secsilm / bifrost

Connect gpus with your eyes.

Bifrost: A Simple GPU Monitor

简体中文

A simple dash-based GPU monitor that uses web pages to continuously view GPU usage.

Features

• Use the webpage to continuously view basic GPU usage, including GPU usage, remaining graphics memory, temperature, and fan speed
• Check who is using the GPU, including user, process ID, process creation time, process creation commands, etc. For details, see the description below.

Dependencies

Bifrost relies mainly on the following:

• Python 3.6+
• dash
• psutil
• pandas
• numpy
• nvidia-ml-py3

You can install these dependencies with the following command:

pip install dash psutil pandas numpy nvidia-ml-py3

You can also use requirements.txt to install them:

pip install -r requirements.txt

Usage

python app.py

or using gunicorn:

gunicorn -c config.py app:server

Then open http://locahost:8150 or http://{your_ip}:8150 in your browser. Bazinga!

Field description

• gpu: GPU ID
• pid: The process ID
• used_gpu_mem: The GPU memory used by the process
• username: The owner of the process
• name: The Process name
• create_time: The process creation time
• status: The process status
• cpu_percent: CPU utilization of the process
• cpu_num: Which CPU this process is currently running on
• memory_percent: Memory utilization of the process
• num_threads: The number of threads currently used by this process
• cmdline: The command line this process has been called

LICENSE

MIT.

bifrost's Issues

The cpu_percent always return 0

For exzample,

The official document explains the reason:

When interval is 0.0 or None compares process times to system CPU times elapsed since last call, returning immediately. That means the first time this is called it will return a meaningless 0.0 value which you are supposed to ignore.