The last time Hackerfall tried to access this page, it returned a not found error. A cached version of the page is below, or clickhereto continue anyway

Install Keras + tensorflow-gpu with a NVidia Card | Deep Learning | The Geek Legacy

Hello, You must check the compute capability of you card on this link : en.wikipedia.org/wiki/CUDA

For the Nvidia Card NVS 5200M, your compute capability is only of 2.1. Unfortunately, that won't be enough to drive the tensorflow-gpu driver underneath Keras. Indeed tensorflow-gpu only supports a compute capability equal or superior to 3.0.

This information will save you hours of installating the NVidia driver and Visual Studio.

For others that have a greater capability or want to check if their cards will work with CUDA, here are the steps.

On windows 7/8 or 10 (64 bits), it won't work on a 32 bits version of windows 1) Check the name of your graphic card (for example with dxdiag on windows) 2) Check the compute capability on en.wikipedia.org/wiki/CUDA (your card must be on the list and have a compute capability >=3.0) 3) Install Visual Studio Studio 2010 or 2012 or 2012 or 2015 or Visual Studio Community 2015 (Visual Studio 2017 won't work) => check more on this link : docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/#axzz4nbYyqG4A 4) Install Cuda Toolkit 8.0 : developer.nvidia.com/cuda-downloads 5) Install tensorflow-gpu pip install --upgrade tensorflow-gpu You can find complementary informations on www.tensorflow.org/install/install_windows

I have not tried the installation on Ubuntu but I know that the step 2 applies as well and you have to work on a version linux 64 bits such as Ubuntu 64 bits (it won't work on a 32 bits version of Linux)

Hope it helps!

Continue reading on www.thegeeklegacy.com