Open ShinoNia opened 11 months ago
that's mine conifg/.
I'm not a rocket scientist... but
Why do you think CUDA CORE usage will appear as GPU utilisation? Aren't they completely different cores?
Somehow I had made a hack. Just clear the cache by adding code in read.py. it will automatically keep frreeing your gpu.
I expected this to use >90% of GPU and generate voices in crazy speed like 10 seconds of generating time per 1 second of audio. But it only uses about 10% of CPU and 0.3% of GPU and it takes about 7 minutes to generate a 3 seconds audio. I increased the batch size from 1 to 6 (I have 6gb of VRAM) and it did speed up the process, it uses 10% of CPU and 0.7% of GPU, takes about 3 minutes to generate 3 seconds of audio. However, I expected this to use more of GPU and faster. Is this normal or only me?
Sorry for bad English. Thanks in advance.
More info: