Description: This seminar will cover basics of the Rust programming language. We will cover installation on HPCMP systems, building and running code, and scaling Rust across multiple nodes with MPI. Unlike some compiled languages, Rust is memory safe and eliminates many potential memory errors at compile time. It also offers better performance than interpreted languages like Python. The build tools included with Rust simplify compilation and removes the need for complicated environment setup.

CUI

Description: Multiprocessing pools are often the go-to method of parallelizing Python code. However, these pools limit scalability on HPC to a single node. This seminar will demonstrate how to convert multiprocessing pools to mpi4py with minimal code changes. We will also briefly discuss MPI concepts and learn how to convert code to traditional MPI.

Description: Offloading computations from a CPU to a GPU is becoming increasingly the preferred method for performing computationally intensive parallel tasks. This is particularly useful for data analysis, scientific computation, and machine learning. While offloading computations can significantly increase performance, there is an inherent overhead associated with transferring data and managing synchronization between the CPU and GPU. Four GPU programming paradigms will be discussed that can address this issue.

This seminar will be going over OpenACC, OpenMP Offloading, CUDA Fortran and 'Do Concurrent' Fortran MPI examples that have run successfully on the DoD HPCMP systems. The examples will include Module Lists, Makefiles, source code, batch scripts and output examples, as well as comparison of the various GPUs and Programming Paradigms. Special emphasis will be given to Blueback's AMD MI300A APUs which combine the CPU and GPU on the same chip, drastically simplifying the effort of porting a CPU code to a GPU.