{
  "tags": [],
  "title": "NVIDIA DCGM Exporter Dashboard",
  "widgets": [
    {
      "layout": {
        "x": 0,
        "y": 0,
        "width": 9,
        "height": 2
      },
      "definition": {
        "type": "timeseries",
        "title": "GPU Temperature",
        "yaxis": {
          "max": "auto",
          "min": "auto",
          "label": "",
          "scale": "linear",
          "include_zero": true
        },
        "requests": [
          {
            "style": {
              "palette": "dog_classic",
              "line_type": "solid",
              "line_width": "normal"
            },
            "queries": [
              {
                "name": "query1",
                "query": "avg:DCGM_FI_DEV_GPU_TEMP{$instance,$gpu} by {gpu}",
                "data_source": "metrics"
              }
            ],
            "formulas": [
              {
                "formula": "query1"
              }
            ],
            "display_type": "area",
            "response_format": "timeseries"
          }
        ],
        "show_legend": true,
        "legend_layout": "auto",
        "legend_columns": [
          "avg",
          "min",
          "max",
          "value",
          "sum"
        ]
      }
    },
    {
      "layout": {
        "x": 9,
        "y": 0,
        "width": 3,
        "height": 2
      },
      "definition": {
        "type": "query_value",
        "title": "GPU Avg. Temp",
        "requests": [
          {
            "queries": [
              {
                "name": "query1",
                "query": "avg:DCGM_FI_DEV_GPU_TEMP{$instance,$gpu}",
                "aggregator": "last",
                "data_source": "metrics"
              }
            ],
            "formulas": [
              {
                "formula": "query1"
              }
            ],
            "response_format": "scalar"
          }
        ],
        "autoscale": true,
        "precision": 2,
        "text_align": "center"
      }
    },
    {
      "layout": {
        "x": 0,
        "y": 2,
        "width": 9,
        "height": 2
      },
      "definition": {
        "type": "timeseries",
        "title": "GPU Power Usage",
        "yaxis": {
          "max": "auto",
          "min": "auto",
          "label": "",
          "scale": "linear",
          "include_zero": true
        },
        "requests": [
          {
            "style": {
              "palette": "dog_classic",
              "line_type": "solid",
              "line_width": "normal"
            },
            "queries": [
              {
                "name": "query1",
                "query": "avg:DCGM_FI_DEV_POWER_USAGE{$instance,$gpu} by {gpu}",
                "data_source": "metrics"
              }
            ],
            "formulas": [
              {
                "formula": "query1"
              }
            ],
            "display_type": "area",
            "response_format": "timeseries"
          }
        ],
        "show_legend": true,
        "legend_layout": "auto",
        "legend_columns": [
          "avg",
          "min",
          "max",
          "value",
          "sum"
        ]
      }
    },
    {
      "layout": {
        "x": 9,
        "y": 2,
        "width": 3,
        "height": 2
      },
      "definition": {
        "type": "query_value",
        "title": "GPU Power Total",
        "requests": [
          {
            "queries": [
              {
                "name": "query1",
                "query": "sum:DCGM_FI_DEV_POWER_USAGE{$instance,$gpu}",
                "aggregator": "last",
                "data_source": "metrics"
              }
            ],
            "formulas": [
              {
                "formula": "query1"
              }
            ],
            "response_format": "scalar"
          }
        ],
        "autoscale": true,
        "precision": 2,
        "text_align": "center"
      }
    },
    {
      "layout": {
        "x": 0,
        "y": 4,
        "width": 6,
        "height": 2
      },
      "definition": {
        "type": "timeseries",
        "title": "GPU SM Clocks",
        "yaxis": {
          "max": "auto",
          "min": "auto",
          "label": "",
          "scale": "linear",
          "include_zero": true
        },
        "requests": [
          {
            "style": {
              "palette": "dog_classic",
              "line_type": "solid",
              "line_width": "normal"
            },
            "queries": [
              {
                "name": "query1",
                "query": "avg:DCGM_FI_DEV_SM_CLOCK{$instance,$gpu} by {gpu}",
