After some searching , I didn't find any standard way to release GPU memory that is allocated to Tensorflow.So I want to know is there a way to manually shrink VRAM objects that are no longer needed without actually rebooting?(It has hugged almost all of my VRAM)
It's the duty of the OS to release system memory after process termination , but it cannot release GPU memory since it has been controlled by the DMA and the graphics driver.
OS : Ubuntu 18.04
Driver : nVidia 430
GPU : GeForce GT 740M
Any help is appreciated.
Thanks.
0 Answers