Segmentation fault (core dumped) when I run 'cutorch.synchronizeAll()'

[zerchen]

The environment is ubuntu16.04, CUDA 8, cudnn 6, GPU is Nvidia Tesla P100
I run the torch in nvidia docker container. However, when I run 'require cudnn ‘ or 'cutorch.synchronizeAll()'
, it returns 'Segmentation fault (core dumped)'
I analyze it with the help of gdb and the output is:

th> require 'cutorch'
[New Thread 0x7ffe4c232700 (LWP 160)]
th> cutorch.synchronizeAll()
[New Thread 0x7ffe48231700 (LWP 161)]
[New Thread 0x7ffe3ffff700 (LWP 162)]

Thread 67 "luajit" received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7ffe3ffff700 (LWP 162)]
0x00007ffe525868df in ?? () from /usr/local/nvidia/lib64/libcuda.so.1
(gdb) bt
#0 0x00007ffe525868df in ?? () from /usr/local/nvidia/lib64/libcuda.so.1
#1 0x00007ffe526e1014 in ?? () from /usr/local/nvidia/lib64/libcuda.so.1
#2 0x00007ffe5267a468 in ?? () from /usr/local/nvidia/lib64/libcuda.so.1
#3 0x00007ffff746e6ba in start_thread () from /lib/x86_64-linux-gnu/libpthread.so.0
#4 0x00007ffff6f8e82d in clone () from /lib/x86_64-linux-gnu/libc.so.6