Windows support, GPU temperature and numpy dep. removed

Added

• Added automatic detection of nvidia-smi on windows
• Added GPU temperature
• Added example for monitoring GPU usage from a separat thread

Removed

• Removed numpy dependency

Fixed

• Fixed crashing when GPUs returned by nvidia-smi
• Fixed potential deadlock when calling nvidia-smi
• Fixed issue with python v.2 unicode
• Fixed various spelling errors in readme

Supporting "not supported" fields and GPU exclusion

• Added support for GPU fields (such as memory and load), which were not supported by the GPU.
• Added UUID to GPUs.
• Added support for excluding GPUs based on ID or UUID
• Added __version__
• Moved main part of code from __init__.py to GPUtil.py
• Updated showUtilization() with increased flexibitlity.
• Updated readme

Correct GPU utilization

A minor, but very important bug has been fixed in regards to calculating the memory utilization, which is used to determined if a GPU is available. The bug meant that in Python 2.X the memory utilization would always show as 0%. Bug was not present in 3.X, as dividing two integers in 3.X are automatically converted to float.

Python 2 support added

Python 2 is now supported!

Renamed to GPUtil and ready for PyPI

Project have been renamed to GPUtil to avoid confusion with the slightly similar GPUstats python script here on github.
The project have also been prepared for PyPI, so that it can easily be installed and upgraded through pip install.

Wait for available GPU

GPUstats

• Added new functionality to getFirstAvailable(...) to support multiple attempts to find available GPU.
  • Added 4 new input arguments:
    • order - specifies, the ordering of the GPUs. Same functionality as for getAvailable(...)
    • attempts - number of attempts for locating available GPU.
    • interval - interval in seconds between each attempt.
    • verbose - print attempt counter

Improved documentation and examples

GPUstats

• Same python code as the initial release v1.0.0, but with improved documentation and examples have been added.

Initial release

GPUstats:

• Get memory usage and utility of all CUDA compatible GPUs using Python (and nvidia-smi)
• Get the first, last and random available GPU from all available GPUs
• Example showing how to programmatically only allocate memory on 1 GPU when using TensorFlow