Just as a clarification: this has nothing to do with the wrapper. This is a limitation of NVCC itself. The error check Stan mentioned is inside a Cuda library header file.
You should be able to add “-ccbin icpc” to your CXX Flags while setting OMPI_CXX=nvcc_wrapper and you get intel as the host compiler. But if you don’t have the exact right version NVCC will complain. Also, as Stan added we do not do nightly testing with NVCC+Intel.