nvidia-docker-smi 0.1.0

A tool to monitor NVIDIA GPU usage across Docker containers

nvidia-docker-smi

A command-line tool to monitor NVIDIA GPU usage across Docker containers.

Installation

pip install nvidia-docker-smi

Usage

Simply run:

> nvidia-docker-smi

--- GPU Summary ---
GPU-0: Util: 0%, Mem: 6735M/11019M (Free: 4284M)
GPU-1: Util: 0%, Mem: 8M/11019M (Free: 11011M)
GPU-2: Util: 0%, Mem: 5938M/11019M (Free: 5081M)
GPU-3: Util: 0%, Mem: 8M/11019M (Free: 11011M)
-------------------

pid      user  proc             type  container             gpu_util  gpu_mem  cpu   mem   gpu
3405     root  Xorg             G     -                     0%        N/A      0.0%  0.0%  GPU-0
902386   root  python3          C     data_collector        0%        439M     0.2%  0.5%  GPU-0
2343459  root  api_server       C     api_server            0%        511M     0.1%  0.1%  GPU-0
2343966  root  python           C     ai_server             0%        5777M    1.9%  3.2%  GPU-0
3405     root  Xorg             G     -                     0%        N/A      0.0%  0.0%  GPU-1
3405     root  Xorg             G     -                     0%        N/A      0.0%  0.0%  GPU-2
1944839  root  api_server       C     api_server_caleb      0%        511M     0.0%  0.1%  GPU-2
1944937  root  python           C     ai_server_caleb       0%        5419M    0.1%  3.2%  GPU-2
3405     root  Xorg             G     -                     0%        N/A      0.0%  0.0%  GPU-3

Add --debug flag for performance profiling:

> nvidia-docker-smi --debug

...

--- Performance Profile ---
refresh_container_pid_map     :   0.0752s ( 18.4%)
get_gpu_memory_usage          :   0.0452s ( 11.1%)
get_gpu_stats                 :   0.0471s ( 11.6%)
get_ps_data                   :   0.1181s ( 29.0%)
get_container_map             :   0.0020s (  0.5%)
display_gpu_summary           :   0.0429s ( 10.5%)
prepare_data_rows             :   0.0737s ( 18.1%)
display_formatted_output      :   0.0036s (  0.9%)
Total execution time          :   0.4079s
---------------------------

Requirements

• Python 3.6 or higher
• NVIDIA GPU with installed drivers
• Docker

5. Install the package locally for development:

cd nvidia_docker_smi
pip install -e .