search

lmnt-com / diffwave

DiffWave is a fast, high-quality neural vocoder and waveform synthesizer.
Apache License 2.0
754 stars 111 forks source link

Why need so many GPU to train the model(DiffWave base)? #35

Closed faceless-rex closed 2 years ago

faceless-rex commented 2 years ago

I follow the paper setting, using a batch size of 16 and a audio length of 16000. But 1 GPU is enough for me to train the model. Why does the original paper use 8 GPUs?

sharvil commented 2 years ago

You can use as many GPUs to train as you'd like. Multi-GPU training may be faster than single-GPU training.