Nvidia SMI Exporter

Nvidia SMI Exporter for GPU statistics from nvidia-smi written in Go.

Supports multiple GPUs.

Building and running

Prerequisites:

• Go compiler

Building:

go get github.com/phstudy/nvidia_smi_exporter/src
cd ${GOPATH-$HOME/go}/src/github.com/phstudy/nvidia_smi_exporter
go build -v -o bin/nvidia_smi_exporter src/nvidia_smi_exporter.go

Running with default port:

bin/nvidia_smi_exporter

Running with specified port:

LISTEN_ADDRESS=9201 bin/nvidia_smi_exporter

Running in test mode:

TEST_MODE=1 bin/nvidia_smi_exporter

Using Nvidia Docker v2

The nvidia_smi_exporter depends on NVIDIA/nvidia-docker.

$ docker run -d \
      --runtime=nvidia \
      --name=nvidia_smi_exporter \
      -p 9202:9202 \
      study/nvidia_smi_exporter

You can check the output of nvidia_smi_exporter with:

$ docker logs -f nvidia_smi_exporter

Metrics Example

nvidia_driver_info{version="390.87"} 1
nvidia_device_count 1
nvidia_fanspeed{minor="0"} 0
nvidia_info{index="0", minor="0", uuid="GPU-7d57c83c-be3c-fc52-9fe2-6eac2185265a", name="GeForce GTX 1080 Ti"} 1
nvidia_memory_total{minor="0"} 11177
nvidia_memory_used{minor="0"} 0
nvidia_memory_free{minor="0"} 11177
nvidia_utilization_gpu{minor="0"} 3
nvidia_utilization_memory{minor="0"} 0
nvidia_temperatures{minor="0"} 55
nvidia_temperatures_max{minor="0"} 96
nvidia_temperatures_slow{minor="0"} 93
nvidia_power_usage{minor="0"} 25.69
nvidia_power_limit{minor="0"} 280.00
nvidia_clock_graphics{minor="0"} 139
nvidia_clock_graphics_max{minor="0"} 1999
nvidia_clock_sm{minor="0"} 139
nvidia_clock_sm_max{minor="0"} 1999
nvidia_clock_mem{minor="0"} 810
nvidia_clock_mem_max{minor="0"} 5505
nvidia_clock_video{minor="0"} 734
nvidia_clock_video_max{minor="0"} 1620

Grafana Dashboard

https://grafana.com/dashboards/6387