#8260 CUDA GPU acceleration unavailable. Verify NVIDIA GPU & CUDA installation (Help -> CUDA Help).

CUDA acceleration in pyVideoTrans is not detected because the software uses its own CUDA runtime detection: it requires an NVIDIA GPU, a recent driver, CUDA Toolkit 12.8+, cuDNN 9.11+, and correct system PATH entries. Missing any one of these makes the "Enable CUDA?" option silently stay unusable.

Verify in this order:

1. Confirm the GPU is NVIDIA

   • Open CMD and run nvidia-smi.
   • If the command is not found or shows no NVIDIA device, CUDA cannot be used at all. AMD / Intel GPUs (including integrated graphics) are not supported.

2. Update the NVIDIA driver

   • Install the latest driver from nvidia.com, then reboot.
   • Run nvidia-smi again and check the "CUDA Version" shown in the top-right. It must be 12.8 or higher.

3. Install CUDA Toolkit 12.8 (or 12.8+ / 13.x)

   • Download from https://developer.nvidia.com/cuda-downloads
   • Verify with CMD: nvcc -V — it must print release 12.8 or above.
   • If nvcc is not recognized, the toolkit is not installed or its bin folder is not in PATH.

4. Install cuDNN 9.11 or higher

   • Download from https://developer.nvidia.com/cudnn (NVIDIA account required).
   • Copy the bin, include, lib folders into the CUDA Toolkit directory, or add the cuDNN bin and lib folders to the system PATH.

5. Check Windows environment variables

   • Open System Properties → Environment Variables → Path.

   • These entries must exist (adjust the version numbers to your install):

     • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8\bin
     • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.8\libnvvp
     • the cuDNN bin / lib folders
   • After editing PATH, fully close and restart pyVideoTrans (PATH is read only at process start).

6. Check the install location / permissions

   • C:/PyVideotrans is fine, but do not place the software under C:\Program Files or C:\Windows.
   • If detection still fails, right-click sp.exe → Run as administrator once.

7. If you have two GPUs (integrated + discrete)

   • The software may attach to GPU index 0. If that is the weak integrated GPU, CUDA init fails or OOM occurs. Force the NVIDIA GPU as the default in NVIDIA Control Panel → Manage 3D settings → Preferred graphics processor.

8. Also reduce memory pressure if CUDA loads but later crashes

   • Menu → Tools → Advanced Options → ASR Settings: set Compute data type to float16 (GPU) or int8, beam_size / best_of to 1, and disable Enable context awareness.
   • Advanced Options → General: keep GPU simultaneous tasks = 1.

Note: even with CUDA correctly installed, only some channels use it — faster-whisper / openai-whisper / Whisper.cpp / FunASR / Qwen-ASR / Huggingface_ASR / Qwen3-TTS / F5-TTS / OmniVoice / Confucius-TTS / ChatterBox / M2M100, etc. Noise reduction, vocal/BGM separation and punctuation restoration always run on CPU and will never show GPU usage.

Please refer to the related documentation:

• Configuring CUDA and cuDNN on Windows
• GPU acceleration and CUDA / VRAM errors (FAQ)