search

ma-xu / pointMLP-pytorch

[ICLR 2022 poster] Official PyTorch implementation of "Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework"
Apache License 2.0
501 stars 65 forks source link

about multi gpu support #50

Closed Emenent758 closed 2 years ago

Emenent758 commented 2 years ago

Hello ! Thank you for really nice work. Please tell me how to run your code on a multi gpu machine? secondly, how much gpu memory is needed for segmentation's task? I have 12 gb gpu memory but when I put the model on training it gives cuda out of memory error right in the beginning, although the classification model trains fine on my machine, Please guide. Thanks

ma-xu commented 2 years ago

Hi, Thanks for your interest. I assume you can directly run on multi-gpu machine. Please let me know if any problems. Should be around 20 GB (not very sure). You can reduce the batch size to fit your gpu.

Emenent758 commented 2 years ago

when on try to run on multi gpu machine, it gets stucks both for classification and part segmentation, it doesnt show any error but doesnt proceed further. image for segmentation image Also please find the screenshot for memory usgae when model is made to run. image Please guide. Thank you for your time

ma-xu commented 2 years ago

@Emenent758 Sorry for the later reply. I have never met this issue previously. Does it work well on one gpu?

ma-xu commented 2 years ago

@Emenent758 I will close this issue since no further discussions. Feel free to reopen it if necessary.