                "data_source": "metrics"
              }
            ],
            "formulas": [
              {
                "formula": "query1 * 1000000"
              }
            ],
            "display_type": "area",
            "response_format": "timeseries"
          }
        ],
        "show_legend": true,
        "legend_layout": "auto",
        "legend_columns": [
          "avg",
          "min",
          "max",
          "value",
          "sum"
        ]
      }
    },
    {
      "layout": {
        "x": 0,
        "y": 6,
        "width": 6,
        "height": 2
      },
      "definition": {
        "type": "timeseries",
        "title": "GPU Utilization",
        "yaxis": {
          "max": "auto",
          "min": "auto",
          "label": "",
          "scale": "linear",
          "include_zero": true
        },
        "requests": [
          {
            "style": {
              "palette": "dog_classic",
              "line_type": "solid",
              "line_width": "normal"
            },
            "queries": [
              {
                "name": "query1",
                "query": "avg:DCGM_FI_DEV_GPU_UTIL{$instance,$gpu} by {gpu}",
                "data_source": "metrics"
              }
            ],
            "formulas": [
              {
                "formula": "query1"
              }
            ],
            "display_type": "area",
            "response_format": "timeseries"
          }
        ],
        "show_legend": true,
        "legend_layout": "auto",
        "legend_columns": [
          "avg",
          "min",
          "max",
          "value",
          "sum"
        ]
      }
    },
    {
      "layout": {
        "x": 0,
        "y": 8,
        "width": 6,
        "height": 2
      },
      "definition": {
        "type": "timeseries",
        "title": "Tensor Core Utilization",
        "yaxis": {
          "max": "auto",
          "min": "auto",
          "label": "",
          "scale": "linear",
          "include_zero": true
        },
        "requests": [
          {
            "style": {
              "palette": "dog_classic",
              "line_type": "solid",
              "line_width": "normal"
            },
            "queries": [
              {
                "name": "query1",
                "query": "avg:DCGM_FI_PROF_PIPE_TENSOR_ACTIVE{$instance,$gpu} by {gpu}",
                "data_source": "metrics"
              }
            ],
            "formulas": [
              {
                "formula": "(query1) * 100"
              }
            ],
            "display_type": "area",
            "response_format": "timeseries"
          }
        ],
        "show_legend": true,
        "legend_layout": "auto",
        "legend_columns": [
          "avg",
          "min",
          "max",
          "value",
          "sum"
        ]
      }
    },
    {
      "layout": {
        "x": 0,
        "y": 10,
        "width": 6,
        "height": 2
      },
      "definition": {
        "type": "timeseries",
        "title": "GPU Framebuffer Mem Used",
        "yaxis": {
          "max": "auto",
          "min": "auto",
          "label": "",
          "scale": "linear",
          "include_zero": true
        },
        "requests": [
          {
            "style": {
              "palette": "dog_classic",
              "line_type": "solid",
              "line_width": "normal"
            },
            "queries": [
              {
                "name": "query1",
                "query": "avg:DCGM_FI_DEV_FB_USED{$instance,$gpu} by {gpu}",
                "data_source": "metrics"
              }
            ],
            "formulas": [
              {
                "formula": "query1"
              }
            ],
            "display_type": "area",
            "response_format": "timeseries"
          }
        ],
        "show_legend": true,
        "legend_layout": "auto",
        "legend_columns": [
          "avg",
          "min",
          "max",
          "value",
          "sum"
        ]
      }
    }
  ],
  "description": "Converted from Grafana dashboard https://grafana.com/grafana/dashboards/12239 (by nvidia). This dashboard is to display the metrics from DCGM Exporter on a Kubernetes (1.13+) cluster",
  "layout_type": "ordered",
  "reflow_type": "fixed",
  "template_variables": [
    {
      "name": "instance",
      "prefix": "instance",
      "default": "*",
      "available_values": []
    },
    {
      "name": "gpu",
      "prefix": "gpu",
      "default": "*",
      "available_values": []
    }
  ]
}