#8786 CUDA GPU acceleration unavailable. Verify NVIDIA GPU & CUDA installation (Help -> CUDA Help).

The core cause is: the program did not detect a usable NVIDIA CUDA runtime environment — either there is no NVIDIA GPU, or the driver / CUDA 12.8 / cuDNN 9.x are not properly installed and the cuDNN bin directory is not on the system PATH.

Step-by-step fix:

1. Confirm the graphics card is NVIDIA. AMD and Intel GPUs (including Intel integrated graphics) do not support CUDA acceleration at all; in that case keep using CPU mode.
2. Download and install the latest NVIDIA driver from the official website and restart the computer. Drivers that are too old do not support CUDA 12.8.
3. Install CUDA Toolkit 12.8.
4. Install cuDNN 9.11, copy the bin and lib folders from the cuDNN package into the corresponding directories of the CUDA installation, and add the cuDNN bin path to the system environment variable Path.
5. Restart the software and check CUDA Acceleration on the main interface; then run Menu -> Help -> CUDA Help to confirm detection passes.
6. If the machine has both an integrated GPU and a discrete NVIDIA GPU, make sure the NVIDIA discrete card is the one being used (v4.18 already has adaptive detection and will prefer the GPU with the most free VRAM).
7. If it still fails, reduce the model size or switch Compute Type to int8 to rule out an out-of-VRAM false negative.

If none of the above restores it, the cause cannot be determined for now. Please provide the last ~30 lines of the latest log file in the logs/ folder under the software installation directory (especially the CUDA detection output) for analysis.

Please refer to the related documentation:

• Configuring CUDA and cuDNN on Windows