Very slow training on CPU

HarryVolek / PyTorch_Speaker_Verification

PyTorch implementation of "Generalized End-to-End Loss for Speaker Verification" by Wan, Li et al.

BSD 3-Clause "New" or "Revised" License

575 stars 165 forks source link

Very slow training on CPU #61

Open chrisspen opened 4 years ago

chrisspen commented 4 years ago

I'm training with the CPU backend for Torch with Python3.7 on anaconda with a quad-core CPU, and it seems very slow. After running for 12 hours, it had only completely 30 epoches out of the max of 950, and my loss had only gotten down to about 12. I estimate a full training would likely take an entire week. Is this normal without cuda? Is there anything I can do to speed up performance without getting a GPU?

MahmoudAliEng commented 4 years ago

Hi @chrisspen , I had the same issue here, did you found a solution ? Could you please tell me how did you solve it?

fmaker commented 4 years ago

Yes, that is normal with big models. You need a GPU.