tensorflow-gpu introduction memo
Introduction
④, ⑤, [⑥](https://qiita.com/ yohiro / items / e9e80183e635e0ac4894), ⑦ I tried to bite sklearn and svm, so I thought I'd try to bite TensorFlow this time, and its introduction memo. Fortunately, I own a GPU-equipped PC, so I'll try installing the GPU version of tensorflow.
Reference materials: [Image judgment AI application development / Part 1] Introduction to image judgment AI application development made with TensorFlow / Python / Flask
My environment
- Windows 10 --Anaconda3 installed --GeForce GTX 1060 installed
Implementation requirements
https://www.tensorflow.org/install/gpu?hl=ja
From the official, the requirements for tensorflow are as follows
It seems that once you have the GPU driver, you can install it with a single dictation command. So, first update the driver.
GPU driver update
First, start Device Manager, right-click the GPU that is the target of "Display Adapter" → click "Update Driver"
If you press "Automatically search for the latest version of driver software", it will be updated automatically.
Install tensorflow-gpu
Execute the following command from the Anaconda environment
conda install tensorflow-gpu
Confirmation of installation
From the Anaconda environment
python
After typing
import tensorflow
Enter. Then
2020-04-15 18:35:41.872948: I tensorflow/stream_executor/platform/default/dso_loader.cc:44] Successfully opened dynamic library cudart64_101.dll
Is displayed, and it seems that the installation was successful.
Where I got stuck
I was trying to install CUDA and cuDNN individually according to the teaching materials, but when I tried to install CUDA, I could not install it because of the same symptom as "Visual Studio Integration failed" as on the following page.
If the installation of "Visual Studio Integration" fails when installing CUDA on a Windows terminal
After all, it seems that the above command will automatically configure what is required for tensorflow-gpu, so CUDA and cuDNN have not been installed individually.
Recommended Posts