Over at the PGIinsider, Brent LeBack writes that new PGI Compiler release 14.7 enables Unified Memory in CUDA Fortran.
I’ve shown a number of ways in which the CUDA 6.0 managed memory feature can make CUDA Fortran programming easier. At this point, managed memory should be viewed as a productivity feature, and not targeted at improved performance. As NVIDIA GPU architectures progress over the next several years, we expect to see better and better performance from the unified memory system. You can start today, and I think you’ll find managed memory is a good tool for learning, coding up prototypes, and simplifying complicated data structures which need to be used on both the host and device. Stay tuned; we have a few more ideas on how unified memory can make additional high-level Fortran language features available to CUDA Fortran programmers in the near future.