could you comment on the usability of CUDA-Fortran? for instance, is it easy to do multithreading/m

A half of HPC is being done in that interesting ancient tongue, Many of us have legacy Fortran codes, and some are thinking of building (nVidia based) GPU-clusters.

I would be grateful for comments from those of you who have tried multi-GPU CUDA computing under PGI Fortran. I’d prefer a GNU
Fortran compiler with CUDA, but have no idea if/when such a thing will be available. I’m trying to decide if the $700 will be well spent…
I know that the code will look simpler and might be easier to write than in CUDA C, and will work ok on one GPU.

Are there any parallel primitives libraries in Fortran (like Thrust and CUPP)?


I suspect if you wait for a GNU Fortran compiler for Fortan, you will have to wait a long time. The PGI Cuda Fortran compiler is here now. I have it, and will be specifically using it with multiple C1060s.


I’m using a C CUDA library inside a legacy Fortran code. It works like a charm for my purposes. A C library could use Thrust or whatever you wish, of course.

Good luck!

wow, that’s right - I’ll try!

on the other hand I already found nice graphics for C++ to replace my atiquated pgplot – mathGL and if write a lot in C then why not go all the way and

compile with one compiler…


While not really a “parallel primitives” library, CULA, a GPU accelerated linear algebra package, includes FORTRAN, C, and C++ libraries that match the LAPACK Fortran interfaces.