Copy.deepcopy(model) won't increase too much cuda memory?

  • I have a cuda model, model = Net.cuda()
  • This model consumes ~510MB cuda memory via nvidia-smi
  • I copy it with model1 = copy.deepcopy(model)
  • I find that cuda memory only increase ~10MB.

Question

  1. Can I use model and model1 as two independent models?
  2. Why cuda memory only increases ~10MB?