gpu4u 0.2.0

A Python Package for Automatically Monitoring & Occupying NVIDIA GPUs

A Python Package for Automatically Monitoring & Occupying NVIDIA GPUs

GPU4U locates all GPUs on the computer, determines their availablity and returns a ordered list of available GPUs. Availablity is based upon the current memory consumption and load of each GPU. The package is written with GPU selection for Deep Learning in mind, but it is not task/library specific and it can be applied to any task, where it may be useful to identify available GPUs.

Requirements

NVIDIA GPU with latest NVIDIA driver installed. GPU4U uses the program nvidia-smi to get the GPU status of all NVIDIA GPUs. nvidia-smi should be installed automatically, when you install your NVIDIA driver.

Python libraris:

• json
• os
• random
• re
• sys
• time
• datetime
• pynvml
• requests

Tested on CUDA driver version 450.102.04 with Python 3.6.10.

Installation

With PIP

pip install gpu4u

With Source Code

git clone https://github.com/imrdong/gpu4u.git
cd gpu4u
python setup.py install

Usage

Base

To combine GPU4U with your Python code, all you have to do is

• Open a terminal in a folder other than the GPU4U folder
• Start a python console by typing python in the terminal
• In the newly opened python console, type:

>>> from gpu4u import auto_monitor
>>> auto_monitor(script="fill_in_your_script_here")

WeChat

Given WeChat token, you can receive notifications from GPU4U in WeChat.

• Go to [PushPlus] official website, scan the QR code through WeChat to log in, and copy your token.
• The next steps are the same as Base section

>>> token = "paste_your_token_here"
>>> from gpu4u import auto_monitor
>>> auto_monitor(script="fill_in_your_script_here", use_wechat_push=True, token=token)

The outputs of GPU4U depending on your number of GPUs and their current usage, see Demo for more details.

Demo

:smile: Script Running with Available GPUs

# Find available GPUs
Find Available GPU: 0, 1, 2, 3. Start Running Your Script.

# Random select one GPU to run your script
Script: CUDA_VISIBLE_DEVICES=1 python train.py --batch_size 64

# The start time of script run
Started @: 2021-02-22 13:08:23

:disappointed: Script Running with No Available GPUs

# No available GPUs, start automatic monitoring with waiting time prompt
No Available GPU for Now, Automatic Monitoring for 0:23:10

# Find available GPUs
Find Available GPU: 2, 3. Start Running Your Script.

# Random select one GPU to run your script
Script: CUDA_VISIBLE_DEVICES=3 python train.py --batch_size 64

# The start time of script run
Started @: 2021-02-22 13:31:33

License

See LICENSE