episode 8: Multi-GPU with MPI

[wikfeldt]

• communication data via host
• communication with GPU-aware MPI with device_ptr

[jussienko]

• querying number of devices (per node)
• how to assign MPI ranks to devices
• best practices for number of tasks per node: one task per GPU or multiple tasks per GPU?

[mabraham]

Stub for this lesson at https://github.com/ENCCS/openmp-gpu/blob/main/content/multi-gpu.rst