Resolve Cudart64_90.dll Missing Error: 4 Proven Solutions
Key Notes
- Ensure you install CUDA Toolkit version 9.
- Manually copy the missing DLL if necessary.
- Always check your PATH variables for accuracy.
How to Fix the Missing cudart64_90.dll Error in TensorFlow
Encountering the “Could not find cudart64_90.dll” error in TensorFlow can interrupt your development work. Understanding how to resolve this issue is critical for developers and data scientists alike.
What Should You Do If cudart64_90.dll is Missing?
Step 1: Install CUDA Toolkit 9
To start resolving the missing DLL issue, you will need to install the CUDA Toolkit version 9:
- Navigate to the CUDA Toolkit 9 download page.
- Select your version and operating system for download.
- Click Download for the Base Installer option and save it to your computer.
- After the download completes, run the setup file and follow the on-screen instructions.
Pro Tip: Some users have had success with the network installer over the offline version; if one fails, try the other.
Step 2: Manually Recover Missing Files
If the installation does not resolve the issue, consider manually copying the missing DLL file:
- Download and extract the CUDA Toolkit 9 files; do not run the setup file yet.
- Press Windows + R and enter %localappdata%.
- Navigate to
Temp\CUDA\cudart\bin. - Locate cudart64_90.dll within that folder and copy it.
- Then, go to
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.1\bin. - Paste the DLL file into this directory.
Pro Tip: The file path in Step 5 may vary based on your specific CUDA installation.
Step 3: Reinstall CUDA Toolkit 9
If you continue to experience issues, a complete reinstallation of CUDA Toolkit 9 could be necessary:
- Press Windows + X and select Apps and Features.
- Find CUDA Toolkit 9, click on the three dots next to it, and select Uninstall.
- Follow the uninstallation instructions that appear.
- Afterward, download the CUDA Toolkit 9 again and proceed with installation.
Pro Tip: Ensure you remove any other CUDA version, such as 9.1, before reinstalling.
Step 4: Update PATH Variables
Correct path settings are essential for CUDA to function properly:
- Verify that you have CUDA and cuDNN installed, specifically version 9.
- Press Windows + S and type “view advanced”.Choose View advanced system settings.
- Click on Environment Variables.
- Locate the Path variable for your user account, and add the paths from CUDA and CuDNN that contain cudart64_90.dll.
- Apply the changes.
- Also, update the System variables’ Path variable by adding the same paths.
- Save the changes to complete the process.
Pro Tip: If using a virtual environment, ensure to include these paths in your environment settings as well.
Additional Tips
- Always verify your backups before making significant changes.
- If issues persist, consider seeking support on relevant online forums.
- Keep your GPU drivers updated for optimal performance.
Summary
Resolving the cudart64_90.dll missing error involves installing and potentially reinstalling the correct version of CUDA Toolkit, manually copying required files, and ensuring that system PATH variables are properly set. These steps can help get your TensorFlow environment back up and running smoothly.
Conclusion
By following the outlined steps, you should be well-equipped to troubleshoot the cudart64_90.dll missing error in TensorFlow. Don’t hesitate to reach out to developer communities for additional support!
FAQ (Frequently Asked Questions)
What is cudart64_90.dll?
The cudart64_90.dll file is a Dynamic Link Library related to the CUDA framework, used by applications such as TensorFlow for GPU acceleration.
How do I know if I need to reinstall CUDA?
You should consider reinstalling CUDA if the missing DLL error persists after installation or if you encounter compatibility issues with TensorFlow.