I was using Linux at work and am used to install tensorflow GPU version via Conda.
The Linux machine has NVIDIA driver, and nvidia-smi runs properly.
The command I use is
conda install python tensorflow-gpu
Magically, conda takes care of everything. It installs cudatoolkit or whatever related tools. Once I run python in that virtual environment, GPU is usable.
Now I switch to work on Windows. The Windows box has NVIDIA driver, and nvidia-smi runs properly. Then I use the same conda command as I used on Linux.
conda install is successful but the python in the virtual env does not detect GPU.
How to install tensorflow 2 gpu on windows 10 with conda? says on Windows I must manually install cudnn and Cuda. Why conda on Windows does not take care of it as Linux one does?
Source: Windows Questions