System crash with GPU

GeForce GTX 1080ti
driver :455.38
docker 21.03-py3
project GitHub - biubug6/Pytorch_Retinaface: Retinaface get 80.99% in widerface hard val using mobilenet0.25.

Q: I train the network with 4 GPU and 24 batch.The system crash when I train.If the system is out of memory, it should show the message to warn .But I directly disconnect.How can I do ?