The Julia language has caught my attention as I am a heavy MATLAB user, but would like more performance. Julia looks attractive as skipping the whole 'prototype in MATLAB then rewrite code in FORTRAN/C' would be nice, because my productivity drops drastically. But it seems that Julia is still quite young and some remarks in this post make me hesitant to try it.
The only parallel scalability results (not FEM related) I could find are some slides https://www.intel.com/content/dam/www/public/us/en/documents/presentation/julia-in-parallel-and-high-performance-computing.pdf
PyOPENCL/PyCUDA seems to have some success using GPUs. Maybe it is not too unreasonable to say the Julia GPU packages would give similar performance?