ciplab-NTIRE-2020 icon indicating copy to clipboard operation
ciplab-NTIRE-2020 copied to clipboard

Published 20 hours ago • verified publisher icon yhjo09

Training problem

Open xiaoyi-st opened this issue 4 years ago • 2 comments

I use a small data set and the running memory is large enough. Why does it show that the running memory is insufficient? RuntimeError: CUDA out of memory. Tried to allocate 384.00 MiB (GPU 0; 31.75 GiB total capacity; 30.61 GiB already allocated; 65.19 MiB free; 21.09 MiB cached)

By the way, can your network architecture be enlarged by 2x, 4x, 8x?

xiaoyi-st avatar Nov 19 '20 06:11 xiaoyi-st

I'm not sure about the GPU Memory. You need to provide more information. E.g. configuration.

Yes you can edit UPSCALE on https://github.com/kingsj0405/ciplab-NTIRE-2020/blob/master/train.py#L25

kingsj0405 avatar Jan 07 '21 10:01 kingsj0405

I'm not sure about the GPU Memory. You need to provide more information. E.g. configuration.

Yes you can edit UPSCALE on https://github.com/kingsj0405/ciplab-NTIRE-2020/blob/master/train.py#L25

my GPU Memory has 32510MiB, but the error is RuntimeError: CUDA out of memory. Tried to allocate 432.00 MiB (GPU 0; 31.75 GiB total capacity; 30.21 GiB already allocated; 304.69 MiB free; 30.23 GiB reserved in total by PyTorch).

xiaoyi-st avatar Jan 15 '21 08:01 xiaoyi-st