#3084 (None, 'RuntimeError: CUDA error: no kernel image is available for execution on the device\nCUDA kernel errors might be

This error indicates that the CUDA version installed on your system is incompatible with the version of PyTorch or the AI model library (likely faster-whisper) used by pyVideoTrans. The error "no kernel image is available for execution on the device" typically means the CUDA compute capability of your GPU is not supported by the compiled binaries.

The most likely cause is that you have an older NVIDIA GPU (compute capability < 5.0) and are trying to use software or models compiled for newer CUDA architectures (e.g., CUDA 12.x with sm_89). pyVideoTrans v3.96 is built to work with CUDA 12.8.

Follow these steps to resolve the issue:

1. Check your GPU's compute capability:

   • Open Command Prompt and run nvidia-smi. Note your GPU model.
   • Visit NVIDIA's CUDA GPU list to find your GPU's compute capability (e.g., 6.1 for GTX 10 series, 7.5 for RTX 20 series).

2. Verify your CUDA Toolkit version:

   • In Command Prompt, run nvcc -V. The output should show V12.8 or higher to be fully compatible with pyVideoTrans v3.96.

3. Solution based on your findings:

   • If your GPU is old (e.g., Maxwell architecture, compute capability 5.x or lower): The pre-built PyTorch/CTranslate2 binaries for CUDA 12.x may not include kernels for your GPU. You have two options:

     • Option A (Recommended): Force the software to use the CPU. In pyVideoTrans, go to Menu -> Settings -> Advanced Options -> General Settings. Find the option related to GPU/CPU usage (e.g., "Device" for faster-whisper) and set it to cpu. This will be slower but will work.
     • Option B: Try installing an older version of CUDA (like 11.8) and corresponding cuDNN, then attempt a source code installation of pyVideoTrans. This is complex and not guaranteed.
   • If nvcc -V shows a CUDA version older than 12.8: You need to upgrade to CUDA 12.8 and cuDNN 9.11+. Follow the installation guide for Windows.
   • If you have a very new GPU (e.g., RTX 50 series): You might need to wait for updated library binaries. As a temporary fix, use the CPU (Option A above).
4. Test after changes: After applying the fix (especially switching to CPU), restart pyVideoTrans and try the task again.

Please refer to the relevant documentation:

• windows上安装CUDA12.8和cuDNN9.11
• 常见错误与解决方法