PyCharm中无法导入TensorFlow-gpu 1.8.0的问题求助

Hey there, since you’ve confirmed the issue only pops up with your Anaconda-based PyCharm interpreter (and works fine in CMD/another non-Anaconda user account), let’s walk through targeted fixes to get this sorted:

1. Verify PyCharm’s Interpreter Environment Variables #

PyCharm might not be pulling in the system PATH for your Anaconda interpreter. Here’s how to check and adjust it:

• Open PyCharm, navigate to File > Settings > Project: [Your Project Name] > Python Interpreter
• Click the gear icon ⚙️ next to the interpreter dropdown, select Show All
• Locate your Anaconda interpreter (e.g., Python 3.6 (base: conda)), click the pencil icon ✏️ to edit its settings
• Switch to the Environment Variables tab
• Look for the PATH variable: if your cuDNN bin directory (where cudnn64_7.dll is stored) isn’t listed, click the + button to add it:
  • Variable name: PATH
  • Variable value: Paste your cuDNN bin path (e.g., C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\extras\CUDNN\v7.0\bin) followed by a semicolon ; and the existing PATH value
• Save changes, restart PyCharm, and test importing TensorFlow again.

2. Fix Anaconda’s Environment PATH Overrides #

Anaconda sometimes resets the PATH when activating an environment, which can hide your system’s cuDNN path. Try this:

• Open PyCharm’s built-in Terminal (bottom panel)
• Run echo %PATH% and check if your cuDNN bin directory is present in the output
• If it’s missing, open Anaconda Prompt, activate your environment with conda activate [your_env_name], then run:

  conda env config vars set PATH="C:\your\cudnn\bin\path;%PATH%"
  

• Restart PyCharm and reactivate the environment in the terminal to apply changes.

3. Quick Workaround: Copy the DLL to Anaconda’s Library Folder #

As a straightforward fix, copy cudnn64_7.dll from its current location to the Library\bin folder of your Anaconda environment:

• For the base environment: C:\Users\[Your Username]\Anaconda3\Library\bin
• For a custom environment: C:\Users\[Your Username]\Anaconda3\envs\[Your Env Name]\Library\bin
  This ensures the Anaconda interpreter can find the DLL directly without relying on PATH configurations.

4. Create a Clean Anaconda Environment #

If your current environment has conflicting packages or broken PATH settings, start fresh with a dedicated environment:

• Open Anaconda Prompt and create a new environment (TensorFlow 1.8.0 works best with Python 3.6):

  conda create -n tf_gpu_env python=3.6
  

• Activate the environment:

  conda activate tf_gpu_env
  

• Install TensorFlow-GPU 1.8.0:

  pip install tensorflow-gpu==1.8.0
  

• Back in PyCharm, switch your project interpreter to this new tf_gpu_env environment, then test importing TensorFlow.

内容的提问来源于stack exchange，提问作者GuPe