "sampler",
      "interval_sec": 2,
      "cmd": ["ipmitool", "sensor"]
    },
    {
      "name": "gpu-power",
      "type": "sampler",
      "interval_sec": 2,
      "cmd": ["nvidia-smi", "--query-gpu=index,power.draw,temperature.gpu,clocks.sm", "--format=csv"]
    },
    {
      "name": "nvbandwidth-all",
      "type": "command",
      "cmd": ["dcgmi", "diag", "-r", "nvbandwidth", "-i", "{{gpus}}"],
      "gpu_indices": [0, 1, 2, 3, 4, 5]
    }
  ]
}