How much memory of GPU is needed for training?

Question

How much memory of GPU is needed for training?

garrisonz opened this issue 2 years ago · 2 comments

I train Monohuman on a single 16G P100, but it fail with error "cuda out of memory".
May I know how much memory of GPU is needed when training. thanks

Answer 1 · 2023-05-24T07:10:58.000Z

around 30G. 32G v100 is good enough.

Answer 2 · 2023-05-26T05:58:42.000Z

I train Monohuman on a single 16G P100, but it fail with error "cuda out of memory". May I know how much memory of GPU is needed when training. thanks

Hi, if your CUDA memory is not enough, you can reduce the N_patches and size configs and increase sample_subject_ratio configs in https://github.com/Yzmblog/MonoHuman/blob/main/configs/default.yaml#LL132C1-L135C40 to fit your machine.