0

I am trying to train a network on images (48x48px), where I get the following error:

enter image description here

I am training on CloudML with Nvidia_k80.

It just seems unreasonable that 48x48 px blocks run out of memory on 11gb (I have ~300 conv layers with around 64 filters each).

GRS
  • 2,807
  • 4
  • 34
  • 72
  • @paddycakes I tried setting ` gpu_options = GPUOptions(allow_growth=True)`, and other config changes without success. I don't think anyone will be able to solve this issue, however. – GRS Mar 27 '19 at 21:25
  • I suggest you to optimize your network to reduce the model size or use GPUs with large RAM. It's unlikely an issue from Google Cloud ML Engine. You can run your model directly on Compute Engine VM to verify it if you believe so. – Guoqing Xu Apr 09 '19 at 08:15

0 Answers0