I think your 3. and 4. are phrased a bit dismissively, though I don't entirely disagree. Not every workload in scientific computing maps cleanly into the GPGPU

I think your 3. and 4. are phrased a bit dismissively, though I don't entirely disagree. Not every workload in scientific computing maps cleanly into the GPGPU paradigm. Though that's not really a compelling reason to use Fortran over C++ An additional reason I've heard (I don't personally use Fortran) for keeping Fortran around is that it's straightforward to convert Matlab prototype code into Fortran