compile on fronted without GPU


is it possible to compile CUDA fortran code on a fronted without a GPU and then execute on a node that has a GPU?

Thanks, Jan

Sure. Compilation does not need to be on the same system as the system where the binary is executed, nor does the build system need to have GPU attached.

  • Mat


Just to let you know: remember if your frontend is, say, a Haswell, but your GPU node is a Sandy Bridge, you’ll want to compile your code with an appropriate -tp for the GPU node. Otherwise, you’ll get some fun code explosions (usually unknown instruction) that–if you are like me–you’ll try to track down in your GPU code that in the end is a more basic issue.


Hi Matt and Mat,

thanks, I found the problem I had was unrelated to PGI but rather due to a slurm issue. The compiling works fine now.