search

Purfview / whisper-standalone-win

Whisper & Faster-Whisper standalone executables for those who don't want to bother with Python.
1.16k stars 60 forks source link

Runs on CPU instead of GPU #296

Closed keith555 closed 1 month ago

keith555 commented 1 month ago

I installed it automatically via subtitle-edit. I noticed that transcriptions were running on the CPU, not the GPU. I ran it from the command line with the same results. I've tried several different models. The output says that it is "running on CUDA" but the GPU usage remains minimal. I tried explicitly setting cuda with --device: cuda. I also tried --device cuda:1 in case there was some problem with the integrated graphics card, but that failed with invalid device ordinal. If I transcribe the same file using Whisper Desktop, it runs on the GPU. The GPU is an RTX 4080 running on a Gigabyte Auros 17H.

Snag_8cef0a

Using Whisper Desktop: Snag_8d46af

Purfview commented 1 month ago

Compare transcription speed with --device cpu.

keith555 commented 1 month ago

I don't understand your previous comment. It runs much faster on the GPU. When I transcribed the file with Whisper Desktop which uses the GPU, it finished in under 10 minutes. When I ran it using this version, which runs on the CPU, it took almost an hour.

Purfview commented 1 month ago

I meant compare transcription speed of Standalone Faster-Whisper with --device cpu vs --device cuda. I'm not interested in Whisper Desktop, whatever that is.

keith555 commented 1 month ago

-cuda 110 seconds -cpu killed it after 30 minutes and it had transcribed three lines -Whisper Desktop 48 seconds

Purfview commented 1 month ago

-cuda 110 seconds -cpu killed it after 30 minutes and it had transcribed three lines

Obviously it runs on GPU.