GPU energy - NVIDIA SMI - Component

What it does #

This metric provider gets the current GPU power draw from the NVIDIA SMI software.

Classname #

• GpuEnergyNvidiaSmiComponentProvider

Metric Name #

• gpu_energy_nvidia_smi_component

Prerequisites & Installation #

You first must install the CUDA Toolkit from NVIDIA for the metric provider to have the needed libraries and binars. The URL at the time of writing is here: https://developer.nvidia.com/cuda-downloads

You need both:

• Base Installer
• Driver Installer

To check if the installation has succeeded you can run:

$ nvidia-smi -q

After the installation you system can use language bindings for your matching CUDA version. Please check on our Measurement Cluster page which CUDA version is installed.

Debugging #

If you cannot generate any output you should first check if your GPU is supported by NVIDIA CUDA on their list for CUDA support.

Then you should check if the kernel module was corretly loaded with dmesg. Sometimes a message like this appears:

The NVIDIA GPU 0000:01:00.0 (PCI ID: 10de:1081)
NVRM: installed in this system is not supported by open
NVRM: nvidia.ko because it does not include the required GPU
NVRM: System Processor (GSP).

In this case you should switch to the legacy kernel module

Check in sudo dmesg if the kernel module could correctly be lodaded and then verify through cat /proc/driver/nvidia/version. See also details on the NVIDIA support page

Input Parameters #

• args
  • -i: interval in milliseconds

By default the measurement interval is 100 ms.

./metric-provider-nvidia-smi-wrapper.sh -i 100

Output #

This metric provider prints to Stdout a continuous stream of data. The format of the data is as follows:

TIMESTAMP READING

Where:

• TIMESTAMP: Unix timestamp, in microseconds
• READING: The energy used by the GPU in milliWatts (Ex: 12230 for 12.23 Watts)

Any errors are printed to Stderr.

How it works #

The provider uses the nvidia-smi tool to read the data.