search

ark1234 / ICCV2023-HyperCD

The official repository of the paper "Hyperbolic Chamfer Distance for Point Cloud Completion" published at ICCV 2023
GNU General Public License v3.0
18 stars 3 forks source link

Traing Time Question #3

Open Jinpeng-Yu opened 11 months ago

Jinpeng-Yu commented 11 months ago

Hi, thanks for your great work, I found out that you use the seedformer as the backbone and DP (Dataparallel) as the default speedup, I would like to know how much time the training process takes. Because I really found that on my training machine (4 GPUs) it takes more time to use DP than just only single GPU, I don't know if you have encountered this problem, do you have any suggestions, thanks

ark1234 commented 6 months ago

Thanks for asking. I am sorry for the late response; We found multiple gpu training quicker than single in Nvidia quard 6000, but when you have one powerful GPU(like A200), it would be quicker since no need to take extra time on data transfer from different devices