when I train for some epock,raise error;RuntimeError: DataLoader worker (pid 102459) is killed by signal: Segmentation fault.

[shihongji1993]

[mahdizynali]

It regularly happens when your gpu memory is full of data. decrease your batch size into you gpu memory space . for example batch sizes with 32 usually allocates about 7 to 8 gig of memory. in order to find out gpu memory size which is allocated during processing, try this command :

watch nvidia-smi

[shihongji1993]

It regularly happens when your gpu memory is full of data. decrease your batch size into you gpu memory space . for example batch sizes with 32 usually allocates about 7 to 8 gig of memory. in order to find out gpu memory size which is allocated during processing, try this command :

watch nvidia-smi

I have set the batch_size=32,but it also have the problem; when I set the workers_per_gpu=0 or 1,the problem can be solved,but the train speed also will be slower,

[mahdizynali]

What is your gpu model ? Would you check the number of cuda cores ?

[shihongji1993]

What is your gpu model ? Would you check the number of cuda cores ?

my GPU is 2080