CUDA out of memory
zhibaishouheilab opened this issue · 2 comments
Hello！
Thanks for your excellent work! I'm just learning the pytorch and unet network. When I ran your scripts I got the following error:
OutOfMemoryError: CUDA out of memory. Tried to allocate 226.00 MiB (GPU 0; 4.00 GiB total capacity; 3.24 GiB already allocated; 0 bytes free; 3.27 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
The error was got in the step of training. And there were no other process hobbling the gpu.
Thanks for your reply.
The provided information is too generic. It simply means your GPU memory is low and cannot fit the model and its parameters during the training process. You may try
(ordered list)
- lowering the batch size
- using a smaller number of features maps (e.g., 32 instead of 64 in
Conv2d-1; make subsequent changes accordingly) - reducing the input image size either by using patches or downscaling the entire input image
- upgrading to a better GPU with more memory
Good luck!