🐛 Bug

The memory usage comparison between the same data structures implemented with different backends (PyTorch tensors and NumPy arrays) shows over 4x higher usage when using PyTorch. Data structure consists of a list containing 5kk small tensors/arrays.

To Reproduce

Use this Gist: https://gist.github.com/iamhatesz/3ef34254febe482aa48e3e489f89b07b

Expected behavior

The memory usage for both data structures should be similar.

Environment

• PyTorch Version (e.g., 1.0): 1.0.1
• OS (e.g., Linux): Windows 10
• How you installed PyTorch (conda, pip, source): pip
• Build command you used (if compiling from source): -
• Python version: 3.7.1
• CUDA/cuDNN version: -
• GPU models and configuration: -
• Any other relevant information: -

Additional context

none