#7170 CUDA GPU acceleration unavailable. Verify NVIDIA GPU & CUDA installation (Help -> CUDA Help).

This error means the software cannot detect a usable NVIDIA CUDA environment, so GPU acceleration is being disabled. Note that CUDA acceleration only works with NVIDIA graphics cards — AMD and Intel GPUs are not supported.

Also, your install path D:/win-pyvideotrans-v4.12-0908 (1) contains spaces and parentheses, which is not recommended. Move the software to a path with only plain English letters/numbers (e.g. D:/pyVideoTrans).

To enable CUDA, verify all of the following:

1. NVIDIA GPU only — Confirm you have an NVIDIA card. AMD/Intel GPUs cannot use CUDA.
2. CUDA Toolkit 12.8+ — Open CMD and run nvcc -V. The version must be 12.8 or higher (12.8+ and 13.x are compatible).
3. cuDNN 9.11+ — Install cuDNN 9.11 or newer. Both its bin and lib directories must be added to the system Path environment variable.
4. GPU driver — Run nvidia-smi in CMD to check the driver and the max CUDA version it supports. Update your NVIDIA driver to the latest version.
5. Environment variables — Make sure the CUDA bin and lib folders are correctly included in Path.
6. Signed-in check — After fixing, in the software select "Enable CUDA acceleration" (the "CUDA acceleration" checkbox in the ASR row).

Note: some channels (e.g. Qwen-ASR built-in, Qwen-TTS built-in, HuggingFace_ASR built-in, Hy-MT2) always try GPU first regardless of the checkbox, while others (e.g. FireRed3, zipvoice, denoise, vocal/BGM separation, punctuation restore) always use CPU and never use GPU.

If CUDA is still not detected after all of the above, the exact cause cannot be confirmed remotely.

Please refer to the following documents:

• Configure CUDA and cuDNN on Windows
• Install CUDA 12.6 and cuDNN 9.8 on Windows

'nvidia-smi' is not recognized as an internal or external command,
operable program or batch file

nvidia-smi is not recognized means the NVIDIA driver's tool is not in your system PATH — most likely the NVIDIA driver is not installed (or an old/corrupted driver), or you are using an AMD/Intel GPU. This directly explains why CUDA acceleration is unavailable.

Steps to confirm and fix:

1. Confirm you actually have an NVIDIA GPU.

   • Press Win + R, type dxdiag, press Enter, open the Display tab, and check the graphics card name.
   • If it is AMD or Intel (not NVIDIA), CUDA is impossible. Use CPU mode, or for AMD use the Whisper.net channel (source deployment required, advanced).

2. If you do have an NVIDIA card, install/update the driver:

   • Go to https://www.nvidia.com/Download/index.aspx, select your GPU model, download the latest Game Ready / Studio driver, install it, and reboot.
   • After reboot, open CMD and run nvidia-smi again. It should now output a table showing the driver version and the max CUDA version supported.
   • nvidia-smi is normally located at C:\Windows\System32\nvidia-smi.exe. If the file exists but the command still fails, add C:\Windows\System32 to your system PATH.

3. After the driver works, install the CUDA runtime components required:

   • CUDA Toolkit 12.8 or higher (run nvcc -V to verify).
   • cuDNN 9.11 or higher, and add both its bin and lib directories to the system Path.
4. Move the software to a plain path such as D:\pyVideoTrans (your current path D:/win-pyvideotrans-v4.12-0908 (1) contains spaces and parentheses, which is not recommended).
5. Restart the software and try enabling CUDA acceleration in the ASR row.

If you have an NVIDIA card, installed the latest driver, and nvidia-smi still fails, the specific cause cannot be determined. Please provide the exact GPU model (from dxdiag) and the log file under the software's logs/ folder (copy the last ~30 lines).

Please refer to the following documents:

• Configure CUDA and cuDNN on Windows
• Install CUDA 12.6 and cuDNN 9.8 on Windows

sử dụng chế độ CPU như thế nào nếu là intel.

Nếu bạn dùng GPU Intel, CUDA hoàn toàn không khả dụng, và phần mềm sẽ tự động chạy trên CPU — không cần bật bất kỳ tùy chọn nào.Để dùng chế độ CPU:Không tích chọn "启用CUDA加速 / Enable CUDA acceleration" (ô CUDA acceleration ở hàng nhận dạng giọng nói). Cứ để trống.Chọn mô hình nhỏ hơn để CPU chạy nhanh hơn: trong hàng nhận dạng chọn faster-whisper (内置), model nên chọn tiny / base / small thay vì large-v3 (large-v3 trên CPU rất chậm).Có thể vào 菜单 → 工具 → 高级选项 → 语音识别参数:将 计算数据类型 (Compute data type) đặt thành int8 để giảm tài nguyên và tăng tốc trên CPU.将 识别准确度beam_size 和 best_of giảm về 1 để nhanh hơn.Lưu ý về tốc độ: chạy thuần CPU sẽ chậm hơn nhiều so với GPU (khoảng gấp đôi hoặc hơn so với khi có CUDA). Một số kênh như 降噪, 分离人声背景声, 标点恢复 luôn chạy trên CPU (mô hình onnx) và không dùng GPU trong mọi trường hợp.Đối với GPU Intel không có cách nào bật tăng tốc phần cứng trong phần mềm này. Nếu có GPU AMD, bạn có thể dùng kênh Whisper.net (yêu cầu triển khai từ mã nguồn, phức tạp, không khuyến khích cho người mới).