Haven’t noticed it announced here, so: from here, beta of latest version of Portland Group Fortran compiler, supporting CUDA, is available for the first time. Both writing GPU kernels in Fortran, as well as calling CUDA runtime functions from host Fortran code, is supported. I was tracking the development since announced for the first time, and for a Fortran lover like myself, this all (except for the hefty price tag) looks very nice. For example, I especially liked syntax sugar making host code much simpler than with C (like: no cudaMemcpy*() calls, just simple array assignments); also, it’s going to be fun to see the performance compared with the code generated by nvcc, as well as to try some Fortran-specific constructs in the kernel code and check the PTX generated…