We used the CUDACC define for checking if a file is compiled with nvcc or not. Since we installed CUDA 2.0 stable this flag seems to be ignored. If we add a check on NVCC everything is ok again. In the documentation of the nvcc it clearly says that CUDACC should be used for those checks. Till the beta of CUDA 2.0 this was the case but now the flag does not be set anymore?