GPU mode consumes half of available CPUs

[zucher]

on the version 0.4.1, CPU transcription consume less cpu than GPU +CPU.

with: docker run -it -p 9090:9090 ghcr.io/collabora/whisperlive-cpu:latest

With GPU: docker run -it --gpus all -p 9090:9090 ghcr.io/collabora/whisperlive-gpu:latest

The called model was french / small So I don't understand why ? And with such results why still using GPU ;-)

so seriously seems threads not sleeping correctly during GPU processing.

[Asma-droid]

I have the same issue