Pytorch release memory problem

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

Hi,
I have tried the follow code on PC platform, it can release memory successfully, but it doesn’t work on TX2 (Jetpack 4.4) .
Is there other method to release memory with Pytorch object?

import gc
import torch
x=torch.randn((32,3,300,300),device=torch.device(‘cuda’))
del x
gc.collect()
torch.cuda.empty_cache()

3

Hi,

Could you share which PyTorch package do you use?
Do you install it from the link shared in this topic?

Thanks.

4

Hi AastaLLL,
Sorry, I forgot to mention my pytorch version.
My pytorch version is 1.6 (torch-1.6.0-cp36-cp36m-linux_aarch64.whl ), and it was installed from Jetson Zoo.

5

Hi AastaLLL,
It can release memory successfully on TX2, with both pytorch 1.6 and 1.7.
Tensor shape like (32,3,300,300) is too small, can not get the obvious memory release amount.
So when I increase shape to (2000,3,300,300), I see the memory release result.

Thank you for help.