search

LansburyCH / relightable-nr

A Neural Rendering Framework for Free-Viewpoint Relighting (CVPR 2020)
Other
115 stars 25 forks source link

What is the recommended number of GPUs for training? #2

Closed TuotuoLi closed 4 years ago

TuotuoLi commented 4 years ago

Hi Lansbury Thanks for sharing the impressive work! I have running into the following error, any insight on this problem?

Traceback (most recent call last): File "train_rnr.py", line 304, in lp_stitch_resize = torch.from_numpy(lp_stitch_resize).to(device) RuntimeError: CUDA error: invalid device ordinal

Does the code requires more than one GPU?

Best, tuotuo

LansburyCH commented 4 years ago

Hi Tuotuo, The code can also work when there is only one GPU, but you need to change the training option from --gpu_id 0,1 to --gpu_id 0.

When using 2 GPUs, the gcn network is allocated on the first GPU while the rest are allocated on the second GPU. This is only necessary when the RAM of a single GPU is not enough.

TuotuoLi commented 4 years ago

That works, thanks for the help!!