Windows10 uses conda to install the gpu version of tensorflow2.0

Tip: Anaconda has been installed in advance

One, local configuration

  • System version: windows 10 64 bit
  • Graphics card: GeForce RTX 2060

For NVIDIV graphics cards, the CUDA computing power requirement is not less than 3.0. The
NVIDIA driver requires 410.x or higher.
CUDA version: 10.0
cudnn version number must be not less than 7.4.1

computing power. Refer to the
NVIDIA driver version to view: nvidia-smi(Command line Implementation)
NVIDIA driver download address

Two, install CUDA and cudnn

Because my NVIDIA driver meets the requirements, there is no update. Students who need to update can click the link above to download

1. Open Anaconda Prompt

2. Install CUDA

Command line input

conda install cudatoolkit=10.0.130

Recommended installation version: 10.0.130

3. Install cudnn

Command line input

conda install cudnn=7.6.0

Recommended version: 7.6.0

4. Install tensorflow

Command line input

conda install tensorflow-gpu=2.0.0

5. Check whether the installation is successful

Open python on the command line and enter

import tensorflow as tf
tf.test.is_gpu_available()

Guess you like

Origin blog.csdn.net/qq_45465526/article/details/108299866