CUD tensorflow install_error ai_generated true

InternalError: CUDA driver version is insufficient for CUDA runtime version

ID: tensorflow/cuda-driver-version-mismatch

Also available as: JSON · Markdown · 中文
92%Fix Rate
90%Confidence
1Evidence
2024-05-20First Seen

Version Compatibility

VersionStatusIntroducedDeprecatedNotes
tensorflow 2.15 active — — —
tensorflow 2.16 active — — —
tensorflow 2.17 active — — —

Root Cause

The installed NVIDIA driver is older than the version required by the CUDA toolkit that TensorFlow was compiled against.

generic

中文

安装的 NVIDIA 驱动程序版本低于 TensorFlow 编译所依赖的 CUDA 工具包所需版本。

Official Documentation

https://www.tensorflow.org/install/gpu

Workarounds

  1. 92% success Update the NVIDIA driver to the required version. Check current driver with `nvidia-smi` and download the latest driver from https://www.nvidia.com/Download/index.aspx. For Ubuntu: `sudo apt update && sudo apt install nvidia-driver-545`. 
    Update the NVIDIA driver to the required version. Check current driver with `nvidia-smi` and download the latest driver from https://www.nvidia.com/Download/index.aspx. For Ubuntu: `sudo apt update && sudo apt install nvidia-driver-545`.
  2. 70% success Downgrade TensorFlow to a version compatible with the current driver. For example, use `pip install tensorflow==2.12.0` if the driver is older. 
    Downgrade TensorFlow to a version compatible with the current driver. For example, use `pip install tensorflow==2.12.0` if the driver is older.

中文步骤

  1. Update the NVIDIA driver to the required version. Check current driver with `nvidia-smi` and download the latest driver from https://www.nvidia.com/Download/index.aspx. For Ubuntu: `sudo apt update && sudo apt install nvidia-driver-545`.
  2. Downgrade TensorFlow to a version compatible with the current driver. For example, use `pip install tensorflow==2.12.0` if the driver is older.

Dead Ends

Common approaches that don't work:

  1. 95% fail

    Reinstalling TensorFlow alone does not update the NVIDIA driver; the driver is a system-level component.

  2. 80% fail

    Installing a newer CUDA toolkit via conda often fails because it does not update the kernel-mode driver.