When using tools based on CTranslate2, such as faster-whisper or pyVideotrans, you may encounter errors like the following on 50-series GPUs:

cuBLAS failed with status CUBLAS_STATUS_NOT_SUPPORTED

err[msg]=_process:cuBLAS failed with status
CUBLAS_STATUS_NOT_SUPPORTED

This issue occurs because the underlying architecture of 50-series GPUs has changed. By default, CTranslate2 uses auto mode, which automatically selects the int8_float16 data type, which is not compatible with 50-series GPUs. Currently, you can resolve this by manually specifying the compute type.

1. In the command line, you can specify --compute_type='float16'
2. In the pyVideoTrans software, go to Menu → Tools → Advanced Options → Faster/OpenAI Speech Recognition Settings → CUDA Data Type → Select float16, then save and restart the software to resume normal operation.