pyVideoTrans

English

简体中文

English

简体中文

Appearance

A Simple Guide to Fixing the cudaErrorNoKernelImageForDevice Error ​

When running the pyVideoTrans program, if you encounter a long and daunting error message like the one below:

Traceback (most recent call last):
File "videotrans\task\_only_one.py", line 43, in run
File "videotrans\task\trans_create.py", line 371, in recogn
File "videotrans\recognition\__init__.py", line 236, in run
File "videotrans\recognition\_base.py", line 75, in run
File "videotrans\recognition\_overall.py", line 155, in _exec
RuntimeError: err[msg]=_process:parallel_for failed: cudaErrorNoKernelImageForDevice: no kernel image is available for execution on the device

Don't worry, this is usually not a major issue. The key error message is: cudaErrorNoKernelImageForDevice: no kernel image is available for execution on the device

Core Reason: The PyTorch library embedded in your current software is incompatible with your computer's graphics card environment (NVIDIA driver or CUDA version). It's like trying to open a new lock with an old key—they simply can't work together.

Solutions: Finding the Right Fix ​

Please choose one of the following two solutions based on your specific situation.

Case 1: You Upgraded the Software Using a "Small Patch Package" ​

Full pyVideoTrans download link: https://pyvideotrans.com/downpackage

If you upgraded from an older version below v3.76 by downloading a small patch package (e.g., around 360MB) and overwriting the installation, you have likely run into this issue.

  • Reason: To reduce the update package size, the small patch may not include the latest AI computing core libraries that are fully compatible with newer drivers.

  • Solution: Please visit the official website or download page, download the latest full installation package (approx. 2.5GB), and then overwrite the installation again. The full package contains all the necessary components to ensure a perfect match with your graphics card environment.

Case 2: You Are Already Using the Full Package or Encounter the Error on a Fresh Install ​

If you are certain you are using the latest full package but the problem persists, it's highly likely that your computer's graphics card environment needs an update.

Step 1: Check Your Graphics Card Environment ​

  1. Press the Win + R keys on your keyboard to open the 'Run' dialog box.
  2. Type cmd and press Enter to open the Command Prompt window.
  3. In the window, type the following command and press Enter:
    bash
    nvidia-smi
    This command shows the highest CUDA version supported by your NVIDIA graphics card driver. Pay close attention to the CUDA Version in the top-right corner of the output.

  1. If you have the CUDA Toolkit installed, you can also check its version with the nvcc --version command. However, for solving this particular issue, the information from nvidia-smi is more critical.

(Example output of nvcc --version)

Step 2: Upgrade Your Graphics Card Environment ​

If the nvidia-smi command shows your CUDA Version is below 12.0, or if nvcc --version shows a CUDA version below 12.0, you need to upgrade.

  • Solution:
    1. Update NVIDIA Graphics Driver: Go to the official NVIDIA website, download, and install the latest driver for your graphics card model. This is the most crucial step.
    2. Install Matching CUDA and cuDNN: The software relies on the correct CUDA Toolkit and cuDNN libraries to function properly.

To ensure a smooth installation, please follow the step-by-step tutorial at the link below:

Detailed GPU Environment Installation Tutorial ​

The cudaErrorNoKernelImageForDevice error is a clear signal of a "version mismatch."

  • First, check the software: Did you upgrade using a small patch? If so, switch to the full package.
  • Then, check your computer: If the software is fine, upgrade your graphics driver and related components.