Internode Parallelism: GPU-Aware MPI

The Message Passing Interface (MPI) is arguably the primary programming model used for applications' internode parallelism. Developed by the NCI Training Team, the Introduction to MPI workshop demonstrates MPI procedures based on the latest MPI Standard - 4.0, with the hands-on finite difference exercise seen in the Introduction to OpenMP workshop.

This workshop builds upon the Internode Parallelism: Introduction to MPI content, and explores GPU-accelerated distributed computing with CUDA and MPI, focusing on process and GPU-aware communication, memory management, and synchronisation.

If you have any questions regarding this training, please contact training.nci@anu.edu.au.

Prerequisites

1. Basic experience with C/C++ is required. Knowledge about C functions, pointers and memory management is sufficient.

2. Basic experience with bash or similar Unix shells. (Software Carpentry - The Unix Shell )

3. User experience on Gadi is ideal. (Seminar on Getting Started with Gadi