Windows10下TensorFlow-GPU安装失败求助

Hey there, I’ve run into this exact headache with Windows + Anaconda + TensorFlow-GPU before, so let’s break down how to get your setup working smoothly.

First, Let’s Pinpoint the Core Issue #

Your error stems from two main problems: a version mismatch between TensorFlow-GPU and your CUDA/CuDNN setup, plus potential missing system dependencies or misconfigured environment variables. Here’s how to fix each part step by step:

1. Install a TensorFlow-GPU Version Compatible with CUDA 8.0 #

When you run pip install tensorflow-gpu without specifying a version, it grabs the latest TensorFlow release—which no longer supports CUDA 8.0 (CUDA 8.0 works with TensorFlow 1.4.x to 1.10.x).

Activate your tensorflow-gpu environment and run this:

pip install tensorflow-gpu==1.10.0

This version is stable, supports Python 3.5/3.6, and plays perfectly with CUDA 8.0 + CuDNN 6.0.

2. Double-Check CUDA & CuDNN Setup #

Make sure your CuDNN version matches TensorFlow 1.10.0 (you need CuDNN 6.0 for CUDA 8.0):

• Download CuDNN 6.0 for CUDA 8.0 (you’ll need an NVIDIA developer account)
• Extract the zip file, then copy the contents directly to your CUDA 8.0 installation folder:
  • Copy cudnn64_6.dll from the extracted bin folder → C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\bin
  • Copy cudnn.h from extracted include → C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\include
  • Copy cudnn.lib from extracted lib\x64 → C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\lib\x64

Then update your system environment variables (restart your computer after this—critical for changes to take effect!):

• Add C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\bin to your PATH
• Add C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\lib\x64 to your PATH

3. Resolve Anaconda Environment Conflicts #

Sometimes Anaconda’s default packages clash with TensorFlow. Run this in your activated environment to align dependencies:

conda install cudatoolkit=8.0 cudnn=6.0

This ensures your environment has the exact CUDA/CuDNN versions TensorFlow expects, eliminating hidden mismatches.

4. Install Required Visual C++ Redistributable #

Windows needs the Visual C++ 2015/2017 Redistributable for TensorFlow’s DLLs to load properly. Download and install the x64 version of the Visual C++ Redistributable for Visual Studio 2017 (it’s backward-compatible with 2015 requirements).

5. Test Your Installation #

After all steps, activate your environment and fire up Python:

import tensorflow as tf
print(tf.__version__)
print(tf.test.is_gpu_available())

If everything works, you’ll see True for the GPU check—confirming TensorFlow is successfully using your GTX 970.

内容的提问来源于stack exchange，提问作者Gaetan.S