Switched from CUDA8 to CUDA9 - "segmentation fault (core dumped)"


I am a deep learning engineer and I use keras + tensorflow + CUDA to train and test models. I have a testing interface set up which tests multiple models at once (in a loop). When running this program on CUDA8 I had no issues. After making the switch to CUDA9, I can no longer test more than two models at a time. If I try to test more than two models, I get a “segmentation fault (core dumped)” error and the program is forced to quit.

Does anyone know the cause of this issue?


Turns out the issue is with the use of “K.clearsession()” in Keras 2