Install Tensorflow-GPU with conda for GTX 1050 on Windows 10

Question

I am new to tensorflow and wanted some help installing it.

I have Miniconda3 with python 3.7.7.

I have 8GB RAM, i7-7700hq core and GTX-1050, 4GB RAM graphics card with Cuda driver version 9.2.209 version (I saw this from Nvidia control panel in NVCUDA.DLL)

I have tried to follow instructions on medium, youtube but to no use.

Can someone please help me setup tensorflow-gpu(preferably 2.1.0 but I'll be happy with any version as long as it is working) on my laptop?

Please feel free to ask any questions/information that you might need to answer.

Thank you :)

Answer 1

Version Checks-

• Check which GPU you are using (win+X) -> Device Manager -> Display Adapters
• Determine which version of Tensorflow you want to use.
• Determine CUDA version supported by tensorflow version from here. 2.1.0 supports 10.1

Prerequisites-

• CUDA enabled GPU (check here to check)
• Visual Studio. This is required to install GCC compiler. You can also go with Windows SDK but I have not tried that.

Steps-

1. Download CUDA toolkit for the required version here
2. Download latest cuDNN for your cuda version here (You have to register with Nvidia for this).
3. Install CUDA toolkit. It will automatically install the required Nvidia driver too.
4. Install cuDNN using these instructions

Edit - You need to ensure your graphics card has the necessary drivers to support CUDA version required by tensorflow. I had to go through docs of each driver to to this. (Check drivers for your GPU)