Nezha Server Monitoring

English

简体中文

English

简体中文

Appearance

Enable GPU monitoring ​

GPU monitoring is a new feature implemented in Nezha Monitoring v0.17.x. Before using the feature, please ensure your Dashboard version is higher than v0.17.2 and Agent version is higher than v0.17.0.

Enable ​

From Command-Line Flag ​

Append the --gpu flag to the Agent argument. For example:

bash
/opt/nezha/agent/nezha-agent -s example.com:5555 -p example --gpu

From configuration file ​

Execute the following command to modify Agent configuration to enable GPU monitoring.

bash
/opt/nezha/agent/nezha-agent edit

In the interactive menu returned, choose to enable GPU monitoring.

Enable GPU utilization monitoring ​

GPU model and GPU utilization are two different monitor items, which uses different approaches to obtain their value.

Windows and macOS supports getting GPU utilization without extra dependencies, and support multiple graphics card brands.

Linux distros support only NVIDIA and AMD cards and need to install extra dependencies.

Below are the instructions on how to enable GPU utilization monitoring on Linux for NVIDIA / AMD graphics cards.

NVIDIA ​

NVIDIA cards need the nvidia-smi utility to get GPU utilization. This utility is included in the official driver by default.

If you use unofficial drivers like nouveau, then it's not possible to get GPU utilization.

AMD ​

AMD cards need to install the open source amdgpu driver and the rocm-smi utility.

Mainstream distros have already packaged rocm-smi, below are commands to install the utility on these distros:

bash
# Arch Linux
pacman -Sy rocm-smi-lib

# Debian / Ubuntu
apt install rocm-smi

# Fedora / RHEL 8+
dnf install rocm-smi

If your distro doesn't have the package, then you will need to compile rocm_smi_lib manually.

Required dependencies：git cmake gcc

First, clone the git repository of rocm_smi_lib:

bash
git clone https://github.com/ROCm/rocm_smi_lib

Then compile the libraries and install them on your system.

bash
cd rocm_smi_lib
mkdir -p build
cd build
cmake ..
make -j $(nproc)
# Install library file and header; default location is /opt/rocm
make install