Please read this post with guidelines and suggestions for this forum to learn:
- how to correctly quote inputs and outputs so that people can read them (yours are mangled and thus not readable)
- which information to always provide when reporting problems with a job
Without knowing what kind of GPU hardware you have, it is not possible to make any suggestions. Please keep in mind that when using 12 MPI ranks attached to the same GPU your possible acceleration from the GPU is split between the 12 processes and thus there cannot be as much acceleration as would be possible for a single process plus the overhead of launching GPU kernels will apply to each process.