Whisper.AI on ASUS ROG 16 with RTX 3060 – Jupyter Kernel Crashes on CUDA Execution

[SamarSyed]

Hi everyone,

I recently installed Whisper.AI in a fresh Anaconda environment on my ASUS ROG 16 laptop with the following specs:

CPU: AMD Ryzen 9

GPU: NVIDIA RTX 3060 (laptop)

OS: Windows 11

CUDA & PyTorch: properly installed, CUDA is recognized

Whisper / WhisperX: installed via pip in a new conda environment

Environment: running inside Jupyter Notebook

The installation itself went smoothly, and the model downloads correctly. The command torch.cuda.is_available() returns True, and the GPU is properly detected. However, as soon as I start transcription using device='cuda', the Jupyter kernel crashes immediately, with no meaningful error message.

Details:

Using device='cpu' works without issue.

The crash happens right when Whisper attempts to run on the GPU (not during import or setup).

Even when trying smaller models (tiny, base), the same crash occurs.

nvidia-smi detects the GPU, and there are no memory issues before starting the task.

Things I've already tried:

Verified CUDA and torch versions are compatible (torch 2.1, CUDA 11.8)

Tried reinstalling the environment and dependencies

GPU driver is up to date

No other heavy GPU processes running

Also happens in WhisperX

Question:
Has anyone experienced similar crashes with RTX 30-series GPUs on laptops when using Whisper with CUDA?
Is this related to a driver or Jupyter issue on Windows? Any tips on how to debug this more effectively (e.g. view core dumps, detailed logs) or what could be the problem here?

Thanks in advance for any suggestions or ideas!

Best regards

[phineas-pta]

did u try to run in python/ipython console ? and in cli ?