#3943 parallel_for failed: cudaErrorNoKernelImageForDevice: no kernel image is available for execution on the device

This error indicates that the CUDA kernel images are not compatible with your specific NVIDIA GPU model. The software's built-in CUDA 12.8 libraries do not contain a kernel compiled for your GPU's compute capability.

To resolve this, you need to install the correct CUDA Toolkit and cuDNN version that matches your GPU's compute capability and are compatible with the software.

1. Check your GPU's compute capability:

   • Open Command Prompt (CMD) and run nvidia-smi. Note your GPU model (e.g., RTX 3060).
   • Search online for "[Your GPU Model] compute capability" (e.g., "RTX 3060 compute capability").

2. Install the correct CUDA Toolkit:

   • The software is built with CUDA 12.8. You must install CUDA Toolkit 12.8 from the official NVIDIA website.
   • During installation, ensure you select the component "Development" (which includes the necessary runtime libraries).

3. Install the matching cuDNN:

   • Download cuDNN v9.1.1 (or a compatible 9.x version) for CUDA 12.8 from the NVIDIA developer website.
   • Extract the cuDNN files and copy them into your CUDA installation directory (e.g., C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8), merging the bin, include, and lib folders.

4. Verify the installation and environment variables:

   • In CMD, run nvcc -V to confirm CUDA 12.8 is installed.
   • Ensure your system's PATH environment variable includes the paths to the CUDA bin and lib directories (e.g., C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8\bin).
5. Restart your computer and try running pyVideoTrans again with "CUDA加速" checked.

If the error persists after these steps, your GPU might be too old (with a very low compute capability) and not supported by the CUDA 12.8 binaries. In that case, you cannot use GPU acceleration and must rely on CPU processing, which will be significantly slower.

Please refer to the relevant documentation:

• Windows上安装CUDA12.8和cuDNN9.11