Aborted when run 2 training from libdarknet.so at the same time

[wanfade]

Hello, I compiled darknet and libdarknet.so on my macbook pro.
It is fine when I ran "darknet detector train" on the same GPU at the same.
But when turn to libdarknet.so,
the second training will be very slow and finally aborted with no more log info.

I load libdarknet.so in the both python code and C,
and the behaviors are same that training a detector on the one GPU is safe,
or two trainings on the different GPU respectively is also safe,
two training on the one GPU is aborted(in the most case one of the trainings aborted, sometimes both).

Wish your help, thank you.

[wanfade]

Update:
All problems disappeared when I changed the input size from 608 to 416.