Multi-GPU support for training?

Question

Multi-GPU support for training?

Yurri-hub opened this issue 2 years ago · comments

I try to train a large-scale scene with large number of point clouds.
But I received a CUDA out of memory error at the start of training.

Is there any way to train the model on Multi-GPUs?

Artem Sevastopolsky · Answer 1 · Tue Sep 06 2022 16:09:36 GMT+0800 (China Standard Time)

Hi @Yurri-hub, yes, you can specify multigpu: True in train config or --multigpu in command line when running python train.py. The indices of the GPUs used can be specified via CUDA_VISIBLE_DEVICES="0,1,5" python train.py ..., where indices should be specified instead of "0,1,5". Also make sure to set the batch_size accordingly in order to avoid memory issues.

Yu Li · Answer 2 · Thu Sep 08 2022 16:53:41 GMT+0800 (China Standard Time)

Thank you for your reply. It helps a lot.