search

keithito / tacotron

A TensorFlow implementation of Google's Tacotron speech synthesis with pre-trained model (unofficial)
MIT License
2.96k stars 957 forks source link

How to train faster? #217

Open yoosif0 opened 6 years ago

yoosif0 commented 6 years ago

I am currently training with 1080 GPU I think I would move to the cloud for faster training. Any recommendations? Anyone knows if the model is parallelizable (could run on more than one instance).

Thank you

Evil-Spirit commented 6 years ago

The same question. I have notebook GPU GeForce 650M. Now steps computed at 2.5 sec/step (~3000 step). Is it normal? May be I can go faster if I will use non-CUDA version of tensorflow? I mean, I have only 1gb video memory, so this can be bottleneck.

Rai220 commented 6 years ago

1080 takes about 1.750 sec/step, so I think it's normal.