You can do GPGPU programming with Fortran, which is definitely not C-derived, predating C by at least 20 years.

I find Fortran pretty pleasant for scientific work (what it was designed for) and take pleasure in knowing it hasn't changed a whole lot over the years F70, F90...etc.

Sure, and it is a good alternative option that I forgot about, given that it is anyway its strong point, numerical computing.

However it doesn't win many hearts outside HPC nowadays even with the latest standard revisions, and I was thinking more about mainstream adoption.

There are also Accelerate, CUDA4J, ClojureCUDA, Hybridizer and Alea, but I am not sure about their adoption at large.