GPU Exporter

This is a Prometheus Exporter for Collect Nvidia GPU metrics. It uses the Nvidia NVML Go Bindings, not call the nvidia-smi binary.

Build

images

make docker

Pre required

# ls -l /lib/x86_64-linux-gnu | grep libdl
-rw-r--r-- 1 root root   14640 Jan 14  2018 libdl-2.24.so
lrwxrwxrwx 1 root root      13 Jan 14  2018 libdl.so.2 -> libdl-2.24.so

Running

./gpu-exporter run -v 5 --logtostderr

# HELP nvidia_gpu_memory_total_mb Memory Total of the GPU device in MB
# TYPE nvidia_gpu_memory_total_mb gauge
nvidia_gpu_memory_total_mb{name="Tesla P4",path="/dev/nvidia0",uuid="GPU-9bb57e8e-f94d-4b95-7c33-ee279c8fb75c"} 7606
nvidia_gpu_memory_total_mb{name="Tesla P4",path="/dev/nvidia1",uuid="GPU-2d9067e4-8fab-5820-6228-69e2e74b3d58"} 7606

# HELP nvidia_gpu_memory_used_mb Memory used by the GPU device in MB
# TYPE nvidia_gpu_memory_used_mb gauge
nvidia_gpu_memory_used_mb{name="Tesla P4",path="/dev/nvidia0",uuid="GPU-9bb57e8e-f94d-4b95-7c33-ee279c8fb75c"} 0
nvidia_gpu_memory_used_mb{name="Tesla P4",path="/dev/nvidia1",uuid="GPU-2d9067e4-8fab-5820-6228-69e2e74b3d58"} 0

# HELP nvidia_gpu_num_devices Number of Nvidia GPU devices
# TYPE nvidia_gpu_num_devices gauge
nvidia_gpu_num_devices 2
# HELP nvidia_gpu_power_usage Power usage of the GPU device in watts
# TYPE nvidia_gpu_power_usage gauge
nvidia_gpu_power_usage{name="Tesla P4",path="/dev/nvidia0",uuid="GPU-9bb57e8e-f94d-4b95-7c33-ee279c8fb75c"} 7
nvidia_gpu_power_usage{name="Tesla P4",path="/dev/nvidia1",uuid="GPU-2d9067e4-8fab-5820-6228-69e2e74b3d58"} 6

# HELP nvidia_gpu_temperature Temperature of the GPU device in celsius
# TYPE nvidia_gpu_temperature gauge
nvidia_gpu_temperature{name="Tesla P4",path="/dev/nvidia0",uuid="GPU-9bb57e8e-f94d-4b95-7c33-ee279c8fb75c"} 45
nvidia_gpu_temperature{name="Tesla P4",path="/dev/nvidia1",uuid="GPU-2d9067e4-8fab-5820-6228-69e2e74b3d58"} 39