cpu resources costs

[leocheung54]

Checks

• This template is only for usage issues encountered.
• I have thoroughly reviewed the project documentation but couldn't find information to solve my problem.
• I have searched for existing issues, including closed ones, and couldn't find a solution.
• I confirm that I am using English to submit this report in order to facilitate communication.

Environment Details

a100-gpu cuda12.2

Steps to Reproduce

Why does using GPU for inference consume a lot of CPU resources, even though the printed device in load_checkpoint and load_vocoder is 'cuda'?

✔️ Expected Behavior

load model for once, and load wav_scp one line by one, each run infer_process

❌ Actual Behavior

it costs much 1000%cpu resources for 1 thread

[SWivid]

how is gpu usage

[ewwink]

make sure installation step for the following is succeed

# NVIDIA GPU: install pytorch with your CUDA version, e.g.
pip install torch==2.3.0+cu118 torchaudio==2.3.0+cu118 --extra-index-url https://download.pytorch.org/whl/cu118

# AMD GPU: install pytorch with your ROCm version, e.g.
pip install torch==2.5.1+rocm6.2 torchaudio==2.5.1+rocm6.2 --extra-index-url https://download.pytorch.org/whl/rocm6.2

then use the following command to check whether cuda is detected

python -c "import torch; print('is CUDA:', torch.cuda.is_available())"

[leocheung54]

how is gpu usage

gpu memory usage is ~3GB

[leocheung54]

make sure installation step for the following is succeed

# NVIDIA GPU: install pytorch with your CUDA version, e.g.
pip install torch==2.3.0+cu118 torchaudio==2.3.0+cu118 --extra-index-url https://download.pytorch.org/whl/cu118

# AMD GPU: install pytorch with your ROCm version, e.g.
pip install torch==2.5.1+rocm6.2 torchaudio==2.5.1+rocm6.2 --extra-index-url https://download.pytorch.org/whl/rocm6.2

then use the following command to check whether cuda is detected

python -c "import torch; print('is CUDA:', torch.cuda.is_available())"

yes, im pretty sure : torch.cuda.is_avaliable() is True