Hi, CUDA 7.5 / 8.0~RC just works fine with GCC-5, however bad with GCC-6.
Some useful information from ArchLinux community: https://bugs.archlinux.org/task/49272 https://www.reddit.com/r/archlinux/comments/4iczja/a_work_around_if_you_are_using_cuda_and_have/ There is a workaround which makes C++ standards falling back to c++98, but upstream sources written in C++11 would suffer. Besides, C++11 code compiles with the CUDA+GCC-5 combination. Maybe we will have to file RM requests against all the CUDA reverse dependencies, and in the future introduce them back via backports when CUDA works with GCC-6. This can be confirmed as long as CUDA8.0 were released, with the releasing date being unknown. -- Best, Lumin
