Activating Tensorflow with Intel GPU on Windows 10

Tensorflow provides many packages or libraries which can be used to develop and train ML models especially related to deep learning and neural network. To provide support to increasing the training time for the models while using its immensely computational resources consuming algorithms , it provides default support to run thier models on CUDA Graphics Drivers(GPUs) which are basically NVIDIA graphics drivers.

But , these days due to increasing enthusiam in ML , and data getting more and more complex, companies other than NVIDIA have also taken to add wrappers on the tensorflow packages which can take advantage of thier GPUs compational resources to reduce training time while creating ML models using tensorflow libraries.

So, if you wanted to use tensorflow gpu on your windows 10 machine , but have Intel Graphics card installed, you can still do that as Intel too now has a stable release of Intel oneAPI Math Kernel Library, also known as Intel MKL which has a open source performance library of optmized math routines and optimizations for deep learning. This is named as oneAPI-DNN which has a detailed installed guide on Optimization for Tensorflow Installation guide

Although above links has the detailed information , I would like to share a concise way to enable Intel GPU for tensorflow on windows 10 with anaconda installed.

1. Create a new environment - conda create env --name <envname> python=3.8.5
2. Activate the environment and install intel-tensorflow through pip
3. Run the sanity check program sanity check program given by Intel to check if the mkl libraries has been enabled on your machine.

Meanwhile , Microsoft has also released DirectML libraries in preview mode for enabling acceleration for Tensorflow,

Happy Learning .!!