The latest solution for the installed version of tensorflow Gpu

I have been looking for various Chinese materials, but I finally died, and I couldn't find the Dll, and then I gnawed on the English documents in a rage. https://www.tensorflow.org/install/


The documentation clearly states that the installed cuda cuDNN does not meet the requirements and needs to be recompiled. But I don't want to compile because I'm lazy. I got a windfall when using the latest version of conda.


1. First install cuda 8 series. https://developer.nvidia.com/cuda-80-ga2-download-archive


The next step is fine.

2. Don't download cuDNN like most of the Internet says, just conda install tensorflow-gpu directly, you will see conda automatically download it for us.

3. Then directly experiment import tensorflow as tf in python, if no error is reported, it will be successful. The following figure has gpu information.

Guess you like

Origin http://10.200.1.11:23101/article/api/json?id=326676530&siteId=291194637